SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/kthread.h>
  25 #include <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
  75 static noinline int
  76 block_group_cache_done(struct btrfs_block_group_cache *cache)
  77 {
  78         smp_mb();
  79         return cache->cached == BTRFS_CACHE_FINISHED;
  80 }
  81
  82 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  83 {
  84         return (cache->flags & bits) == bits;
  85 }
  86
  87 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
  88 {
  89         atomic_inc(&cache->count);
  90 }
  91
  92 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
  93 {
  94         if (atomic_dec_and_test(&cache->count))
  95                 kfree(cache);
  96 }
  97
  98 /*
  99  * this adds the block group to the fs_info rb tree for the block group
 100  * cache
 101  */
 102 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 103                                 struct btrfs_block_group_cache *block_group)
 104 {
 105         struct rb_node **p;
 106         struct rb_node *parent = NULL;
 107         struct btrfs_block_group_cache *cache;
 108
 109         spin_lock(&info->block_group_cache_lock);
 110         p = &info->block_group_cache_tree.rb_node;
 111
 112         while (*p) {
 113                 parent = *p;
 114                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 115                                  cache_node);
 116                 if (block_group->key.objectid < cache->key.objectid) {
 117                         p = &(*p)->rb_left;
 118                 } else if (block_group->key.objectid > cache->key.objectid) {
 119                         p = &(*p)->rb_right;
 120                 } else {
 121                         spin_unlock(&info->block_group_cache_lock);
 122                         return -EEXIST;
 123                 }
 124         }
 125
 126         rb_link_node(&block_group->cache_node, parent, p);
 127         rb_insert_color(&block_group->cache_node,
 128                         &info->block_group_cache_tree);
 129         spin_unlock(&info->block_group_cache_lock);
 130
 131         return 0;
 132 }
 133
 134 /*
 135  * This will return the block group at or after bytenr if contains is 0, else
 136  * it will return the block group that contains the bytenr
 137  */
 138 static struct btrfs_block_group_cache *
 139 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 140                               int contains)
 141 {
 142         struct btrfs_block_group_cache *cache, *ret = NULL;
 143         struct rb_node *n;
 144         u64 end, start;
 145
 146         spin_lock(&info->block_group_cache_lock);
 147         n = info->block_group_cache_tree.rb_node;
 148
 149         while (n) {
 150                 cache = rb_entry(n, struct btrfs_block_group_cache,
 151                                  cache_node);
 152                 end = cache->key.objectid + cache->key.offset - 1;
 153                 start = cache->key.objectid;
 154
 155                 if (bytenr < start) {
 156                         if (!contains && (!ret || start < ret->key.objectid))
 157                                 ret = cache;
 158                         n = n->rb_left;
 159                 } else if (bytenr > start) {
 160                         if (contains && bytenr <= end) {
 161                                 ret = cache;
 162                                 break;
 163                         }
 164                         n = n->rb_right;
 165                 } else {
 166                         ret = cache;
 167                         break;
 168                 }
 169         }
 170         if (ret)
 171                 btrfs_get_block_group(ret);
 172         spin_unlock(&info->block_group_cache_lock);
 173
 174         return ret;
 175 }
 176
 177 static int add_excluded_extent(struct btrfs_root *root,
 178                                u64 start, u64 num_bytes)
 179 {
 180         u64 end = start + num_bytes - 1;
 181         set_extent_bits(&root->fs_info->freed_extents[0],
 182                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 183         set_extent_bits(&root->fs_info->freed_extents[1],
 184                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 185         return 0;
 186 }
 187
 188 static void free_excluded_extents(struct btrfs_root *root,
 189                                   struct btrfs_block_group_cache *cache)
 190 {
 191         u64 start, end;
 192
 193         start = cache->key.objectid;
 194         end = start + cache->key.offset - 1;
 195
 196         clear_extent_bits(&root->fs_info->freed_extents[0],
 197                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 198         clear_extent_bits(&root->fs_info->freed_extents[1],
 199                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 200 }
 201
 202 static int exclude_super_stripes(struct btrfs_root *root,
 203                                  struct btrfs_block_group_cache *cache)
 204 {
 205         u64 bytenr;
 206         u64 *logical;
 207         int stripe_len;
 208         int i, nr, ret;
 209
 210         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 211                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 212                 cache->bytes_super += stripe_len;
 213                 ret = add_excluded_extent(root, cache->key.objectid,
 214                                           stripe_len);
 215                 BUG_ON(ret);
 216         }
 217
 218         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 219                 bytenr = btrfs_sb_offset(i);
 220                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 221                                        cache->key.objectid, bytenr,
 222                                        0, &logical, &nr, &stripe_len);
 223                 BUG_ON(ret);
 224
 225                 while (nr--) {
 226                         cache->bytes_super += stripe_len;
 227                         ret = add_excluded_extent(root, logical[nr],
 228                                                   stripe_len);
 229                         BUG_ON(ret);
 230                 }
 231
 232                 kfree(logical);
 233         }
 234         return 0;
 235 }
 236
 237 static struct btrfs_caching_control *
 238 get_caching_control(struct btrfs_block_group_cache *cache)
 239 {
 240         struct btrfs_caching_control *ctl;
 241
 242         spin_lock(&cache->lock);
 243         if (cache->cached != BTRFS_CACHE_STARTED) {
 244                 spin_unlock(&cache->lock);
 245                 return NULL;
 246         }
 247
 248         ctl = cache->caching_ctl;
 249         atomic_inc(&ctl->count);
 250         spin_unlock(&cache->lock);
 251         return ctl;
 252 }
 253
 254 static void put_caching_control(struct btrfs_caching_control *ctl)
 255 {
 256         if (atomic_dec_and_test(&ctl->count))
 257                 kfree(ctl);
 258 }
 259
 260 /*
 261  * this is only called by cache_block_group, since we could have freed extents
 262  * we need to check the pinned_extents for any extents that can't be used yet
 263  * since their free space will be released as soon as the transaction commits.
 264  */
 265 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 266                               struct btrfs_fs_info *info, u64 start, u64 end)
 267 {
 268         u64 extent_start, extent_end, size, total_added = 0;
 269         int ret;
 270
 271         while (start < end) {
 272                 ret = find_first_extent_bit(info->pinned_extents, start,
 273                                             &extent_start, &extent_end,
 274                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 275                 if (ret)
 276                         break;
 277
 278                 if (extent_start <= start) {
 279                         start = extent_end + 1;
 280                 } else if (extent_start > start && extent_start < end) {
 281                         size = extent_start - start;
 282                         total_added += size;
 283                         ret = btrfs_add_free_space(block_group, start,
 284                                                    size);
 285                         BUG_ON(ret);
 286                         start = extent_end + 1;
 287                 } else {
 288                         break;
 289                 }
 290         }
 291
 292         if (start < end) {
 293                 size = end - start;
 294                 total_added += size;
 295                 ret = btrfs_add_free_space(block_group, start, size);
 296                 BUG_ON(ret);
 297         }
 298
 299         return total_added;
 300 }
 301
 302 static int caching_kthread(void *data)
 303 {
 304         struct btrfs_block_group_cache *block_group = data;
 305         struct btrfs_fs_info *fs_info = block_group->fs_info;
 306         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 307         struct btrfs_root *extent_root = fs_info->extent_root;
 308         struct btrfs_path *path;
 309         struct extent_buffer *leaf;
 310         struct btrfs_key key;
 311         u64 total_found = 0;
 312         u64 last = 0;
 313         u32 nritems;
 314         int ret = 0;
 315
 316         path = btrfs_alloc_path();
 317         if (!path)
 318                 return -ENOMEM;
 319
 320         exclude_super_stripes(extent_root, block_group);
 321         spin_lock(&block_group->space_info->lock);
 322         block_group->space_info->bytes_super += block_group->bytes_super;
 323         spin_unlock(&block_group->space_info->lock);
 324
 325         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 326
 327         /*
 328          * We don't want to deadlock with somebody trying to allocate a new
 329          * extent for the extent root while also trying to search the extent
 330          * root to add free space.  So we skip locking and search the commit
 331          * root, since its read-only
 332          */
 333         path->skip_locking = 1;
 334         path->search_commit_root = 1;
 335         path->reada = 2;
 336
 337         key.objectid = last;
 338         key.offset = 0;
 339         key.type = BTRFS_EXTENT_ITEM_KEY;
 340 again:
 341         mutex_lock(&caching_ctl->mutex);
 342         /* need to make sure the commit_root doesn't disappear */
 343         down_read(&fs_info->extent_commit_sem);
 344
 345         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 346         if (ret < 0)
 347                 goto err;
 348
 349         leaf = path->nodes[0];
 350         nritems = btrfs_header_nritems(leaf);
 351
 352         while (1) {
 353                 smp_mb();
 354                 if (fs_info->closing > 1) {
 355                         last = (u64)-1;
 356                         break;
 357                 }
 358
 359                 if (path->slots[0] < nritems) {
 360                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 361                 } else {
 362                         ret = find_next_key(path, 0, &key);
 363                         if (ret)
 364                                 break;
 365
 366                         caching_ctl->progress = last;
 367                         btrfs_release_path(extent_root, path);
 368                         up_read(&fs_info->extent_commit_sem);
 369                         mutex_unlock(&caching_ctl->mutex);
 370                         if (btrfs_transaction_in_commit(fs_info))
 371                                 schedule_timeout(1);
 372                         else
 373                                 cond_resched();
 374                         goto again;
 375                 }
 376
 377                 if (key.objectid < block_group->key.objectid) {
 378                         path->slots[0]++;
 379                         continue;
 380                 }
 381
 382                 if (key.objectid >= block_group->key.objectid +
 383                     block_group->key.offset)
 384                         break;
 385
 386                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 387                         total_found += add_new_free_space(block_group,
 388                                                           fs_info, last,
 389                                                           key.objectid);
 390                         last = key.objectid + key.offset;
 391
 392                         if (total_found > (1024 * 1024 * 2)) {
 393                                 total_found = 0;
 394                                 wake_up(&caching_ctl->wait);
 395                         }
 396                 }
 397                 path->slots[0]++;
 398         }
 399         ret = 0;
 400
 401         total_found += add_new_free_space(block_group, fs_info, last,
 402                                           block_group->key.objectid +
 403                                           block_group->key.offset);
 404         caching_ctl->progress = (u64)-1;
 405
 406         spin_lock(&block_group->lock);
 407         block_group->caching_ctl = NULL;
 408         block_group->cached = BTRFS_CACHE_FINISHED;
 409         spin_unlock(&block_group->lock);
 410
 411 err:
 412         btrfs_free_path(path);
 413         up_read(&fs_info->extent_commit_sem);
 414
 415         free_excluded_extents(extent_root, block_group);
 416
 417         mutex_unlock(&caching_ctl->mutex);
 418         wake_up(&caching_ctl->wait);
 419
 420         put_caching_control(caching_ctl);
 421         atomic_dec(&block_group->space_info->caching_threads);
 422         btrfs_put_block_group(block_group);
 423
 424         return 0;
 425 }
 426
 427 static int cache_block_group(struct btrfs_block_group_cache *cache)
 428 {
 429         struct btrfs_fs_info *fs_info = cache->fs_info;
 430         struct btrfs_caching_control *caching_ctl;
 431         struct task_struct *tsk;
 432         int ret = 0;
 433
 434         smp_mb();
 435         if (cache->cached != BTRFS_CACHE_NO)
 436                 return 0;
 437
 438         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
 439         BUG_ON(!caching_ctl);
 440
 441         INIT_LIST_HEAD(&caching_ctl->list);
 442         mutex_init(&caching_ctl->mutex);
 443         init_waitqueue_head(&caching_ctl->wait);
 444         caching_ctl->block_group = cache;
 445         caching_ctl->progress = cache->key.objectid;
 446         /* one for caching kthread, one for caching block group list */
 447         atomic_set(&caching_ctl->count, 2);
 448
 449         spin_lock(&cache->lock);
 450         if (cache->cached != BTRFS_CACHE_NO) {
 451                 spin_unlock(&cache->lock);
 452                 kfree(caching_ctl);
 453                 return 0;
 454         }
 455         cache->caching_ctl = caching_ctl;
 456         cache->cached = BTRFS_CACHE_STARTED;
 457         spin_unlock(&cache->lock);
 458
 459         down_write(&fs_info->extent_commit_sem);
 460         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 461         up_write(&fs_info->extent_commit_sem);
 462
 463         atomic_inc(&cache->space_info->caching_threads);
 464         btrfs_get_block_group(cache);
 465
 466         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 467                           cache->key.objectid);
 468         if (IS_ERR(tsk)) {
 469                 ret = PTR_ERR(tsk);
 470                 printk(KERN_ERR "error running thread %d\n", ret);
 471                 BUG();
 472         }
 473
 474         return ret;
 475 }
 476
 477 /*
 478  * return the block group that starts at or after bytenr
 479  */
 480 static struct btrfs_block_group_cache *
 481 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 482 {
 483         struct btrfs_block_group_cache *cache;
 484
 485         cache = block_group_cache_tree_search(info, bytenr, 0);
 486
 487         return cache;
 488 }
 489
 490 /*
 491  * return the block group that contains the given bytenr
 492  */
 493 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 494                                                  struct btrfs_fs_info *info,
 495                                                  u64 bytenr)
 496 {
 497         struct btrfs_block_group_cache *cache;
 498
 499         cache = block_group_cache_tree_search(info, bytenr, 1);
 500
 501         return cache;
 502 }
 503
 504 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 505                                                   u64 flags)
 506 {
 507         struct list_head *head = &info->space_info;
 508         struct btrfs_space_info *found;
 509
 510         rcu_read_lock();
 511         list_for_each_entry_rcu(found, head, list) {
 512                 if (found->flags == flags) {
 513                         rcu_read_unlock();
 514                         return found;
 515                 }
 516         }
 517         rcu_read_unlock();
 518         return NULL;
 519 }
 520
 521 /*
 522  * after adding space to the filesystem, we need to clear the full flags
 523  * on all the space infos.
 524  */
 525 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 526 {
 527         struct list_head *head = &info->space_info;
 528         struct btrfs_space_info *found;
 529
 530         rcu_read_lock();
 531         list_for_each_entry_rcu(found, head, list)
 532                 found->full = 0;
 533         rcu_read_unlock();
 534 }
 535
 536 static u64 div_factor(u64 num, int factor)
 537 {
 538         if (factor == 10)
 539                 return num;
 540         num *= factor;
 541         do_div(num, 10);
 542         return num;
 543 }
 544
 545 u64 btrfs_find_block_group(struct btrfs_root *root,
 546                            u64 search_start, u64 search_hint, int owner)
 547 {
 548         struct btrfs_block_group_cache *cache;
 549         u64 used;
 550         u64 last = max(search_hint, search_start);
 551         u64 group_start = 0;
 552         int full_search = 0;
 553         int factor = 9;
 554         int wrapped = 0;
 555 again:
 556         while (1) {
 557                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 558                 if (!cache)
 559                         break;
 560
 561                 spin_lock(&cache->lock);
 562                 last = cache->key.objectid + cache->key.offset;
 563                 used = btrfs_block_group_used(&cache->item);
 564
 565                 if ((full_search || !cache->ro) &&
 566                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 567                         if (used + cache->pinned + cache->reserved <
 568                             div_factor(cache->key.offset, factor)) {
 569                                 group_start = cache->key.objectid;
 570                                 spin_unlock(&cache->lock);
 571                                 btrfs_put_block_group(cache);
 572                                 goto found;
 573                         }
 574                 }
 575                 spin_unlock(&cache->lock);
 576                 btrfs_put_block_group(cache);
 577                 cond_resched();
 578         }
 579         if (!wrapped) {
 580                 last = search_start;
 581                 wrapped = 1;
 582                 goto again;
 583         }
 584         if (!full_search && factor < 10) {
 585                 last = search_start;
 586                 full_search = 1;
 587                 factor = 10;
 588                 goto again;
 589         }
 590 found:
 591         return group_start;
 592 }
 593
 594 /* simple helper to search for an existing extent at a given offset */
 595 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 596 {
 597         int ret;
 598         struct btrfs_key key;
 599         struct btrfs_path *path;
 600
 601         path = btrfs_alloc_path();
 602         BUG_ON(!path);
 603         key.objectid = start;
 604         key.offset = len;
 605         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 606         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 607                                 0, 0);
 608         btrfs_free_path(path);
 609         return ret;
 610 }
 611
 612 /*
 613  * Back reference rules.  Back refs have three main goals:
 614  *
 615  * 1) differentiate between all holders of references to an extent so that
 616  *    when a reference is dropped we can make sure it was a valid reference
 617  *    before freeing the extent.
 618  *
 619  * 2) Provide enough information to quickly find the holders of an extent
 620  *    if we notice a given block is corrupted or bad.
 621  *
 622  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 623  *    maintenance.  This is actually the same as #2, but with a slightly
 624  *    different use case.
 625  *
 626  * There are two kinds of back refs. The implicit back refs is optimized
 627  * for pointers in non-shared tree blocks. For a given pointer in a block,
 628  * back refs of this kind provide information about the block's owner tree
 629  * and the pointer's key. These information allow us to find the block by
 630  * b-tree searching. The full back refs is for pointers in tree blocks not
 631  * referenced by their owner trees. The location of tree block is recorded
 632  * in the back refs. Actually the full back refs is generic, and can be
 633  * used in all cases the implicit back refs is used. The major shortcoming
 634  * of the full back refs is its overhead. Every time a tree block gets
 635  * COWed, we have to update back refs entry for all pointers in it.
 636  *
 637  * For a newly allocated tree block, we use implicit back refs for
 638  * pointers in it. This means most tree related operations only involve
 639  * implicit back refs. For a tree block created in old transaction, the
 640  * only way to drop a reference to it is COW it. So we can detect the
 641  * event that tree block loses its owner tree's reference and do the
 642  * back refs conversion.
 643  *
 644  * When a tree block is COW'd through a tree, there are four cases:
 645  *
 646  * The reference count of the block is one and the tree is the block's
 647  * owner tree. Nothing to do in this case.
 648  *
 649  * The reference count of the block is one and the tree is not the
 650  * block's owner tree. In this case, full back refs is used for pointers
 651  * in the block. Remove these full back refs, add implicit back refs for
 652  * every pointers in the new block.
 653  *
 654  * The reference count of the block is greater than one and the tree is
 655  * the block's owner tree. In this case, implicit back refs is used for
 656  * pointers in the block. Add full back refs for every pointers in the
 657  * block, increase lower level extents' reference counts. The original
 658  * implicit back refs are entailed to the new block.
 659  *
 660  * The reference count of the block is greater than one and the tree is
 661  * not the block's owner tree. Add implicit back refs for every pointer in
 662  * the new block, increase lower level extents' reference count.
 663  *
 664  * Back Reference Key composing:
 665  *
 666  * The key objectid corresponds to the first byte in the extent,
 667  * The key type is used to differentiate between types of back refs.
 668  * There are different meanings of the key offset for different types
 669  * of back refs.
 670  *
 671  * File extents can be referenced by:
 672  *
 673  * - multiple snapshots, subvolumes, or different generations in one subvol
 674  * - different files inside a single subvolume
 675  * - different offsets inside a file (bookend extents in file.c)
 676  *
 677  * The extent ref structure for the implicit back refs has fields for:
 678  *
 679  * - Objectid of the subvolume root
 680  * - objectid of the file holding the reference
 681  * - original offset in the file
 682  * - how many bookend extents
 683  *
 684  * The key offset for the implicit back refs is hash of the first
 685  * three fields.
 686  *
 687  * The extent ref structure for the full back refs has field for:
 688  *
 689  * - number of pointers in the tree leaf
 690  *
 691  * The key offset for the implicit back refs is the first byte of
 692  * the tree leaf
 693  *
 694  * When a file extent is allocated, The implicit back refs is used.
 695  * the fields are filled in:
 696  *
 697  *     (root_key.objectid, inode objectid, offset in file, 1)
 698  *
 699  * When a file extent is removed file truncation, we find the
 700  * corresponding implicit back refs and check the following fields:
 701  *
 702  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 703  *
 704  * Btree extents can be referenced by:
 705  *
 706  * - Different subvolumes
 707  *
 708  * Both the implicit back refs and the full back refs for tree blocks
 709  * only consist of key. The key offset for the implicit back refs is
 710  * objectid of block's owner tree. The key offset for the full back refs
 711  * is the first byte of parent block.
 712  *
 713  * When implicit back refs is used, information about the lowest key and
 714  * level of the tree block are required. These information are stored in
 715  * tree block info structure.
 716  */
 717
 718 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 719 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 720                                   struct btrfs_root *root,
 721                                   struct btrfs_path *path,
 722                                   u64 owner, u32 extra_size)
 723 {
 724         struct btrfs_extent_item *item;
 725         struct btrfs_extent_item_v0 *ei0;
 726         struct btrfs_extent_ref_v0 *ref0;
 727         struct btrfs_tree_block_info *bi;
 728         struct extent_buffer *leaf;
 729         struct btrfs_key key;
 730         struct btrfs_key found_key;
 731         u32 new_size = sizeof(*item);
 732         u64 refs;
 733         int ret;
 734
 735         leaf = path->nodes[0];
 736         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 737
 738         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 739         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 740                              struct btrfs_extent_item_v0);
 741         refs = btrfs_extent_refs_v0(leaf, ei0);
 742
 743         if (owner == (u64)-1) {
 744                 while (1) {
 745                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 746                                 ret = btrfs_next_leaf(root, path);
 747                                 if (ret < 0)
 748                                         return ret;
 749                                 BUG_ON(ret > 0);
 750                                 leaf = path->nodes[0];
 751                         }
 752                         btrfs_item_key_to_cpu(leaf, &found_key,
 753                                               path->slots[0]);
 754                         BUG_ON(key.objectid != found_key.objectid);
 755                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 756                                 path->slots[0]++;
 757                                 continue;
 758                         }
 759                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 760                                               struct btrfs_extent_ref_v0);
 761                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 762                         break;
 763                 }
 764         }
 765         btrfs_release_path(root, path);
 766
 767         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 768                 new_size += sizeof(*bi);
 769
 770         new_size -= sizeof(*ei0);
 771         ret = btrfs_search_slot(trans, root, &key, path,
 772                                 new_size + extra_size, 1);
 773         if (ret < 0)
 774                 return ret;
 775         BUG_ON(ret);
 776
 777         ret = btrfs_extend_item(trans, root, path, new_size);
 778         BUG_ON(ret);
 779
 780         leaf = path->nodes[0];
 781         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 782         btrfs_set_extent_refs(leaf, item, refs);
 783         /* FIXME: get real generation */
 784         btrfs_set_extent_generation(leaf, item, 0);
 785         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 786                 btrfs_set_extent_flags(leaf, item,
 787                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 788                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 789                 bi = (struct btrfs_tree_block_info *)(item + 1);
 790                 /* FIXME: get first key of the block */
 791                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 792                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 793         } else {
 794                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 795         }
 796         btrfs_mark_buffer_dirty(leaf);
 797         return 0;
 798 }
 799 #endif
 800
 801 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 802 {
 803         u32 high_crc = ~(u32)0;
 804         u32 low_crc = ~(u32)0;
 805         __le64 lenum;
 806
 807         lenum = cpu_to_le64(root_objectid);
 808         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 809         lenum = cpu_to_le64(owner);
 810         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 811         lenum = cpu_to_le64(offset);
 812         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 813
 814         return ((u64)high_crc << 31) ^ (u64)low_crc;
 815 }
 816
 817 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 818                                      struct btrfs_extent_data_ref *ref)
 819 {
 820         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 821                                     btrfs_extent_data_ref_objectid(leaf, ref),
 822                                     btrfs_extent_data_ref_offset(leaf, ref));
 823 }
 824
 825 static int match_extent_data_ref(struct extent_buffer *leaf,
 826                                  struct btrfs_extent_data_ref *ref,
 827                                  u64 root_objectid, u64 owner, u64 offset)
 828 {
 829         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 830             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 831             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 832                 return 0;
 833         return 1;
 834 }
 835
 836 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 837                                            struct btrfs_root *root,
 838                                            struct btrfs_path *path,
 839                                            u64 bytenr, u64 parent,
 840                                            u64 root_objectid,
 841                                            u64 owner, u64 offset)
 842 {
 843         struct btrfs_key key;
 844         struct btrfs_extent_data_ref *ref;
 845         struct extent_buffer *leaf;
 846         u32 nritems;
 847         int ret;
 848         int recow;
 849         int err = -ENOENT;
 850
 851         key.objectid = bytenr;
 852         if (parent) {
 853                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 854                 key.offset = parent;
 855         } else {
 856                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 857                 key.offset = hash_extent_data_ref(root_objectid,
 858                                                   owner, offset);
 859         }
 860 again:
 861         recow = 0;
 862         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 863         if (ret < 0) {
 864                 err = ret;
 865                 goto fail;
 866         }
 867
 868         if (parent) {
 869                 if (!ret)
 870                         return 0;
 871 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 872                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 873                 btrfs_release_path(root, path);
 874                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 875                 if (ret < 0) {
 876                         err = ret;
 877                         goto fail;
 878                 }
 879                 if (!ret)
 880                         return 0;
 881 #endif
 882                 goto fail;
 883         }
 884
 885         leaf = path->nodes[0];
 886         nritems = btrfs_header_nritems(leaf);
 887         while (1) {
 888                 if (path->slots[0] >= nritems) {
 889                         ret = btrfs_next_leaf(root, path);
 890                         if (ret < 0)
 891                                 err = ret;
 892                         if (ret)
 893                                 goto fail;
 894
 895                         leaf = path->nodes[0];
 896                         nritems = btrfs_header_nritems(leaf);
 897                         recow = 1;
 898                 }
 899
 900                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 901                 if (key.objectid != bytenr ||
 902                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
 903                         goto fail;
 904
 905                 ref = btrfs_item_ptr(leaf, path->slots[0],
 906                                      struct btrfs_extent_data_ref);
 907
 908                 if (match_extent_data_ref(leaf, ref, root_objectid,
 909                                           owner, offset)) {
 910                         if (recow) {
 911                                 btrfs_release_path(root, path);
 912                                 goto again;
 913                         }
 914                         err = 0;
 915                         break;
 916                 }
 917                 path->slots[0]++;
 918         }
 919 fail:
 920         return err;
 921 }
 922
 923 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
 924                                            struct btrfs_root *root,
 925                                            struct btrfs_path *path,
 926                                            u64 bytenr, u64 parent,
 927                                            u64 root_objectid, u64 owner,
 928                                            u64 offset, int refs_to_add)
 929 {
 930         struct btrfs_key key;
 931         struct extent_buffer *leaf;
 932         u32 size;
 933         u32 num_refs;
 934         int ret;
 935
 936         key.objectid = bytenr;
 937         if (parent) {
 938                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 939                 key.offset = parent;
 940                 size = sizeof(struct btrfs_shared_data_ref);
 941         } else {
 942                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 943                 key.offset = hash_extent_data_ref(root_objectid,
 944                                                   owner, offset);
 945                 size = sizeof(struct btrfs_extent_data_ref);
 946         }
 947
 948         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
 949         if (ret && ret != -EEXIST)
 950                 goto fail;
 951
 952         leaf = path->nodes[0];
 953         if (parent) {
 954                 struct btrfs_shared_data_ref *ref;
 955                 ref = btrfs_item_ptr(leaf, path->slots[0],
 956                                      struct btrfs_shared_data_ref);
 957                 if (ret == 0) {
 958                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
 959                 } else {
 960                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
 961                         num_refs += refs_to_add;
 962                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
 963                 }
 964         } else {
 965                 struct btrfs_extent_data_ref *ref;
 966                 while (ret == -EEXIST) {
 967                         ref = btrfs_item_ptr(leaf, path->slots[0],
 968                                              struct btrfs_extent_data_ref);
 969                         if (match_extent_data_ref(leaf, ref, root_objectid,
 970                                                   owner, offset))
 971                                 break;
 972                         btrfs_release_path(root, path);
 973                         key.offset++;
 974                         ret = btrfs_insert_empty_item(trans, root, path, &key,
 975                                                       size);
 976                         if (ret && ret != -EEXIST)
 977                                 goto fail;
 978
 979                         leaf = path->nodes[0];
 980                 }
 981                 ref = btrfs_item_ptr(leaf, path->slots[0],
 982                                      struct btrfs_extent_data_ref);
 983                 if (ret == 0) {
 984                         btrfs_set_extent_data_ref_root(leaf, ref,
 985                                                        root_objectid);
 986                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
 987                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
 988                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
 989                 } else {
 990                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
 991                         num_refs += refs_to_add;
 992                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
 993                 }
 994         }
 995         btrfs_mark_buffer_dirty(leaf);
 996         ret = 0;
 997 fail:
 998         btrfs_release_path(root, path);
 999         return ret;
1000 }
1001
1002 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1003                                            struct btrfs_root *root,
1004                                            struct btrfs_path *path,
1005                                            int refs_to_drop)
1006 {
1007         struct btrfs_key key;
1008         struct btrfs_extent_data_ref *ref1 = NULL;
1009         struct btrfs_shared_data_ref *ref2 = NULL;
1010         struct extent_buffer *leaf;
1011         u32 num_refs = 0;
1012         int ret = 0;
1013
1014         leaf = path->nodes[0];
1015         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1016
1017         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1018                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1019                                       struct btrfs_extent_data_ref);
1020                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1021         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1022                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1023                                       struct btrfs_shared_data_ref);
1024                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1025 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1026         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1027                 struct btrfs_extent_ref_v0 *ref0;
1028                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1029                                       struct btrfs_extent_ref_v0);
1030                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1031 #endif
1032         } else {
1033                 BUG();
1034         }
1035
1036         BUG_ON(num_refs < refs_to_drop);
1037         num_refs -= refs_to_drop;
1038
1039         if (num_refs == 0) {
1040                 ret = btrfs_del_item(trans, root, path);
1041         } else {
1042                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1043                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1044                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1045                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1046 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1047                 else {
1048                         struct btrfs_extent_ref_v0 *ref0;
1049                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1050                                         struct btrfs_extent_ref_v0);
1051                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1052                 }
1053 #endif
1054                 btrfs_mark_buffer_dirty(leaf);
1055         }
1056         return ret;
1057 }
1058
1059 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1060                                           struct btrfs_path *path,
1061                                           struct btrfs_extent_inline_ref *iref)
1062 {
1063         struct btrfs_key key;
1064         struct extent_buffer *leaf;
1065         struct btrfs_extent_data_ref *ref1;
1066         struct btrfs_shared_data_ref *ref2;
1067         u32 num_refs = 0;
1068
1069         leaf = path->nodes[0];
1070         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1071         if (iref) {
1072                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1073                     BTRFS_EXTENT_DATA_REF_KEY) {
1074                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1075                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1076                 } else {
1077                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1078                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1079                 }
1080         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1081                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1082                                       struct btrfs_extent_data_ref);
1083                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1084         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1085                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1086                                       struct btrfs_shared_data_ref);
1087                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1088 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1089         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1090                 struct btrfs_extent_ref_v0 *ref0;
1091                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1092                                       struct btrfs_extent_ref_v0);
1093                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1094 #endif
1095         } else {
1096                 WARN_ON(1);
1097         }
1098         return num_refs;
1099 }
1100
1101 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1102                                           struct btrfs_root *root,
1103                                           struct btrfs_path *path,
1104                                           u64 bytenr, u64 parent,
1105                                           u64 root_objectid)
1106 {
1107         struct btrfs_key key;
1108         int ret;
1109
1110         key.objectid = bytenr;
1111         if (parent) {
1112                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1113                 key.offset = parent;
1114         } else {
1115                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1116                 key.offset = root_objectid;
1117         }
1118
1119         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1120         if (ret > 0)
1121                 ret = -ENOENT;
1122 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1123         if (ret == -ENOENT && parent) {
1124                 btrfs_release_path(root, path);
1125                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1126                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1127                 if (ret > 0)
1128                         ret = -ENOENT;
1129         }
1130 #endif
1131         return ret;
1132 }
1133
1134 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1135                                           struct btrfs_root *root,
1136                                           struct btrfs_path *path,
1137                                           u64 bytenr, u64 parent,
1138                                           u64 root_objectid)
1139 {
1140         struct btrfs_key key;
1141         int ret;
1142
1143         key.objectid = bytenr;
1144         if (parent) {
1145                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1146                 key.offset = parent;
1147         } else {
1148                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1149                 key.offset = root_objectid;
1150         }
1151
1152         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1153         btrfs_release_path(root, path);
1154         return ret;
1155 }
1156
1157 static inline int extent_ref_type(u64 parent, u64 owner)
1158 {
1159         int type;
1160         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1161                 if (parent > 0)
1162                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1163                 else
1164                         type = BTRFS_TREE_BLOCK_REF_KEY;
1165         } else {
1166                 if (parent > 0)
1167                         type = BTRFS_SHARED_DATA_REF_KEY;
1168                 else
1169                         type = BTRFS_EXTENT_DATA_REF_KEY;
1170         }
1171         return type;
1172 }
1173
1174 static int find_next_key(struct btrfs_path *path, int level,
1175                          struct btrfs_key *key)
1176
1177 {
1178         for (; level < BTRFS_MAX_LEVEL; level++) {
1179                 if (!path->nodes[level])
1180                         break;
1181                 if (path->slots[level] + 1 >=
1182                     btrfs_header_nritems(path->nodes[level]))
1183                         continue;
1184                 if (level == 0)
1185                         btrfs_item_key_to_cpu(path->nodes[level], key,
1186                                               path->slots[level] + 1);
1187                 else
1188                         btrfs_node_key_to_cpu(path->nodes[level], key,
1189                                               path->slots[level] + 1);
1190                 return 0;
1191         }
1192         return 1;
1193 }
1194
1195 /*
1196  * look for inline back ref. if back ref is found, *ref_ret is set
1197  * to the address of inline back ref, and 0 is returned.
1198  *
1199  * if back ref isn't found, *ref_ret is set to the address where it
1200  * should be inserted, and -ENOENT is returned.
1201  *
1202  * if insert is true and there are too many inline back refs, the path
1203  * points to the extent item, and -EAGAIN is returned.
1204  *
1205  * NOTE: inline back refs are ordered in the same way that back ref
1206  *       items in the tree are ordered.
1207  */
1208 static noinline_for_stack
1209 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1210                                  struct btrfs_root *root,
1211                                  struct btrfs_path *path,
1212                                  struct btrfs_extent_inline_ref **ref_ret,
1213                                  u64 bytenr, u64 num_bytes,
1214                                  u64 parent, u64 root_objectid,
1215                                  u64 owner, u64 offset, int insert)
1216 {
1217         struct btrfs_key key;
1218         struct extent_buffer *leaf;
1219         struct btrfs_extent_item *ei;
1220         struct btrfs_extent_inline_ref *iref;
1221         u64 flags;
1222         u64 item_size;
1223         unsigned long ptr;
1224         unsigned long end;
1225         int extra_size;
1226         int type;
1227         int want;
1228         int ret;
1229         int err = 0;
1230
1231         key.objectid = bytenr;
1232         key.type = BTRFS_EXTENT_ITEM_KEY;
1233         key.offset = num_bytes;
1234
1235         want = extent_ref_type(parent, owner);
1236         if (insert) {
1237                 extra_size = btrfs_extent_inline_ref_size(want);
1238                 path->keep_locks = 1;
1239         } else
1240                 extra_size = -1;
1241         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1242         if (ret < 0) {
1243                 err = ret;
1244                 goto out;
1245         }
1246         BUG_ON(ret);
1247
1248         leaf = path->nodes[0];
1249         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1250 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1251         if (item_size < sizeof(*ei)) {
1252                 if (!insert) {
1253                         err = -ENOENT;
1254                         goto out;
1255                 }
1256                 ret = convert_extent_item_v0(trans, root, path, owner,
1257                                              extra_size);
1258                 if (ret < 0) {
1259                         err = ret;
1260                         goto out;
1261                 }
1262                 leaf = path->nodes[0];
1263                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1264         }
1265 #endif
1266         BUG_ON(item_size < sizeof(*ei));
1267
1268         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1269         flags = btrfs_extent_flags(leaf, ei);
1270
1271         ptr = (unsigned long)(ei + 1);
1272         end = (unsigned long)ei + item_size;
1273
1274         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1275                 ptr += sizeof(struct btrfs_tree_block_info);
1276                 BUG_ON(ptr > end);
1277         } else {
1278                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1279         }
1280
1281         err = -ENOENT;
1282         while (1) {
1283                 if (ptr >= end) {
1284                         WARN_ON(ptr > end);
1285                         break;
1286                 }
1287                 iref = (struct btrfs_extent_inline_ref *)ptr;
1288                 type = btrfs_extent_inline_ref_type(leaf, iref);
1289                 if (want < type)
1290                         break;
1291                 if (want > type) {
1292                         ptr += btrfs_extent_inline_ref_size(type);
1293                         continue;
1294                 }
1295
1296                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1297                         struct btrfs_extent_data_ref *dref;
1298                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1299                         if (match_extent_data_ref(leaf, dref, root_objectid,
1300                                                   owner, offset)) {
1301                                 err = 0;
1302                                 break;
1303                         }
1304                         if (hash_extent_data_ref_item(leaf, dref) <
1305                             hash_extent_data_ref(root_objectid, owner, offset))
1306                                 break;
1307                 } else {
1308                         u64 ref_offset;
1309                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1310                         if (parent > 0) {
1311                                 if (parent == ref_offset) {
1312                                         err = 0;
1313                                         break;
1314                                 }
1315                                 if (ref_offset < parent)
1316                                         break;
1317                         } else {
1318                                 if (root_objectid == ref_offset) {
1319                                         err = 0;
1320                                         break;
1321                                 }
1322                                 if (ref_offset < root_objectid)
1323                                         break;
1324                         }
1325                 }
1326                 ptr += btrfs_extent_inline_ref_size(type);
1327         }
1328         if (err == -ENOENT && insert) {
1329                 if (item_size + extra_size >=
1330                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1331                         err = -EAGAIN;
1332                         goto out;
1333                 }
1334                 /*
1335                  * To add new inline back ref, we have to make sure
1336                  * there is no corresponding back ref item.
1337                  * For simplicity, we just do not add new inline back
1338                  * ref if there is any kind of item for this block
1339                  */
1340                 if (find_next_key(path, 0, &key) == 0 &&
1341                     key.objectid == bytenr &&
1342                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1343                         err = -EAGAIN;
1344                         goto out;
1345                 }
1346         }
1347         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1348 out:
1349         if (insert) {
1350                 path->keep_locks = 0;
1351                 btrfs_unlock_up_safe(path, 1);
1352         }
1353         return err;
1354 }
1355
1356 /*
1357  * helper to add new inline back ref
1358  */
1359 static noinline_for_stack
1360 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1361                                 struct btrfs_root *root,
1362                                 struct btrfs_path *path,
1363                                 struct btrfs_extent_inline_ref *iref,
1364                                 u64 parent, u64 root_objectid,
1365                                 u64 owner, u64 offset, int refs_to_add,
1366                                 struct btrfs_delayed_extent_op *extent_op)
1367 {
1368         struct extent_buffer *leaf;
1369         struct btrfs_extent_item *ei;
1370         unsigned long ptr;
1371         unsigned long end;
1372         unsigned long item_offset;
1373         u64 refs;
1374         int size;
1375         int type;
1376         int ret;
1377
1378         leaf = path->nodes[0];
1379         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1380         item_offset = (unsigned long)iref - (unsigned long)ei;
1381
1382         type = extent_ref_type(parent, owner);
1383         size = btrfs_extent_inline_ref_size(type);
1384
1385         ret = btrfs_extend_item(trans, root, path, size);
1386         BUG_ON(ret);
1387
1388         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1389         refs = btrfs_extent_refs(leaf, ei);
1390         refs += refs_to_add;
1391         btrfs_set_extent_refs(leaf, ei, refs);
1392         if (extent_op)
1393                 __run_delayed_extent_op(extent_op, leaf, ei);
1394
1395         ptr = (unsigned long)ei + item_offset;
1396         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1397         if (ptr < end - size)
1398                 memmove_extent_buffer(leaf, ptr + size, ptr,
1399                                       end - size - ptr);
1400
1401         iref = (struct btrfs_extent_inline_ref *)ptr;
1402         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1403         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1404                 struct btrfs_extent_data_ref *dref;
1405                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1406                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1407                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1408                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1409                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1410         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1411                 struct btrfs_shared_data_ref *sref;
1412                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1413                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1414                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1415         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1416                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1417         } else {
1418                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1419         }
1420         btrfs_mark_buffer_dirty(leaf);
1421         return 0;
1422 }
1423
1424 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1425                                  struct btrfs_root *root,
1426                                  struct btrfs_path *path,
1427                                  struct btrfs_extent_inline_ref **ref_ret,
1428                                  u64 bytenr, u64 num_bytes, u64 parent,
1429                                  u64 root_objectid, u64 owner, u64 offset)
1430 {
1431         int ret;
1432
1433         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1434                                            bytenr, num_bytes, parent,
1435                                            root_objectid, owner, offset, 0);
1436         if (ret != -ENOENT)
1437                 return ret;
1438
1439         btrfs_release_path(root, path);
1440         *ref_ret = NULL;
1441
1442         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1443                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1444                                             root_objectid);
1445         } else {
1446                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1447                                              root_objectid, owner, offset);
1448         }
1449         return ret;
1450 }
1451
1452 /*
1453  * helper to update/remove inline back ref
1454  */
1455 static noinline_for_stack
1456 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1457                                  struct btrfs_root *root,
1458                                  struct btrfs_path *path,
1459                                  struct btrfs_extent_inline_ref *iref,
1460                                  int refs_to_mod,
1461                                  struct btrfs_delayed_extent_op *extent_op)
1462 {
1463         struct extent_buffer *leaf;
1464         struct btrfs_extent_item *ei;
1465         struct btrfs_extent_data_ref *dref = NULL;
1466         struct btrfs_shared_data_ref *sref = NULL;
1467         unsigned long ptr;
1468         unsigned long end;
1469         u32 item_size;
1470         int size;
1471         int type;
1472         int ret;
1473         u64 refs;
1474
1475         leaf = path->nodes[0];
1476         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1477         refs = btrfs_extent_refs(leaf, ei);
1478         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1479         refs += refs_to_mod;
1480         btrfs_set_extent_refs(leaf, ei, refs);
1481         if (extent_op)
1482                 __run_delayed_extent_op(extent_op, leaf, ei);
1483
1484         type = btrfs_extent_inline_ref_type(leaf, iref);
1485
1486         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1487                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1488                 refs = btrfs_extent_data_ref_count(leaf, dref);
1489         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1490                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1491                 refs = btrfs_shared_data_ref_count(leaf, sref);
1492         } else {
1493                 refs = 1;
1494                 BUG_ON(refs_to_mod != -1);
1495         }
1496
1497         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1498         refs += refs_to_mod;
1499
1500         if (refs > 0) {
1501                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1502                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1503                 else
1504                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1505         } else {
1506                 size =  btrfs_extent_inline_ref_size(type);
1507                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1508                 ptr = (unsigned long)iref;
1509                 end = (unsigned long)ei + item_size;
1510                 if (ptr + size < end)
1511                         memmove_extent_buffer(leaf, ptr, ptr + size,
1512                                               end - ptr - size);
1513                 item_size -= size;
1514                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1515                 BUG_ON(ret);
1516         }
1517         btrfs_mark_buffer_dirty(leaf);
1518         return 0;
1519 }
1520
1521 static noinline_for_stack
1522 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1523                                  struct btrfs_root *root,
1524                                  struct btrfs_path *path,
1525                                  u64 bytenr, u64 num_bytes, u64 parent,
1526                                  u64 root_objectid, u64 owner,
1527                                  u64 offset, int refs_to_add,
1528                                  struct btrfs_delayed_extent_op *extent_op)
1529 {
1530         struct btrfs_extent_inline_ref *iref;
1531         int ret;
1532
1533         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1534                                            bytenr, num_bytes, parent,
1535                                            root_objectid, owner, offset, 1);
1536         if (ret == 0) {
1537                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1538                 ret = update_inline_extent_backref(trans, root, path, iref,
1539                                                    refs_to_add, extent_op);
1540         } else if (ret == -ENOENT) {
1541                 ret = setup_inline_extent_backref(trans, root, path, iref,
1542                                                   parent, root_objectid,
1543                                                   owner, offset, refs_to_add,
1544                                                   extent_op);
1545         }
1546         return ret;
1547 }
1548
1549 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1550                                  struct btrfs_root *root,
1551                                  struct btrfs_path *path,
1552                                  u64 bytenr, u64 parent, u64 root_objectid,
1553                                  u64 owner, u64 offset, int refs_to_add)
1554 {
1555         int ret;
1556         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1557                 BUG_ON(refs_to_add != 1);
1558                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1559                                             parent, root_objectid);
1560         } else {
1561                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1562                                              parent, root_objectid,
1563                                              owner, offset, refs_to_add);
1564         }
1565         return ret;
1566 }
1567
1568 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1569                                  struct btrfs_root *root,
1570                                  struct btrfs_path *path,
1571                                  struct btrfs_extent_inline_ref *iref,
1572                                  int refs_to_drop, int is_data)
1573 {
1574         int ret;
1575
1576         BUG_ON(!is_data && refs_to_drop != 1);
1577         if (iref) {
1578                 ret = update_inline_extent_backref(trans, root, path, iref,
1579                                                    -refs_to_drop, NULL);
1580         } else if (is_data) {
1581                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1582         } else {
1583                 ret = btrfs_del_item(trans, root, path);
1584         }
1585         return ret;
1586 }
1587
1588 static void btrfs_issue_discard(struct block_device *bdev,
1589                                 u64 start, u64 len)
1590 {
1591         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1592                              DISCARD_FL_BARRIER);
1593 }
1594
1595 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1596                                 u64 num_bytes)
1597 {
1598         int ret;
1599         u64 map_length = num_bytes;
1600         struct btrfs_multi_bio *multi = NULL;
1601
1602         if (!btrfs_test_opt(root, DISCARD))
1603                 return 0;
1604
1605         /* Tell the block device(s) that the sectors can be discarded */
1606         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1607                               bytenr, &map_length, &multi, 0);
1608         if (!ret) {
1609                 struct btrfs_bio_stripe *stripe = multi->stripes;
1610                 int i;
1611
1612                 if (map_length > num_bytes)
1613                         map_length = num_bytes;
1614
1615                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1616                         btrfs_issue_discard(stripe->dev->bdev,
1617                                             stripe->physical,
1618                                             map_length);
1619                 }
1620                 kfree(multi);
1621         }
1622
1623         return ret;
1624 }
1625
1626 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1627                          struct btrfs_root *root,
1628                          u64 bytenr, u64 num_bytes, u64 parent,
1629                          u64 root_objectid, u64 owner, u64 offset)
1630 {
1631         int ret;
1632         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1633                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1634
1635         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1636                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1637                                         parent, root_objectid, (int)owner,
1638                                         BTRFS_ADD_DELAYED_REF, NULL);
1639         } else {
1640                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1641                                         parent, root_objectid, owner, offset,
1642                                         BTRFS_ADD_DELAYED_REF, NULL);
1643         }
1644         return ret;
1645 }
1646
1647 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1648                                   struct btrfs_root *root,
1649                                   u64 bytenr, u64 num_bytes,
1650                                   u64 parent, u64 root_objectid,
1651                                   u64 owner, u64 offset, int refs_to_add,
1652                                   struct btrfs_delayed_extent_op *extent_op)
1653 {
1654         struct btrfs_path *path;
1655         struct extent_buffer *leaf;
1656         struct btrfs_extent_item *item;
1657         u64 refs;
1658         int ret;
1659         int err = 0;
1660
1661         path = btrfs_alloc_path();
1662         if (!path)
1663                 return -ENOMEM;
1664
1665         path->reada = 1;
1666         path->leave_spinning = 1;
1667         /* this will setup the path even if it fails to insert the back ref */
1668         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1669                                            path, bytenr, num_bytes, parent,
1670                                            root_objectid, owner, offset,
1671                                            refs_to_add, extent_op);
1672         if (ret == 0)
1673                 goto out;
1674
1675         if (ret != -EAGAIN) {
1676                 err = ret;
1677                 goto out;
1678         }
1679
1680         leaf = path->nodes[0];
1681         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1682         refs = btrfs_extent_refs(leaf, item);
1683         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1684         if (extent_op)
1685                 __run_delayed_extent_op(extent_op, leaf, item);
1686
1687         btrfs_mark_buffer_dirty(leaf);
1688         btrfs_release_path(root->fs_info->extent_root, path);
1689
1690         path->reada = 1;
1691         path->leave_spinning = 1;
1692
1693         /* now insert the actual backref */
1694         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1695                                     path, bytenr, parent, root_objectid,
1696                                     owner, offset, refs_to_add);
1697         BUG_ON(ret);
1698 out:
1699         btrfs_free_path(path);
1700         return err;
1701 }
1702
1703 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1704                                 struct btrfs_root *root,
1705                                 struct btrfs_delayed_ref_node *node,
1706                                 struct btrfs_delayed_extent_op *extent_op,
1707                                 int insert_reserved)
1708 {
1709         int ret = 0;
1710         struct btrfs_delayed_data_ref *ref;
1711         struct btrfs_key ins;
1712         u64 parent = 0;
1713         u64 ref_root = 0;
1714         u64 flags = 0;
1715
1716         ins.objectid = node->bytenr;
1717         ins.offset = node->num_bytes;
1718         ins.type = BTRFS_EXTENT_ITEM_KEY;
1719
1720         ref = btrfs_delayed_node_to_data_ref(node);
1721         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1722                 parent = ref->parent;
1723         else
1724                 ref_root = ref->root;
1725
1726         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1727                 if (extent_op) {
1728                         BUG_ON(extent_op->update_key);
1729                         flags |= extent_op->flags_to_set;
1730                 }
1731                 ret = alloc_reserved_file_extent(trans, root,
1732                                                  parent, ref_root, flags,
1733                                                  ref->objectid, ref->offset,
1734                                                  &ins, node->ref_mod);
1735         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1736                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1737                                              node->num_bytes, parent,
1738                                              ref_root, ref->objectid,
1739                                              ref->offset, node->ref_mod,
1740                                              extent_op);
1741         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1742                 ret = __btrfs_free_extent(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 {
1748                 BUG();
1749         }
1750         return ret;
1751 }
1752
1753 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1754                                     struct extent_buffer *leaf,
1755                                     struct btrfs_extent_item *ei)
1756 {
1757         u64 flags = btrfs_extent_flags(leaf, ei);
1758         if (extent_op->update_flags) {
1759                 flags |= extent_op->flags_to_set;
1760                 btrfs_set_extent_flags(leaf, ei, flags);
1761         }
1762
1763         if (extent_op->update_key) {
1764                 struct btrfs_tree_block_info *bi;
1765                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1766                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1767                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1768         }
1769 }
1770
1771 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1772                                  struct btrfs_root *root,
1773                                  struct btrfs_delayed_ref_node *node,
1774                                  struct btrfs_delayed_extent_op *extent_op)
1775 {
1776         struct btrfs_key key;
1777         struct btrfs_path *path;
1778         struct btrfs_extent_item *ei;
1779         struct extent_buffer *leaf;
1780         u32 item_size;
1781         int ret;
1782         int err = 0;
1783
1784         path = btrfs_alloc_path();
1785         if (!path)
1786                 return -ENOMEM;
1787
1788         key.objectid = node->bytenr;
1789         key.type = BTRFS_EXTENT_ITEM_KEY;
1790         key.offset = node->num_bytes;
1791
1792         path->reada = 1;
1793         path->leave_spinning = 1;
1794         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1795                                 path, 0, 1);
1796         if (ret < 0) {
1797                 err = ret;
1798                 goto out;
1799         }
1800         if (ret > 0) {
1801                 err = -EIO;
1802                 goto out;
1803         }
1804
1805         leaf = path->nodes[0];
1806         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1807 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1808         if (item_size < sizeof(*ei)) {
1809                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1810                                              path, (u64)-1, 0);
1811                 if (ret < 0) {
1812                         err = ret;
1813                         goto out;
1814                 }
1815                 leaf = path->nodes[0];
1816                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1817         }
1818 #endif
1819         BUG_ON(item_size < sizeof(*ei));
1820         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1821         __run_delayed_extent_op(extent_op, leaf, ei);
1822
1823         btrfs_mark_buffer_dirty(leaf);
1824 out:
1825         btrfs_free_path(path);
1826         return err;
1827 }
1828
1829 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1830                                 struct btrfs_root *root,
1831                                 struct btrfs_delayed_ref_node *node,
1832                                 struct btrfs_delayed_extent_op *extent_op,
1833                                 int insert_reserved)
1834 {
1835         int ret = 0;
1836         struct btrfs_delayed_tree_ref *ref;
1837         struct btrfs_key ins;
1838         u64 parent = 0;
1839         u64 ref_root = 0;
1840
1841         ins.objectid = node->bytenr;
1842         ins.offset = node->num_bytes;
1843         ins.type = BTRFS_EXTENT_ITEM_KEY;
1844
1845         ref = btrfs_delayed_node_to_tree_ref(node);
1846         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1847                 parent = ref->parent;
1848         else
1849                 ref_root = ref->root;
1850
1851         BUG_ON(node->ref_mod != 1);
1852         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1853                 BUG_ON(!extent_op || !extent_op->update_flags ||
1854                        !extent_op->update_key);
1855                 ret = alloc_reserved_tree_block(trans, root,
1856                                                 parent, ref_root,
1857                                                 extent_op->flags_to_set,
1858                                                 &extent_op->key,
1859                                                 ref->level, &ins);
1860         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1861                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1862                                              node->num_bytes, parent, ref_root,
1863                                              ref->level, 0, 1, extent_op);
1864         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1865                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1866                                           node->num_bytes, parent, ref_root,
1867                                           ref->level, 0, 1, extent_op);
1868         } else {
1869                 BUG();
1870         }
1871         return ret;
1872 }
1873
1874
1875 /* helper function to actually process a single delayed ref entry */
1876 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1877                                struct btrfs_root *root,
1878                                struct btrfs_delayed_ref_node *node,
1879                                struct btrfs_delayed_extent_op *extent_op,
1880                                int insert_reserved)
1881 {
1882         int ret;
1883         if (btrfs_delayed_ref_is_head(node)) {
1884                 struct btrfs_delayed_ref_head *head;
1885                 /*
1886                  * we've hit the end of the chain and we were supposed
1887                  * to insert this extent into the tree.  But, it got
1888                  * deleted before we ever needed to insert it, so all
1889                  * we have to do is clean up the accounting
1890                  */
1891                 BUG_ON(extent_op);
1892                 head = btrfs_delayed_node_to_head(node);
1893                 if (insert_reserved) {
1894                         int mark_free = 0;
1895                         struct extent_buffer *must_clean = NULL;
1896
1897                         ret = pin_down_bytes(trans, root, NULL,
1898                                              node->bytenr, node->num_bytes,
1899                                              head->is_data, 1, &must_clean);
1900                         if (ret > 0)
1901                                 mark_free = 1;
1902
1903                         if (must_clean) {
1904                                 clean_tree_block(NULL, root, must_clean);
1905                                 btrfs_tree_unlock(must_clean);
1906                                 free_extent_buffer(must_clean);
1907                         }
1908                         if (head->is_data) {
1909                                 ret = btrfs_del_csums(trans, root,
1910                                                       node->bytenr,
1911                                                       node->num_bytes);
1912                                 BUG_ON(ret);
1913                         }
1914                         if (mark_free) {
1915                                 ret = btrfs_free_reserved_extent(root,
1916                                                         node->bytenr,
1917                                                         node->num_bytes);
1918                                 BUG_ON(ret);
1919                         }
1920                 }
1921                 mutex_unlock(&head->mutex);
1922                 return 0;
1923         }
1924
1925         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1926             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1927                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1928                                            insert_reserved);
1929         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1930                  node->type == BTRFS_SHARED_DATA_REF_KEY)
1931                 ret = run_delayed_data_ref(trans, root, node, extent_op,
1932                                            insert_reserved);
1933         else
1934                 BUG();
1935         return ret;
1936 }
1937
1938 static noinline struct btrfs_delayed_ref_node *
1939 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1940 {
1941         struct rb_node *node;
1942         struct btrfs_delayed_ref_node *ref;
1943         int action = BTRFS_ADD_DELAYED_REF;
1944 again:
1945         /*
1946          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1947          * this prevents ref count from going down to zero when
1948          * there still are pending delayed ref.
1949          */
1950         node = rb_prev(&head->node.rb_node);
1951         while (1) {
1952                 if (!node)
1953                         break;
1954                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1955                                 rb_node);
1956                 if (ref->bytenr != head->node.bytenr)
1957                         break;
1958                 if (ref->action == action)
1959                         return ref;
1960                 node = rb_prev(node);
1961         }
1962         if (action == BTRFS_ADD_DELAYED_REF) {
1963                 action = BTRFS_DROP_DELAYED_REF;
1964                 goto again;
1965         }
1966         return NULL;
1967 }
1968
1969 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1970                                        struct btrfs_root *root,
1971                                        struct list_head *cluster)
1972 {
1973         struct btrfs_delayed_ref_root *delayed_refs;
1974         struct btrfs_delayed_ref_node *ref;
1975         struct btrfs_delayed_ref_head *locked_ref = NULL;
1976         struct btrfs_delayed_extent_op *extent_op;
1977         int ret;
1978         int count = 0;
1979         int must_insert_reserved = 0;
1980
1981         delayed_refs = &trans->transaction->delayed_refs;
1982         while (1) {
1983                 if (!locked_ref) {
1984                         /* pick a new head ref from the cluster list */
1985                         if (list_empty(cluster))
1986                                 break;
1987
1988                         locked_ref = list_entry(cluster->next,
1989                                      struct btrfs_delayed_ref_head, cluster);
1990
1991                         /* grab the lock that says we are going to process
1992                          * all the refs for this head */
1993                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
1994
1995                         /*
1996                          * we may have dropped the spin lock to get the head
1997                          * mutex lock, and that might have given someone else
1998                          * time to free the head.  If that's true, it has been
1999                          * removed from our list and we can move on.
2000                          */
2001                         if (ret == -EAGAIN) {
2002                                 locked_ref = NULL;
2003                                 count++;
2004                                 continue;
2005                         }
2006                 }
2007
2008                 /*
2009                  * record the must insert reserved flag before we
2010                  * drop the spin lock.
2011                  */
2012                 must_insert_reserved = locked_ref->must_insert_reserved;
2013                 locked_ref->must_insert_reserved = 0;
2014
2015                 extent_op = locked_ref->extent_op;
2016                 locked_ref->extent_op = NULL;
2017
2018                 /*
2019                  * locked_ref is the head node, so we have to go one
2020                  * node back for any delayed ref updates
2021                  */
2022                 ref = select_delayed_ref(locked_ref);
2023                 if (!ref) {
2024                         /* All delayed refs have been processed, Go ahead
2025                          * and send the head node to run_one_delayed_ref,
2026                          * so that any accounting fixes can happen
2027                          */
2028                         ref = &locked_ref->node;
2029
2030                         if (extent_op && must_insert_reserved) {
2031                                 kfree(extent_op);
2032                                 extent_op = NULL;
2033                         }
2034
2035                         if (extent_op) {
2036                                 spin_unlock(&delayed_refs->lock);
2037
2038                                 ret = run_delayed_extent_op(trans, root,
2039                                                             ref, extent_op);
2040                                 BUG_ON(ret);
2041                                 kfree(extent_op);
2042
2043                                 cond_resched();
2044                                 spin_lock(&delayed_refs->lock);
2045                                 continue;
2046                         }
2047
2048                         list_del_init(&locked_ref->cluster);
2049                         locked_ref = NULL;
2050                 }
2051
2052                 ref->in_tree = 0;
2053                 rb_erase(&ref->rb_node, &delayed_refs->root);
2054                 delayed_refs->num_entries--;
2055
2056                 spin_unlock(&delayed_refs->lock);
2057
2058                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2059                                           must_insert_reserved);
2060                 BUG_ON(ret);
2061
2062                 btrfs_put_delayed_ref(ref);
2063                 kfree(extent_op);
2064                 count++;
2065
2066                 cond_resched();
2067                 spin_lock(&delayed_refs->lock);
2068         }
2069         return count;
2070 }
2071
2072 /*
2073  * this starts processing the delayed reference count updates and
2074  * extent insertions we have queued up so far.  count can be
2075  * 0, which means to process everything in the tree at the start
2076  * of the run (but not newly added entries), or it can be some target
2077  * number you'd like to process.
2078  */
2079 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2080                            struct btrfs_root *root, unsigned long count)
2081 {
2082         struct rb_node *node;
2083         struct btrfs_delayed_ref_root *delayed_refs;
2084         struct btrfs_delayed_ref_node *ref;
2085         struct list_head cluster;
2086         int ret;
2087         int run_all = count == (unsigned long)-1;
2088         int run_most = 0;
2089
2090         if (root == root->fs_info->extent_root)
2091                 root = root->fs_info->tree_root;
2092
2093         delayed_refs = &trans->transaction->delayed_refs;
2094         INIT_LIST_HEAD(&cluster);
2095 again:
2096         spin_lock(&delayed_refs->lock);
2097         if (count == 0) {
2098                 count = delayed_refs->num_entries * 2;
2099                 run_most = 1;
2100         }
2101         while (1) {
2102                 if (!(run_all || run_most) &&
2103                     delayed_refs->num_heads_ready < 64)
2104                         break;
2105
2106                 /*
2107                  * go find something we can process in the rbtree.  We start at
2108                  * the beginning of the tree, and then build a cluster
2109                  * of refs to process starting at the first one we are able to
2110                  * lock
2111                  */
2112                 ret = btrfs_find_ref_cluster(trans, &cluster,
2113                                              delayed_refs->run_delayed_start);
2114                 if (ret)
2115                         break;
2116
2117                 ret = run_clustered_refs(trans, root, &cluster);
2118                 BUG_ON(ret < 0);
2119
2120                 count -= min_t(unsigned long, ret, count);
2121
2122                 if (count == 0)
2123                         break;
2124         }
2125
2126         if (run_all) {
2127                 node = rb_first(&delayed_refs->root);
2128                 if (!node)
2129                         goto out;
2130                 count = (unsigned long)-1;
2131
2132                 while (node) {
2133                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2134                                        rb_node);
2135                         if (btrfs_delayed_ref_is_head(ref)) {
2136                                 struct btrfs_delayed_ref_head *head;
2137
2138                                 head = btrfs_delayed_node_to_head(ref);
2139                                 atomic_inc(&ref->refs);
2140
2141                                 spin_unlock(&delayed_refs->lock);
2142                                 mutex_lock(&head->mutex);
2143                                 mutex_unlock(&head->mutex);
2144
2145                                 btrfs_put_delayed_ref(ref);
2146                                 cond_resched();
2147                                 goto again;
2148                         }
2149                         node = rb_next(node);
2150                 }
2151                 spin_unlock(&delayed_refs->lock);
2152                 schedule_timeout(1);
2153                 goto again;
2154         }
2155 out:
2156         spin_unlock(&delayed_refs->lock);
2157         return 0;
2158 }
2159
2160 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2161                                 struct btrfs_root *root,
2162                                 u64 bytenr, u64 num_bytes, u64 flags,
2163                                 int is_data)
2164 {
2165         struct btrfs_delayed_extent_op *extent_op;
2166         int ret;
2167
2168         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2169         if (!extent_op)
2170                 return -ENOMEM;
2171
2172         extent_op->flags_to_set = flags;
2173         extent_op->update_flags = 1;
2174         extent_op->update_key = 0;
2175         extent_op->is_data = is_data ? 1 : 0;
2176
2177         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2178         if (ret)
2179                 kfree(extent_op);
2180         return ret;
2181 }
2182
2183 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2184                                       struct btrfs_root *root,
2185                                       struct btrfs_path *path,
2186                                       u64 objectid, u64 offset, u64 bytenr)
2187 {
2188         struct btrfs_delayed_ref_head *head;
2189         struct btrfs_delayed_ref_node *ref;
2190         struct btrfs_delayed_data_ref *data_ref;
2191         struct btrfs_delayed_ref_root *delayed_refs;
2192         struct rb_node *node;
2193         int ret = 0;
2194
2195         ret = -ENOENT;
2196         delayed_refs = &trans->transaction->delayed_refs;
2197         spin_lock(&delayed_refs->lock);
2198         head = btrfs_find_delayed_ref_head(trans, bytenr);
2199         if (!head)
2200                 goto out;
2201
2202         if (!mutex_trylock(&head->mutex)) {
2203                 atomic_inc(&head->node.refs);
2204                 spin_unlock(&delayed_refs->lock);
2205
2206                 btrfs_release_path(root->fs_info->extent_root, path);
2207
2208                 mutex_lock(&head->mutex);
2209                 mutex_unlock(&head->mutex);
2210                 btrfs_put_delayed_ref(&head->node);
2211                 return -EAGAIN;
2212         }
2213
2214         node = rb_prev(&head->node.rb_node);
2215         if (!node)
2216                 goto out_unlock;
2217
2218         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2219
2220         if (ref->bytenr != bytenr)
2221                 goto out_unlock;
2222
2223         ret = 1;
2224         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2225                 goto out_unlock;
2226
2227         data_ref = btrfs_delayed_node_to_data_ref(ref);
2228
2229         node = rb_prev(node);
2230         if (node) {
2231                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2232                 if (ref->bytenr == bytenr)
2233                         goto out_unlock;
2234         }
2235
2236         if (data_ref->root != root->root_key.objectid ||
2237             data_ref->objectid != objectid || data_ref->offset != offset)
2238                 goto out_unlock;
2239
2240         ret = 0;
2241 out_unlock:
2242         mutex_unlock(&head->mutex);
2243 out:
2244         spin_unlock(&delayed_refs->lock);
2245         return ret;
2246 }
2247
2248 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2249                                         struct btrfs_root *root,
2250                                         struct btrfs_path *path,
2251                                         u64 objectid, u64 offset, u64 bytenr)
2252 {
2253         struct btrfs_root *extent_root = root->fs_info->extent_root;
2254         struct extent_buffer *leaf;
2255         struct btrfs_extent_data_ref *ref;
2256         struct btrfs_extent_inline_ref *iref;
2257         struct btrfs_extent_item *ei;
2258         struct btrfs_key key;
2259         u32 item_size;
2260         int ret;
2261
2262         key.objectid = bytenr;
2263         key.offset = (u64)-1;
2264         key.type = BTRFS_EXTENT_ITEM_KEY;
2265
2266         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2267         if (ret < 0)
2268                 goto out;
2269         BUG_ON(ret == 0);
2270
2271         ret = -ENOENT;
2272         if (path->slots[0] == 0)
2273                 goto out;
2274
2275         path->slots[0]--;
2276         leaf = path->nodes[0];
2277         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2278
2279         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2280                 goto out;
2281
2282         ret = 1;
2283         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2284 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2285         if (item_size < sizeof(*ei)) {
2286                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2287                 goto out;
2288         }
2289 #endif
2290         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2291
2292         if (item_size != sizeof(*ei) +
2293             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2294                 goto out;
2295
2296         if (btrfs_extent_generation(leaf, ei) <=
2297             btrfs_root_last_snapshot(&root->root_item))
2298                 goto out;
2299
2300         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2301         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2302             BTRFS_EXTENT_DATA_REF_KEY)
2303                 goto out;
2304
2305         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2306         if (btrfs_extent_refs(leaf, ei) !=
2307             btrfs_extent_data_ref_count(leaf, ref) ||
2308             btrfs_extent_data_ref_root(leaf, ref) !=
2309             root->root_key.objectid ||
2310             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2311             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2312                 goto out;
2313
2314         ret = 0;
2315 out:
2316         return ret;
2317 }
2318
2319 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2320                           struct btrfs_root *root,
2321                           u64 objectid, u64 offset, u64 bytenr)
2322 {
2323         struct btrfs_path *path;
2324         int ret;
2325         int ret2;
2326
2327         path = btrfs_alloc_path();
2328         if (!path)
2329                 return -ENOENT;
2330
2331         do {
2332                 ret = check_committed_ref(trans, root, path, objectid,
2333                                           offset, bytenr);
2334                 if (ret && ret != -ENOENT)
2335                         goto out;
2336
2337                 ret2 = check_delayed_ref(trans, root, path, objectid,
2338                                          offset, bytenr);
2339         } while (ret2 == -EAGAIN);
2340
2341         if (ret2 && ret2 != -ENOENT) {
2342                 ret = ret2;
2343                 goto out;
2344         }
2345
2346         if (ret != -ENOENT || ret2 != -ENOENT)
2347                 ret = 0;
2348 out:
2349         btrfs_free_path(path);
2350         return ret;
2351 }
2352
2353 #if 0
2354 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2355                     struct extent_buffer *buf, u32 nr_extents)
2356 {
2357         struct btrfs_key key;
2358         struct btrfs_file_extent_item *fi;
2359         u64 root_gen;
2360         u32 nritems;
2361         int i;
2362         int level;
2363         int ret = 0;
2364         int shared = 0;
2365
2366         if (!root->ref_cows)
2367                 return 0;
2368
2369         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2370                 shared = 0;
2371                 root_gen = root->root_key.offset;
2372         } else {
2373                 shared = 1;
2374                 root_gen = trans->transid - 1;
2375         }
2376
2377         level = btrfs_header_level(buf);
2378         nritems = btrfs_header_nritems(buf);
2379
2380         if (level == 0) {
2381                 struct btrfs_leaf_ref *ref;
2382                 struct btrfs_extent_info *info;
2383
2384                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2385                 if (!ref) {
2386                         ret = -ENOMEM;
2387                         goto out;
2388                 }
2389
2390                 ref->root_gen = root_gen;
2391                 ref->bytenr = buf->start;
2392                 ref->owner = btrfs_header_owner(buf);
2393                 ref->generation = btrfs_header_generation(buf);
2394                 ref->nritems = nr_extents;
2395                 info = ref->extents;
2396
2397                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2398                         u64 disk_bytenr;
2399                         btrfs_item_key_to_cpu(buf, &key, i);
2400                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2401                                 continue;
2402                         fi = btrfs_item_ptr(buf, i,
2403                                             struct btrfs_file_extent_item);
2404                         if (btrfs_file_extent_type(buf, fi) ==
2405                             BTRFS_FILE_EXTENT_INLINE)
2406                                 continue;
2407                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2408                         if (disk_bytenr == 0)
2409                                 continue;
2410
2411                         info->bytenr = disk_bytenr;
2412                         info->num_bytes =
2413                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2414                         info->objectid = key.objectid;
2415                         info->offset = key.offset;
2416                         info++;
2417                 }
2418
2419                 ret = btrfs_add_leaf_ref(root, ref, shared);
2420                 if (ret == -EEXIST && shared) {
2421                         struct btrfs_leaf_ref *old;
2422                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2423                         BUG_ON(!old);
2424                         btrfs_remove_leaf_ref(root, old);
2425                         btrfs_free_leaf_ref(root, old);
2426                         ret = btrfs_add_leaf_ref(root, ref, shared);
2427                 }
2428                 WARN_ON(ret);
2429                 btrfs_free_leaf_ref(root, ref);
2430         }
2431 out:
2432         return ret;
2433 }
2434
2435 /* when a block goes through cow, we update the reference counts of
2436  * everything that block points to.  The internal pointers of the block
2437  * can be in just about any order, and it is likely to have clusters of
2438  * things that are close together and clusters of things that are not.
2439  *
2440  * To help reduce the seeks that come with updating all of these reference
2441  * counts, sort them by byte number before actual updates are done.
2442  *
2443  * struct refsort is used to match byte number to slot in the btree block.
2444  * we sort based on the byte number and then use the slot to actually
2445  * find the item.
2446  *
2447  * struct refsort is smaller than strcut btrfs_item and smaller than
2448  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2449  * for a btree block, there's no way for a kmalloc of refsorts for a
2450  * single node to be bigger than a page.
2451  */
2452 struct refsort {
2453         u64 bytenr;
2454         u32 slot;
2455 };
2456
2457 /*
2458  * for passing into sort()
2459  */
2460 static int refsort_cmp(const void *a_void, const void *b_void)
2461 {
2462         const struct refsort *a = a_void;
2463         const struct refsort *b = b_void;
2464
2465         if (a->bytenr < b->bytenr)
2466                 return -1;
2467         if (a->bytenr > b->bytenr)
2468                 return 1;
2469         return 0;
2470 }
2471 #endif
2472
2473 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2474                            struct btrfs_root *root,
2475                            struct extent_buffer *buf,
2476                            int full_backref, int inc)
2477 {
2478         u64 bytenr;
2479         u64 num_bytes;
2480         u64 parent;
2481         u64 ref_root;
2482         u32 nritems;
2483         struct btrfs_key key;
2484         struct btrfs_file_extent_item *fi;
2485         int i;
2486         int level;
2487         int ret = 0;
2488         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2489                             u64, u64, u64, u64, u64, u64);
2490
2491         ref_root = btrfs_header_owner(buf);
2492         nritems = btrfs_header_nritems(buf);
2493         level = btrfs_header_level(buf);
2494
2495         if (!root->ref_cows && level == 0)
2496                 return 0;
2497
2498         if (inc)
2499                 process_func = btrfs_inc_extent_ref;
2500         else
2501                 process_func = btrfs_free_extent;
2502
2503         if (full_backref)
2504                 parent = buf->start;
2505         else
2506                 parent = 0;
2507
2508         for (i = 0; i < nritems; i++) {
2509                 if (level == 0) {
2510                         btrfs_item_key_to_cpu(buf, &key, i);
2511                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2512                                 continue;
2513                         fi = btrfs_item_ptr(buf, i,
2514                                             struct btrfs_file_extent_item);
2515                         if (btrfs_file_extent_type(buf, fi) ==
2516                             BTRFS_FILE_EXTENT_INLINE)
2517                                 continue;
2518                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2519                         if (bytenr == 0)
2520                                 continue;
2521
2522                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2523                         key.offset -= btrfs_file_extent_offset(buf, fi);
2524                         ret = process_func(trans, root, bytenr, num_bytes,
2525                                            parent, ref_root, key.objectid,
2526                                            key.offset);
2527                         if (ret)
2528                                 goto fail;
2529                 } else {
2530                         bytenr = btrfs_node_blockptr(buf, i);
2531                         num_bytes = btrfs_level_size(root, level - 1);
2532                         ret = process_func(trans, root, bytenr, num_bytes,
2533                                            parent, ref_root, level - 1, 0);
2534                         if (ret)
2535                                 goto fail;
2536                 }
2537         }
2538         return 0;
2539 fail:
2540         BUG();
2541         return ret;
2542 }
2543
2544 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2545                   struct extent_buffer *buf, int full_backref)
2546 {
2547         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2548 }
2549
2550 int btrfs_dec_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, 0);
2554 }
2555
2556 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2557                                  struct btrfs_root *root,
2558                                  struct btrfs_path *path,
2559                                  struct btrfs_block_group_cache *cache)
2560 {
2561         int ret;
2562         struct btrfs_root *extent_root = root->fs_info->extent_root;
2563         unsigned long bi;
2564         struct extent_buffer *leaf;
2565
2566         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2567         if (ret < 0)
2568                 goto fail;
2569         BUG_ON(ret);
2570
2571         leaf = path->nodes[0];
2572         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2573         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2574         btrfs_mark_buffer_dirty(leaf);
2575         btrfs_release_path(extent_root, path);
2576 fail:
2577         if (ret)
2578                 return ret;
2579         return 0;
2580
2581 }
2582
2583 static struct btrfs_block_group_cache *
2584 next_block_group(struct btrfs_root *root,
2585                  struct btrfs_block_group_cache *cache)
2586 {
2587         struct rb_node *node;
2588         spin_lock(&root->fs_info->block_group_cache_lock);
2589         node = rb_next(&cache->cache_node);
2590         btrfs_put_block_group(cache);
2591         if (node) {
2592                 cache = rb_entry(node, struct btrfs_block_group_cache,
2593                                  cache_node);
2594                 btrfs_get_block_group(cache);
2595         } else
2596                 cache = NULL;
2597         spin_unlock(&root->fs_info->block_group_cache_lock);
2598         return cache;
2599 }
2600
2601 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2602                                    struct btrfs_root *root)
2603 {
2604         struct btrfs_block_group_cache *cache;
2605         int err = 0;
2606         struct btrfs_path *path;
2607         u64 last = 0;
2608
2609         path = btrfs_alloc_path();
2610         if (!path)
2611                 return -ENOMEM;
2612
2613         while (1) {
2614                 if (last == 0) {
2615                         err = btrfs_run_delayed_refs(trans, root,
2616                                                      (unsigned long)-1);
2617                         BUG_ON(err);
2618                 }
2619
2620                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2621                 while (cache) {
2622                         if (cache->dirty)
2623                                 break;
2624                         cache = next_block_group(root, cache);
2625                 }
2626                 if (!cache) {
2627                         if (last == 0)
2628                                 break;
2629                         last = 0;
2630                         continue;
2631                 }
2632
2633                 cache->dirty = 0;
2634                 last = cache->key.objectid + cache->key.offset;
2635
2636                 err = write_one_cache_group(trans, root, path, cache);
2637                 BUG_ON(err);
2638                 btrfs_put_block_group(cache);
2639         }
2640
2641         btrfs_free_path(path);
2642         return 0;
2643 }
2644
2645 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2646 {
2647         struct btrfs_block_group_cache *block_group;
2648         int readonly = 0;
2649
2650         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2651         if (!block_group || block_group->ro)
2652                 readonly = 1;
2653         if (block_group)
2654                 btrfs_put_block_group(block_group);
2655         return readonly;
2656 }
2657
2658 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2659                              u64 total_bytes, u64 bytes_used,
2660                              struct btrfs_space_info **space_info)
2661 {
2662         struct btrfs_space_info *found;
2663
2664         found = __find_space_info(info, flags);
2665         if (found) {
2666                 spin_lock(&found->lock);
2667                 found->total_bytes += total_bytes;
2668                 found->bytes_used += bytes_used;
2669                 found->full = 0;
2670                 spin_unlock(&found->lock);
2671                 *space_info = found;
2672                 return 0;
2673         }
2674         found = kzalloc(sizeof(*found), GFP_NOFS);
2675         if (!found)
2676                 return -ENOMEM;
2677
2678         INIT_LIST_HEAD(&found->block_groups);
2679         init_rwsem(&found->groups_sem);
2680         spin_lock_init(&found->lock);
2681         found->flags = flags;
2682         found->total_bytes = total_bytes;
2683         found->bytes_used = bytes_used;
2684         found->bytes_pinned = 0;
2685         found->bytes_reserved = 0;
2686         found->bytes_readonly = 0;
2687         found->bytes_delalloc = 0;
2688         found->full = 0;
2689         found->force_alloc = 0;
2690         *space_info = found;
2691         list_add_rcu(&found->list, &info->space_info);
2692         atomic_set(&found->caching_threads, 0);
2693         return 0;
2694 }
2695
2696 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2697 {
2698         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2699                                    BTRFS_BLOCK_GROUP_RAID1 |
2700                                    BTRFS_BLOCK_GROUP_RAID10 |
2701                                    BTRFS_BLOCK_GROUP_DUP);
2702         if (extra_flags) {
2703                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2704                         fs_info->avail_data_alloc_bits |= extra_flags;
2705                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2706                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2707                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2708                         fs_info->avail_system_alloc_bits |= extra_flags;
2709         }
2710 }
2711
2712 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2713 {
2714         spin_lock(&cache->space_info->lock);
2715         spin_lock(&cache->lock);
2716         if (!cache->ro) {
2717                 cache->space_info->bytes_readonly += cache->key.offset -
2718                                         btrfs_block_group_used(&cache->item);
2719                 cache->ro = 1;
2720         }
2721         spin_unlock(&cache->lock);
2722         spin_unlock(&cache->space_info->lock);
2723 }
2724
2725 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2726 {
2727         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2728
2729         if (num_devices == 1)
2730                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2731         if (num_devices < 4)
2732                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2733
2734         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2735             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2736                       BTRFS_BLOCK_GROUP_RAID10))) {
2737                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2738         }
2739
2740         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2741             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2742                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2743         }
2744
2745         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2746             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2747              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2748              (flags & BTRFS_BLOCK_GROUP_DUP)))
2749                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2750         return flags;
2751 }
2752
2753 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2754 {
2755         struct btrfs_fs_info *info = root->fs_info;
2756         u64 alloc_profile;
2757
2758         if (data) {
2759                 alloc_profile = info->avail_data_alloc_bits &
2760                         info->data_alloc_profile;
2761                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2762         } else if (root == root->fs_info->chunk_root) {
2763                 alloc_profile = info->avail_system_alloc_bits &
2764                         info->system_alloc_profile;
2765                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2766         } else {
2767                 alloc_profile = info->avail_metadata_alloc_bits &
2768                         info->metadata_alloc_profile;
2769                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2770         }
2771
2772         return btrfs_reduce_alloc_profile(root, data);
2773 }
2774
2775 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2776 {
2777         u64 alloc_target;
2778
2779         alloc_target = btrfs_get_alloc_profile(root, 1);
2780         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2781                                                        alloc_target);
2782 }
2783
2784 static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
2785 {
2786         u64 num_bytes;
2787         int level;
2788
2789         level = BTRFS_MAX_LEVEL - 2;
2790         /*
2791          * NOTE: these calculations are absolutely the worst possible case.
2792          * This assumes that _every_ item we insert will require a new leaf, and
2793          * that the tree has grown to its maximum level size.
2794          */
2795
2796         /*
2797          * for every item we insert we could insert both an extent item and a
2798          * extent ref item.  Then for ever item we insert, we will need to cow
2799          * both the original leaf, plus the leaf to the left and right of it.
2800          *
2801          * Unless we are talking about the extent root, then we just want the
2802          * number of items * 2, since we just need the extent item plus its ref.
2803          */
2804         if (root == root->fs_info->extent_root)
2805                 num_bytes = num_items * 2;
2806         else
2807                 num_bytes = (num_items + (2 * num_items)) * 3;
2808
2809         /*
2810          * num_bytes is total number of leaves we could need times the leaf
2811          * size, and then for every leaf we could end up cow'ing 2 nodes per
2812          * level, down to the leaf level.
2813          */
2814         num_bytes = (num_bytes * root->leafsize) +
2815                 (num_bytes * (level * 2)) * root->nodesize;
2816
2817         return num_bytes;
2818 }
2819
2820 /*
2821  * Unreserve metadata space for delalloc.  If we have less reserved credits than
2822  * we have extents, this function does nothing.
2823  */
2824 int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2825                                           struct inode *inode, int num_items)
2826 {
2827         struct btrfs_fs_info *info = root->fs_info;
2828         struct btrfs_space_info *meta_sinfo;
2829         u64 num_bytes;
2830         u64 alloc_target;
2831         bool bug = false;
2832
2833         /* get the space info for where the metadata will live */
2834         alloc_target = btrfs_get_alloc_profile(root, 0);
2835         meta_sinfo = __find_space_info(info, alloc_target);
2836
2837         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2838                                            num_items);
2839
2840         spin_lock(&meta_sinfo->lock);
2841         spin_lock(&BTRFS_I(inode)->accounting_lock);
2842         if (BTRFS_I(inode)->reserved_extents <=
2843             BTRFS_I(inode)->outstanding_extents) {
2844                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2845                 spin_unlock(&meta_sinfo->lock);
2846                 return 0;
2847         }
2848         spin_unlock(&BTRFS_I(inode)->accounting_lock);
2849
2850         BTRFS_I(inode)->reserved_extents--;
2851         BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
2852
2853         if (meta_sinfo->bytes_delalloc < num_bytes) {
2854                 bug = true;
2855                 meta_sinfo->bytes_delalloc = 0;
2856         } else {
2857                 meta_sinfo->bytes_delalloc -= num_bytes;
2858         }
2859         spin_unlock(&meta_sinfo->lock);
2860
2861         BUG_ON(bug);
2862
2863         return 0;
2864 }
2865
2866 static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
2867 {
2868         u64 thresh;
2869
2870         thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2871                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2872                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2873                 meta_sinfo->bytes_may_use;
2874
2875         thresh = meta_sinfo->total_bytes - thresh;
2876         thresh *= 80;
2877         do_div(thresh, 100);
2878         if (thresh <= meta_sinfo->bytes_delalloc)
2879                 meta_sinfo->force_delalloc = 1;
2880         else
2881                 meta_sinfo->force_delalloc = 0;
2882 }
2883
2884 struct async_flush {
2885         struct btrfs_root *root;
2886         struct btrfs_space_info *info;
2887         struct btrfs_work work;
2888 };
2889
2890 static noinline void flush_delalloc_async(struct btrfs_work *work)
2891 {
2892         struct async_flush *async;
2893         struct btrfs_root *root;
2894         struct btrfs_space_info *info;
2895
2896         async = container_of(work, struct async_flush, work);
2897         root = async->root;
2898         info = async->info;
2899
2900         btrfs_start_delalloc_inodes(root, 0);
2901         wake_up(&info->flush_wait);
2902         btrfs_wait_ordered_extents(root, 0, 0);
2903
2904         spin_lock(&info->lock);
2905         info->flushing = 0;
2906         spin_unlock(&info->lock);
2907         wake_up(&info->flush_wait);
2908
2909         kfree(async);
2910 }
2911
2912 static void wait_on_flush(struct btrfs_space_info *info)
2913 {
2914         DEFINE_WAIT(wait);
2915         u64 used;
2916
2917         while (1) {
2918                 prepare_to_wait(&info->flush_wait, &wait,
2919                                 TASK_UNINTERRUPTIBLE);
2920                 spin_lock(&info->lock);
2921                 if (!info->flushing) {
2922                         spin_unlock(&info->lock);
2923                         break;
2924                 }
2925
2926                 used = info->bytes_used + info->bytes_reserved +
2927                         info->bytes_pinned + info->bytes_readonly +
2928                         info->bytes_super + info->bytes_root +
2929                         info->bytes_may_use + info->bytes_delalloc;
2930                 if (used < info->total_bytes) {
2931                         spin_unlock(&info->lock);
2932                         break;
2933                 }
2934                 spin_unlock(&info->lock);
2935                 schedule();
2936         }
2937         finish_wait(&info->flush_wait, &wait);
2938 }
2939
2940 static void flush_delalloc(struct btrfs_root *root,
2941                                  struct btrfs_space_info *info)
2942 {
2943         struct async_flush *async;
2944         bool wait = false;
2945
2946         spin_lock(&info->lock);
2947
2948         if (!info->flushing) {
2949                 info->flushing = 1;
2950                 init_waitqueue_head(&info->flush_wait);
2951         } else {
2952                 wait = true;
2953         }
2954
2955         spin_unlock(&info->lock);
2956
2957         if (wait) {
2958                 wait_on_flush(info);
2959                 return;
2960         }
2961
2962         async = kzalloc(sizeof(*async), GFP_NOFS);
2963         if (!async)
2964                 goto flush;
2965
2966         async->root = root;
2967         async->info = info;
2968         async->work.func = flush_delalloc_async;
2969
2970         btrfs_queue_worker(&root->fs_info->enospc_workers,
2971                            &async->work);
2972         wait_on_flush(info);
2973         return;
2974
2975 flush:
2976         btrfs_start_delalloc_inodes(root, 0);
2977         btrfs_wait_ordered_extents(root, 0, 0);
2978
2979         spin_lock(&info->lock);
2980         info->flushing = 0;
2981         spin_unlock(&info->lock);
2982         wake_up(&info->flush_wait);
2983 }
2984
2985 static int maybe_allocate_chunk(struct btrfs_root *root,
2986                                  struct btrfs_space_info *info)
2987 {
2988         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
2989         struct btrfs_trans_handle *trans;
2990         bool wait = false;
2991         int ret = 0;
2992         u64 min_metadata;
2993         u64 free_space;
2994
2995         free_space = btrfs_super_total_bytes(disk_super);
2996         /*
2997          * we allow the metadata to grow to a max of either 10gb or 5% of the
2998          * space in the volume.
2999          */
3000         min_metadata = min((u64)10 * 1024 * 1024 * 1024,
3001                              div64_u64(free_space * 5, 100));
3002         if (info->total_bytes >= min_metadata) {
3003                 spin_unlock(&info->lock);
3004                 return 0;
3005         }
3006
3007         if (info->full) {
3008                 spin_unlock(&info->lock);
3009                 return 0;
3010         }
3011
3012         if (!info->allocating_chunk) {
3013                 info->force_alloc = 1;
3014                 info->allocating_chunk = 1;
3015                 init_waitqueue_head(&info->allocate_wait);
3016         } else {
3017                 wait = true;
3018         }
3019
3020         spin_unlock(&info->lock);
3021
3022         if (wait) {
3023                 wait_event(info->allocate_wait,
3024                            !info->allocating_chunk);
3025                 return 1;
3026         }
3027
3028         trans = btrfs_start_transaction(root, 1);
3029         if (!trans) {
3030                 ret = -ENOMEM;
3031                 goto out;
3032         }
3033
3034         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3035                              4096 + 2 * 1024 * 1024,
3036                              info->flags, 0);
3037         btrfs_end_transaction(trans, root);
3038         if (ret)
3039                 goto out;
3040 out:
3041         spin_lock(&info->lock);
3042         info->allocating_chunk = 0;
3043         spin_unlock(&info->lock);
3044         wake_up(&info->allocate_wait);
3045
3046         if (ret)
3047                 return 0;
3048         return 1;
3049 }
3050
3051 /*
3052  * Reserve metadata space for delalloc.
3053  */
3054 int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
3055                                         struct inode *inode, int num_items)
3056 {
3057         struct btrfs_fs_info *info = root->fs_info;
3058         struct btrfs_space_info *meta_sinfo;
3059         u64 num_bytes;
3060         u64 used;
3061         u64 alloc_target;
3062         int flushed = 0;
3063         int force_delalloc;
3064
3065         /* get the space info for where the metadata will live */
3066         alloc_target = btrfs_get_alloc_profile(root, 0);
3067         meta_sinfo = __find_space_info(info, alloc_target);
3068
3069         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
3070                                            num_items);
3071 again:
3072         spin_lock(&meta_sinfo->lock);
3073
3074         force_delalloc = meta_sinfo->force_delalloc;
3075
3076         if (unlikely(!meta_sinfo->bytes_root))
3077                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3078
3079         if (!flushed)
3080                 meta_sinfo->bytes_delalloc += num_bytes;
3081
3082         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3083                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3084                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3085                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3086
3087         if (used > meta_sinfo->total_bytes) {
3088                 flushed++;
3089
3090                 if (flushed == 1) {
3091                         if (maybe_allocate_chunk(root, meta_sinfo))
3092                                 goto again;
3093                         flushed++;
3094                 } else {
3095                         spin_unlock(&meta_sinfo->lock);
3096                 }
3097
3098                 if (flushed == 2) {
3099                         filemap_flush(inode->i_mapping);
3100                         goto again;
3101                 } else if (flushed == 3) {
3102                         flush_delalloc(root, meta_sinfo);
3103                         goto again;
3104                 }
3105                 spin_lock(&meta_sinfo->lock);
3106                 meta_sinfo->bytes_delalloc -= num_bytes;
3107                 spin_unlock(&meta_sinfo->lock);
3108                 printk(KERN_ERR "enospc, has %d, reserved %d\n",
3109                        BTRFS_I(inode)->outstanding_extents,
3110                        BTRFS_I(inode)->reserved_extents);
3111                 dump_space_info(meta_sinfo, 0, 0);
3112                 return -ENOSPC;
3113         }
3114
3115         BTRFS_I(inode)->reserved_extents++;
3116         check_force_delalloc(meta_sinfo);
3117         spin_unlock(&meta_sinfo->lock);
3118
3119         if (!flushed && force_delalloc)
3120                 filemap_flush(inode->i_mapping);
3121
3122         return 0;
3123 }
3124
3125 /*
3126  * unreserve num_items number of items worth of metadata space.  This needs to
3127  * be paired with btrfs_reserve_metadata_space.
3128  *
3129  * NOTE: if you have the option, run this _AFTER_ you do a
3130  * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
3131  * oprations which will result in more used metadata, so we want to make sure we
3132  * can do that without issue.
3133  */
3134 int btrfs_unreserve_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 alloc_target;
3140         bool bug = false;
3141
3142         /* get the space info for where the metadata will live */
3143         alloc_target = btrfs_get_alloc_profile(root, 0);
3144         meta_sinfo = __find_space_info(info, alloc_target);
3145
3146         num_bytes = calculate_bytes_needed(root, num_items);
3147
3148         spin_lock(&meta_sinfo->lock);
3149         if (meta_sinfo->bytes_may_use < num_bytes) {
3150                 bug = true;
3151                 meta_sinfo->bytes_may_use = 0;
3152         } else {
3153                 meta_sinfo->bytes_may_use -= num_bytes;
3154         }
3155         spin_unlock(&meta_sinfo->lock);
3156
3157         BUG_ON(bug);
3158
3159         return 0;
3160 }
3161
3162 /*
3163  * Reserve some metadata space for use.  We'll calculate the worste case number
3164  * of bytes that would be needed to modify num_items number of items.  If we
3165  * have space, fantastic, if not, you get -ENOSPC.  Please call
3166  * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3167  * items you reserved, since whatever metadata you needed should have already
3168  * been allocated.
3169  *
3170  * This will commit the transaction to make more space if we don't have enough
3171  * metadata space.  THe only time we don't do this is if we're reserving space
3172  * inside of a transaction, then we will just return -ENOSPC and it is the
3173  * callers responsibility to handle it properly.
3174  */
3175 int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
3176 {
3177         struct btrfs_fs_info *info = root->fs_info;
3178         struct btrfs_space_info *meta_sinfo;
3179         u64 num_bytes;
3180         u64 used;
3181         u64 alloc_target;
3182         int retries = 0;
3183
3184         /* get the space info for where the metadata will live */
3185         alloc_target = btrfs_get_alloc_profile(root, 0);
3186         meta_sinfo = __find_space_info(info, alloc_target);
3187
3188         num_bytes = calculate_bytes_needed(root, num_items);
3189 again:
3190         spin_lock(&meta_sinfo->lock);
3191
3192         if (unlikely(!meta_sinfo->bytes_root))
3193                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3194
3195         if (!retries)
3196                 meta_sinfo->bytes_may_use += num_bytes;
3197
3198         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3199                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3200                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3201                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3202
3203         if (used > meta_sinfo->total_bytes) {
3204                 retries++;
3205                 if (retries == 1) {
3206                         if (maybe_allocate_chunk(root, meta_sinfo))
3207                                 goto again;
3208                         retries++;
3209                 } else {
3210                         spin_unlock(&meta_sinfo->lock);
3211                 }
3212
3213                 if (retries == 2) {
3214                         flush_delalloc(root, meta_sinfo);
3215                         goto again;
3216                 }
3217                 spin_lock(&meta_sinfo->lock);
3218                 meta_sinfo->bytes_may_use -= num_bytes;
3219                 spin_unlock(&meta_sinfo->lock);
3220
3221                 dump_space_info(meta_sinfo, 0, 0);
3222                 return -ENOSPC;
3223         }
3224
3225         check_force_delalloc(meta_sinfo);
3226         spin_unlock(&meta_sinfo->lock);
3227
3228         return 0;
3229 }
3230
3231 /*
3232  * This will check the space that the inode allocates from to make sure we have
3233  * enough space for bytes.
3234  */
3235 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
3236                                 u64 bytes)
3237 {
3238         struct btrfs_space_info *data_sinfo;
3239         int ret = 0, committed = 0;
3240
3241         /* make sure bytes are sectorsize aligned */
3242         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3243
3244         data_sinfo = BTRFS_I(inode)->space_info;
3245         if (!data_sinfo)
3246                 goto alloc;
3247
3248 again:
3249         /* make sure we have enough space to handle the data first */
3250         spin_lock(&data_sinfo->lock);
3251         if (data_sinfo->total_bytes - data_sinfo->bytes_used -
3252             data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
3253             data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
3254             data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
3255                 struct btrfs_trans_handle *trans;
3256
3257                 /*
3258                  * if we don't have enough free bytes in this space then we need
3259                  * to alloc a new chunk.
3260                  */
3261                 if (!data_sinfo->full) {
3262                         u64 alloc_target;
3263
3264                         data_sinfo->force_alloc = 1;
3265                         spin_unlock(&data_sinfo->lock);
3266 alloc:
3267                         alloc_target = btrfs_get_alloc_profile(root, 1);
3268                         trans = btrfs_start_transaction(root, 1);
3269                         if (!trans)
3270                                 return -ENOMEM;
3271
3272                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3273                                              bytes + 2 * 1024 * 1024,
3274                                              alloc_target, 0);
3275                         btrfs_end_transaction(trans, root);
3276                         if (ret)
3277                                 return ret;
3278
3279                         if (!data_sinfo) {
3280                                 btrfs_set_inode_space_info(root, inode);
3281                                 data_sinfo = BTRFS_I(inode)->space_info;
3282                         }
3283                         goto again;
3284                 }
3285                 spin_unlock(&data_sinfo->lock);
3286
3287                 /* commit the current transaction and try again */
3288                 if (!committed && !root->fs_info->open_ioctl_trans) {
3289                         committed = 1;
3290                         trans = btrfs_join_transaction(root, 1);
3291                         if (!trans)
3292                                 return -ENOMEM;
3293                         ret = btrfs_commit_transaction(trans, root);
3294                         if (ret)
3295                                 return ret;
3296                         goto again;
3297                 }
3298
3299                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
3300                        ", %llu bytes_used, %llu bytes_reserved, "
3301                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
3302                        "%llu total\n", (unsigned long long)bytes,
3303                        (unsigned long long)data_sinfo->bytes_delalloc,
3304                        (unsigned long long)data_sinfo->bytes_used,
3305                        (unsigned long long)data_sinfo->bytes_reserved,
3306                        (unsigned long long)data_sinfo->bytes_pinned,
3307                        (unsigned long long)data_sinfo->bytes_readonly,
3308                        (unsigned long long)data_sinfo->bytes_may_use,
3309                        (unsigned long long)data_sinfo->total_bytes);
3310                 return -ENOSPC;
3311         }
3312         data_sinfo->bytes_may_use += bytes;
3313         BTRFS_I(inode)->reserved_bytes += bytes;
3314         spin_unlock(&data_sinfo->lock);
3315
3316         return 0;
3317 }
3318
3319 /*
3320  * if there was an error for whatever reason after calling
3321  * btrfs_check_data_free_space, call this so we can cleanup the counters.
3322  */
3323 void btrfs_free_reserved_data_space(struct btrfs_root *root,
3324                                     struct inode *inode, u64 bytes)
3325 {
3326         struct btrfs_space_info *data_sinfo;
3327
3328         /* make sure bytes are sectorsize aligned */
3329         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3330
3331         data_sinfo = BTRFS_I(inode)->space_info;
3332         spin_lock(&data_sinfo->lock);
3333         data_sinfo->bytes_may_use -= bytes;
3334         BTRFS_I(inode)->reserved_bytes -= bytes;
3335         spin_unlock(&data_sinfo->lock);
3336 }
3337
3338 /* called when we are adding a delalloc extent to the inode's io_tree */
3339 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
3340                                   u64 bytes)
3341 {
3342         struct btrfs_space_info *data_sinfo;
3343
3344         /* get the space info for where this inode will be storing its data */
3345         data_sinfo = BTRFS_I(inode)->space_info;
3346
3347         /* make sure we have enough space to handle the data first */
3348         spin_lock(&data_sinfo->lock);
3349         data_sinfo->bytes_delalloc += bytes;
3350
3351         /*
3352          * we are adding a delalloc extent without calling
3353          * btrfs_check_data_free_space first.  This happens on a weird
3354          * writepage condition, but shouldn't hurt our accounting
3355          */
3356         if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
3357                 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
3358                 BTRFS_I(inode)->reserved_bytes = 0;
3359         } else {
3360                 data_sinfo->bytes_may_use -= bytes;
3361                 BTRFS_I(inode)->reserved_bytes -= bytes;
3362         }
3363
3364         spin_unlock(&data_sinfo->lock);
3365 }
3366
3367 /* called when we are clearing an delalloc extent from the inode's io_tree */
3368 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
3369                               u64 bytes)
3370 {
3371         struct btrfs_space_info *info;
3372
3373         info = BTRFS_I(inode)->space_info;
3374
3375         spin_lock(&info->lock);
3376         info->bytes_delalloc -= bytes;
3377         spin_unlock(&info->lock);
3378 }
3379
3380 static void force_metadata_allocation(struct btrfs_fs_info *info)
3381 {
3382         struct list_head *head = &info->space_info;
3383         struct btrfs_space_info *found;
3384
3385         rcu_read_lock();
3386         list_for_each_entry_rcu(found, head, list) {
3387                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3388                         found->force_alloc = 1;
3389         }
3390         rcu_read_unlock();
3391 }
3392
3393 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3394                           struct btrfs_root *extent_root, u64 alloc_bytes,
3395                           u64 flags, int force)
3396 {
3397         struct btrfs_space_info *space_info;
3398         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3399         u64 thresh;
3400         int ret = 0;
3401
3402         mutex_lock(&fs_info->chunk_mutex);
3403
3404         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3405
3406         space_info = __find_space_info(extent_root->fs_info, flags);
3407         if (!space_info) {
3408                 ret = update_space_info(extent_root->fs_info, flags,
3409                                         0, 0, &space_info);
3410                 BUG_ON(ret);
3411         }
3412         BUG_ON(!space_info);
3413
3414         spin_lock(&space_info->lock);
3415         if (space_info->force_alloc)
3416                 force = 1;
3417         if (space_info->full) {
3418                 spin_unlock(&space_info->lock);
3419                 goto out;
3420         }
3421
3422         thresh = space_info->total_bytes - space_info->bytes_readonly;
3423         thresh = div_factor(thresh, 8);
3424         if (!force &&
3425            (space_info->bytes_used + space_info->bytes_pinned +
3426             space_info->bytes_reserved + alloc_bytes) < thresh) {
3427                 spin_unlock(&space_info->lock);
3428                 goto out;
3429         }
3430         spin_unlock(&space_info->lock);
3431
3432         /*
3433          * if we're doing a data chunk, go ahead and make sure that
3434          * we keep a reasonable number of metadata chunks allocated in the
3435          * FS as well.
3436          */
3437         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3438                 fs_info->data_chunk_allocations++;
3439                 if (!(fs_info->data_chunk_allocations %
3440                       fs_info->metadata_ratio))
3441                         force_metadata_allocation(fs_info);
3442         }
3443
3444         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3445         spin_lock(&space_info->lock);
3446         if (ret)
3447                 space_info->full = 1;
3448         space_info->force_alloc = 0;
3449         spin_unlock(&space_info->lock);
3450 out:
3451         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3452         return ret;
3453 }
3454
3455 static int update_block_group(struct btrfs_trans_handle *trans,
3456                               struct btrfs_root *root,
3457                               u64 bytenr, u64 num_bytes, int alloc,
3458                               int mark_free)
3459 {
3460         struct btrfs_block_group_cache *cache;
3461         struct btrfs_fs_info *info = root->fs_info;
3462         u64 total = num_bytes;
3463         u64 old_val;
3464         u64 byte_in_group;
3465
3466         /* block accounting for super block */
3467         spin_lock(&info->delalloc_lock);
3468         old_val = btrfs_super_bytes_used(&info->super_copy);
3469         if (alloc)
3470                 old_val += num_bytes;
3471         else
3472                 old_val -= num_bytes;
3473         btrfs_set_super_bytes_used(&info->super_copy, old_val);
3474         spin_unlock(&info->delalloc_lock);
3475
3476         while (total) {
3477                 cache = btrfs_lookup_block_group(info, bytenr);
3478                 if (!cache)
3479                         return -1;
3480                 byte_in_group = bytenr - cache->key.objectid;
3481                 WARN_ON(byte_in_group > cache->key.offset);
3482
3483                 spin_lock(&cache->space_info->lock);
3484                 spin_lock(&cache->lock);
3485                 cache->dirty = 1;
3486                 old_val = btrfs_block_group_used(&cache->item);
3487                 num_bytes = min(total, cache->key.offset - byte_in_group);
3488                 if (alloc) {
3489                         old_val += num_bytes;
3490                         btrfs_set_block_group_used(&cache->item, old_val);
3491                         cache->reserved -= num_bytes;
3492                         cache->space_info->bytes_used += num_bytes;
3493                         cache->space_info->bytes_reserved -= num_bytes;
3494                         if (cache->ro)
3495                                 cache->space_info->bytes_readonly -= num_bytes;
3496                         spin_unlock(&cache->lock);
3497                         spin_unlock(&cache->space_info->lock);
3498                 } else {
3499                         old_val -= num_bytes;
3500                         cache->space_info->bytes_used -= num_bytes;
3501                         if (cache->ro)
3502                                 cache->space_info->bytes_readonly += num_bytes;
3503                         btrfs_set_block_group_used(&cache->item, old_val);
3504                         spin_unlock(&cache->lock);
3505                         spin_unlock(&cache->space_info->lock);
3506                         if (mark_free) {
3507                                 int ret;
3508
3509                                 ret = btrfs_discard_extent(root, bytenr,
3510                                                            num_bytes);
3511                                 WARN_ON(ret);
3512
3513                                 ret = btrfs_add_free_space(cache, bytenr,
3514                                                            num_bytes);
3515                                 WARN_ON(ret);
3516                         }
3517                 }
3518                 btrfs_put_block_group(cache);
3519                 total -= num_bytes;
3520                 bytenr += num_bytes;
3521         }
3522         return 0;
3523 }
3524
3525 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3526 {
3527         struct btrfs_block_group_cache *cache;
3528         u64 bytenr;
3529
3530         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3531         if (!cache)
3532                 return 0;
3533
3534         bytenr = cache->key.objectid;
3535         btrfs_put_block_group(cache);
3536
3537         return bytenr;
3538 }
3539
3540 /*
3541  * this function must be called within transaction
3542  */
3543 int btrfs_pin_extent(struct btrfs_root *root,
3544                      u64 bytenr, u64 num_bytes, int reserved)
3545 {
3546         struct btrfs_fs_info *fs_info = root->fs_info;
3547         struct btrfs_block_group_cache *cache;
3548
3549         cache = btrfs_lookup_block_group(fs_info, bytenr);
3550         BUG_ON(!cache);
3551
3552         spin_lock(&cache->space_info->lock);
3553         spin_lock(&cache->lock);
3554         cache->pinned += num_bytes;
3555         cache->space_info->bytes_pinned += num_bytes;
3556         if (reserved) {
3557                 cache->reserved -= num_bytes;
3558                 cache->space_info->bytes_reserved -= num_bytes;
3559         }
3560         spin_unlock(&cache->lock);
3561         spin_unlock(&cache->space_info->lock);
3562
3563         btrfs_put_block_group(cache);
3564
3565         set_extent_dirty(fs_info->pinned_extents,
3566                          bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3567         return 0;
3568 }
3569
3570 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3571                                    u64 num_bytes, int reserve)
3572 {
3573         spin_lock(&cache->space_info->lock);
3574         spin_lock(&cache->lock);
3575         if (reserve) {
3576                 cache->reserved += num_bytes;
3577                 cache->space_info->bytes_reserved += num_bytes;
3578         } else {
3579                 cache->reserved -= num_bytes;
3580                 cache->space_info->bytes_reserved -= num_bytes;
3581         }
3582         spin_unlock(&cache->lock);
3583         spin_unlock(&cache->space_info->lock);
3584         return 0;
3585 }
3586
3587 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3588                                 struct btrfs_root *root)
3589 {
3590         struct btrfs_fs_info *fs_info = root->fs_info;
3591         struct btrfs_caching_control *next;
3592         struct btrfs_caching_control *caching_ctl;
3593         struct btrfs_block_group_cache *cache;
3594
3595         down_write(&fs_info->extent_commit_sem);
3596
3597         list_for_each_entry_safe(caching_ctl, next,
3598                                  &fs_info->caching_block_groups, list) {
3599                 cache = caching_ctl->block_group;
3600                 if (block_group_cache_done(cache)) {
3601                         cache->last_byte_to_unpin = (u64)-1;
3602                         list_del_init(&caching_ctl->list);
3603                         put_caching_control(caching_ctl);
3604                 } else {
3605                         cache->last_byte_to_unpin = caching_ctl->progress;
3606                 }
3607         }
3608
3609         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3610                 fs_info->pinned_extents = &fs_info->freed_extents[1];
3611         else
3612                 fs_info->pinned_extents = &fs_info->freed_extents[0];
3613
3614         up_write(&fs_info->extent_commit_sem);
3615         return 0;
3616 }
3617
3618 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3619 {
3620         struct btrfs_fs_info *fs_info = root->fs_info;
3621         struct btrfs_block_group_cache *cache = NULL;
3622         u64 len;
3623
3624         while (start <= end) {
3625                 if (!cache ||
3626                     start >= cache->key.objectid + cache->key.offset) {
3627                         if (cache)
3628                                 btrfs_put_block_group(cache);
3629                         cache = btrfs_lookup_block_group(fs_info, start);
3630                         BUG_ON(!cache);
3631                 }
3632
3633                 len = cache->key.objectid + cache->key.offset - start;
3634                 len = min(len, end + 1 - start);
3635
3636                 if (start < cache->last_byte_to_unpin) {
3637                         len = min(len, cache->last_byte_to_unpin - start);
3638                         btrfs_add_free_space(cache, start, len);
3639                 }
3640
3641                 spin_lock(&cache->space_info->lock);
3642                 spin_lock(&cache->lock);
3643                 cache->pinned -= len;
3644                 cache->space_info->bytes_pinned -= len;
3645                 spin_unlock(&cache->lock);
3646                 spin_unlock(&cache->space_info->lock);
3647
3648                 start += len;
3649         }
3650
3651         if (cache)
3652                 btrfs_put_block_group(cache);
3653         return 0;
3654 }
3655
3656 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3657                                struct btrfs_root *root)
3658 {
3659         struct btrfs_fs_info *fs_info = root->fs_info;
3660         struct extent_io_tree *unpin;
3661         u64 start;
3662         u64 end;
3663         int ret;
3664
3665         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3666                 unpin = &fs_info->freed_extents[1];
3667         else
3668                 unpin = &fs_info->freed_extents[0];
3669
3670         while (1) {
3671                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3672                                             EXTENT_DIRTY);
3673                 if (ret)
3674                         break;
3675
3676                 ret = btrfs_discard_extent(root, start, end + 1 - start);
3677
3678                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3679                 unpin_extent_range(root, start, end);
3680                 cond_resched();
3681         }
3682
3683         return ret;
3684 }
3685
3686 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3687                           struct btrfs_root *root,
3688                           struct btrfs_path *path,
3689                           u64 bytenr, u64 num_bytes,
3690                           int is_data, int reserved,
3691                           struct extent_buffer **must_clean)
3692 {
3693         int err = 0;
3694         struct extent_buffer *buf;
3695
3696         if (is_data)
3697                 goto pinit;
3698
3699         /*
3700          * discard is sloooow, and so triggering discards on
3701          * individual btree blocks isn't a good plan.  Just
3702          * pin everything in discard mode.
3703          */
3704         if (btrfs_test_opt(root, DISCARD))
3705                 goto pinit;
3706
3707         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3708         if (!buf)
3709                 goto pinit;
3710
3711         /* we can reuse a block if it hasn't been written
3712          * and it is from this transaction.  We can't
3713          * reuse anything from the tree log root because
3714          * it has tiny sub-transactions.
3715          */
3716         if (btrfs_buffer_uptodate(buf, 0) &&
3717             btrfs_try_tree_lock(buf)) {
3718                 u64 header_owner = btrfs_header_owner(buf);
3719                 u64 header_transid = btrfs_header_generation(buf);
3720                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3721                     header_transid == trans->transid &&
3722                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3723                         *must_clean = buf;
3724                         return 1;
3725                 }
3726                 btrfs_tree_unlock(buf);
3727         }
3728         free_extent_buffer(buf);
3729 pinit:
3730         if (path)
3731                 btrfs_set_path_blocking(path);
3732         /* unlocks the pinned mutex */
3733         btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3734
3735         BUG_ON(err < 0);
3736         return 0;
3737 }
3738
3739 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3740                                 struct btrfs_root *root,
3741                                 u64 bytenr, u64 num_bytes, u64 parent,
3742                                 u64 root_objectid, u64 owner_objectid,
3743                                 u64 owner_offset, int refs_to_drop,
3744                                 struct btrfs_delayed_extent_op *extent_op)
3745 {
3746         struct btrfs_key key;
3747         struct btrfs_path *path;
3748         struct btrfs_fs_info *info = root->fs_info;
3749         struct btrfs_root *extent_root = info->extent_root;
3750         struct extent_buffer *leaf;
3751         struct btrfs_extent_item *ei;
3752         struct btrfs_extent_inline_ref *iref;
3753         int ret;
3754         int is_data;
3755         int extent_slot = 0;
3756         int found_extent = 0;
3757         int num_to_del = 1;
3758         u32 item_size;
3759         u64 refs;
3760
3761         path = btrfs_alloc_path();
3762         if (!path)
3763                 return -ENOMEM;
3764
3765         path->reada = 1;
3766         path->leave_spinning = 1;
3767
3768         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3769         BUG_ON(!is_data && refs_to_drop != 1);
3770
3771         ret = lookup_extent_backref(trans, extent_root, path, &iref,
3772                                     bytenr, num_bytes, parent,
3773                                     root_objectid, owner_objectid,
3774                                     owner_offset);
3775         if (ret == 0) {
3776                 extent_slot = path->slots[0];
3777                 while (extent_slot >= 0) {
3778                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3779                                               extent_slot);
3780                         if (key.objectid != bytenr)
3781                                 break;
3782                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3783                             key.offset == num_bytes) {
3784                                 found_extent = 1;
3785                                 break;
3786                         }
3787                         if (path->slots[0] - extent_slot > 5)
3788                                 break;
3789                         extent_slot--;
3790                 }
3791 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3792                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3793                 if (found_extent && item_size < sizeof(*ei))
3794                         found_extent = 0;
3795 #endif
3796                 if (!found_extent) {
3797                         BUG_ON(iref);
3798                         ret = remove_extent_backref(trans, extent_root, path,
3799                                                     NULL, refs_to_drop,
3800                                                     is_data);
3801                         BUG_ON(ret);
3802                         btrfs_release_path(extent_root, path);
3803                         path->leave_spinning = 1;
3804
3805                         key.objectid = bytenr;
3806                         key.type = BTRFS_EXTENT_ITEM_KEY;
3807                         key.offset = num_bytes;
3808
3809                         ret = btrfs_search_slot(trans, extent_root,
3810                                                 &key, path, -1, 1);
3811                         if (ret) {
3812                                 printk(KERN_ERR "umm, got %d back from search"
3813                                        ", was looking for %llu\n", ret,
3814                                        (unsigned long long)bytenr);
3815                                 btrfs_print_leaf(extent_root, path->nodes[0]);
3816                         }
3817                         BUG_ON(ret);
3818                         extent_slot = path->slots[0];
3819                 }
3820         } else {
3821                 btrfs_print_leaf(extent_root, path->nodes[0]);
3822                 WARN_ON(1);
3823                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3824                        "parent %llu root %llu  owner %llu offset %llu\n",
3825                        (unsigned long long)bytenr,
3826                        (unsigned long long)parent,
3827                        (unsigned long long)root_objectid,
3828                        (unsigned long long)owner_objectid,
3829                        (unsigned long long)owner_offset);
3830         }
3831
3832         leaf = path->nodes[0];
3833         item_size = btrfs_item_size_nr(leaf, extent_slot);
3834 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3835         if (item_size < sizeof(*ei)) {
3836                 BUG_ON(found_extent || extent_slot != path->slots[0]);
3837                 ret = convert_extent_item_v0(trans, extent_root, path,
3838                                              owner_objectid, 0);
3839                 BUG_ON(ret < 0);
3840
3841                 btrfs_release_path(extent_root, path);
3842                 path->leave_spinning = 1;
3843
3844                 key.objectid = bytenr;
3845                 key.type = BTRFS_EXTENT_ITEM_KEY;
3846                 key.offset = num_bytes;
3847
3848                 ret = btrfs_search_slot(trans, extent_root, &key, path,
3849                                         -1, 1);
3850                 if (ret) {
3851                         printk(KERN_ERR "umm, got %d back from search"
3852                                ", was looking for %llu\n", ret,
3853                                (unsigned long long)bytenr);
3854                         btrfs_print_leaf(extent_root, path->nodes[0]);
3855                 }
3856                 BUG_ON(ret);
3857                 extent_slot = path->slots[0];
3858                 leaf = path->nodes[0];
3859                 item_size = btrfs_item_size_nr(leaf, extent_slot);
3860         }
3861 #endif
3862         BUG_ON(item_size < sizeof(*ei));
3863         ei = btrfs_item_ptr(leaf, extent_slot,
3864                             struct btrfs_extent_item);
3865         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3866                 struct btrfs_tree_block_info *bi;
3867                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3868                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3869                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3870         }
3871
3872         refs = btrfs_extent_refs(leaf, ei);
3873         BUG_ON(refs < refs_to_drop);
3874         refs -= refs_to_drop;
3875
3876         if (refs > 0) {
3877                 if (extent_op)
3878                         __run_delayed_extent_op(extent_op, leaf, ei);
3879                 /*
3880                  * In the case of inline back ref, reference count will
3881                  * be updated by remove_extent_backref
3882                  */
3883                 if (iref) {
3884                         BUG_ON(!found_extent);
3885                 } else {
3886                         btrfs_set_extent_refs(leaf, ei, refs);
3887                         btrfs_mark_buffer_dirty(leaf);
3888                 }
3889                 if (found_extent) {
3890                         ret = remove_extent_backref(trans, extent_root, path,
3891                                                     iref, refs_to_drop,
3892                                                     is_data);
3893                         BUG_ON(ret);
3894                 }
3895         } else {
3896                 int mark_free = 0;
3897                 struct extent_buffer *must_clean = NULL;
3898
3899                 if (found_extent) {
3900                         BUG_ON(is_data && refs_to_drop !=
3901                                extent_data_ref_count(root, path, iref));
3902                         if (iref) {
3903                                 BUG_ON(path->slots[0] != extent_slot);
3904                         } else {
3905                                 BUG_ON(path->slots[0] != extent_slot + 1);
3906                                 path->slots[0] = extent_slot;
3907                                 num_to_del = 2;
3908                         }
3909                 }
3910
3911                 ret = pin_down_bytes(trans, root, path, bytenr,
3912                                      num_bytes, is_data, 0, &must_clean);
3913                 if (ret > 0)
3914                         mark_free = 1;
3915                 BUG_ON(ret < 0);
3916                 /*
3917                  * it is going to be very rare for someone to be waiting
3918                  * on the block we're freeing.  del_items might need to
3919                  * schedule, so rather than get fancy, just force it
3920                  * to blocking here
3921                  */
3922                 if (must_clean)
3923                         btrfs_set_lock_blocking(must_clean);
3924
3925                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3926                                       num_to_del);
3927                 BUG_ON(ret);
3928                 btrfs_release_path(extent_root, path);
3929
3930                 if (must_clean) {
3931                         clean_tree_block(NULL, root, must_clean);
3932                         btrfs_tree_unlock(must_clean);
3933                         free_extent_buffer(must_clean);
3934                 }
3935
3936                 if (is_data) {
3937                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3938                         BUG_ON(ret);
3939                 } else {
3940                         invalidate_mapping_pages(info->btree_inode->i_mapping,
3941                              bytenr >> PAGE_CACHE_SHIFT,
3942                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3943                 }
3944
3945                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3946                                          mark_free);
3947                 BUG_ON(ret);
3948         }
3949         btrfs_free_path(path);
3950         return ret;
3951 }
3952
3953 /*
3954  * when we free an extent, it is possible (and likely) that we free the last
3955  * delayed ref for that extent as well.  This searches the delayed ref tree for
3956  * a given extent, and if there are no other delayed refs to be processed, it
3957  * removes it from the tree.
3958  */
3959 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3960                                       struct btrfs_root *root, u64 bytenr)
3961 {
3962         struct btrfs_delayed_ref_head *head;
3963         struct btrfs_delayed_ref_root *delayed_refs;
3964         struct btrfs_delayed_ref_node *ref;
3965         struct rb_node *node;
3966         int ret;
3967
3968         delayed_refs = &trans->transaction->delayed_refs;
3969         spin_lock(&delayed_refs->lock);
3970         head = btrfs_find_delayed_ref_head(trans, bytenr);
3971         if (!head)
3972                 goto out;
3973
3974         node = rb_prev(&head->node.rb_node);
3975         if (!node)
3976                 goto out;
3977
3978         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3979
3980         /* there are still entries for this ref, we can't drop it */
3981         if (ref->bytenr == bytenr)
3982                 goto out;
3983
3984         if (head->extent_op) {
3985                 if (!head->must_insert_reserved)
3986                         goto out;
3987                 kfree(head->extent_op);
3988                 head->extent_op = NULL;
3989         }
3990
3991         /*
3992          * waiting for the lock here would deadlock.  If someone else has it
3993          * locked they are already in the process of dropping it anyway
3994          */
3995         if (!mutex_trylock(&head->mutex))
3996                 goto out;
3997
3998         /*
3999          * at this point we have a head with no other entries.  Go
4000          * ahead and process it.
4001          */
4002         head->node.in_tree = 0;
4003         rb_erase(&head->node.rb_node, &delayed_refs->root);
4004
4005         delayed_refs->num_entries--;
4006
4007         /*
4008          * we don't take a ref on the node because we're removing it from the
4009          * tree, so we just steal the ref the tree was holding.
4010          */
4011         delayed_refs->num_heads--;
4012         if (list_empty(&head->cluster))
4013                 delayed_refs->num_heads_ready--;
4014
4015         list_del_init(&head->cluster);
4016         spin_unlock(&delayed_refs->lock);
4017
4018         ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
4019                                   &head->node, head->extent_op,
4020                                   head->must_insert_reserved);
4021         BUG_ON(ret);
4022         btrfs_put_delayed_ref(&head->node);
4023         return 0;
4024 out:
4025         spin_unlock(&delayed_refs->lock);
4026         return 0;
4027 }
4028
4029 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4030                       struct btrfs_root *root,
4031                       u64 bytenr, u64 num_bytes, u64 parent,
4032                       u64 root_objectid, u64 owner, u64 offset)
4033 {
4034         int ret;
4035
4036         /*
4037          * tree log blocks never actually go into the extent allocation
4038          * tree, just update pinning info and exit early.
4039          */
4040         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4041                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4042                 /* unlocks the pinned mutex */
4043                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4044                 ret = 0;
4045         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4046                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4047                                         parent, root_objectid, (int)owner,
4048                                         BTRFS_DROP_DELAYED_REF, NULL);
4049                 BUG_ON(ret);
4050                 ret = check_ref_cleanup(trans, root, bytenr);
4051                 BUG_ON(ret);
4052         } else {
4053                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4054                                         parent, root_objectid, owner,
4055                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4056                 BUG_ON(ret);
4057         }
4058         return ret;
4059 }
4060
4061 int btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4062                           struct btrfs_root *root,
4063                           u64 bytenr, u32 blocksize,
4064                           u64 parent, u64 root_objectid, int level)
4065 {
4066         u64 used;
4067         spin_lock(&root->node_lock);
4068         used = btrfs_root_used(&root->root_item) - blocksize;
4069         btrfs_set_root_used(&root->root_item, used);
4070         spin_unlock(&root->node_lock);
4071
4072         return btrfs_free_extent(trans, root, bytenr, blocksize,
4073                                  parent, root_objectid, level, 0);
4074 }
4075
4076 static u64 stripe_align(struct btrfs_root *root, u64 val)
4077 {
4078         u64 mask = ((u64)root->stripesize - 1);
4079         u64 ret = (val + mask) & ~mask;
4080         return ret;
4081 }
4082
4083 /*
4084  * when we wait for progress in the block group caching, its because
4085  * our allocation attempt failed at least once.  So, we must sleep
4086  * and let some progress happen before we try again.
4087  *
4088  * This function will sleep at least once waiting for new free space to
4089  * show up, and then it will check the block group free space numbers
4090  * for our min num_bytes.  Another option is to have it go ahead
4091  * and look in the rbtree for a free extent of a given size, but this
4092  * is a good start.
4093  */
4094 static noinline int
4095 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4096                                 u64 num_bytes)
4097 {
4098         struct btrfs_caching_control *caching_ctl;
4099         DEFINE_WAIT(wait);
4100
4101         caching_ctl = get_caching_control(cache);
4102         if (!caching_ctl)
4103                 return 0;
4104
4105         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4106                    (cache->free_space >= num_bytes));
4107
4108         put_caching_control(caching_ctl);
4109         return 0;
4110 }
4111
4112 static noinline int
4113 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4114 {
4115         struct btrfs_caching_control *caching_ctl;
4116         DEFINE_WAIT(wait);
4117
4118         caching_ctl = get_caching_control(cache);
4119         if (!caching_ctl)
4120                 return 0;
4121
4122         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4123
4124         put_caching_control(caching_ctl);
4125         return 0;
4126 }
4127
4128 enum btrfs_loop_type {
4129         LOOP_FIND_IDEAL = 0,
4130         LOOP_CACHING_NOWAIT = 1,
4131         LOOP_CACHING_WAIT = 2,
4132         LOOP_ALLOC_CHUNK = 3,
4133         LOOP_NO_EMPTY_SIZE = 4,
4134 };
4135
4136 /*
4137  * walks the btree of allocated extents and find a hole of a given size.
4138  * The key ins is changed to record the hole:
4139  * ins->objectid == block start
4140  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4141  * ins->offset == number of blocks
4142  * Any available blocks before search_start are skipped.
4143  */
4144 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4145                                      struct btrfs_root *orig_root,
4146                                      u64 num_bytes, u64 empty_size,
4147                                      u64 search_start, u64 search_end,
4148                                      u64 hint_byte, struct btrfs_key *ins,
4149                                      u64 exclude_start, u64 exclude_nr,
4150                                      int data)
4151 {
4152         int ret = 0;
4153         struct btrfs_root *root = orig_root->fs_info->extent_root;
4154         struct btrfs_free_cluster *last_ptr = NULL;
4155         struct btrfs_block_group_cache *block_group = NULL;
4156         int empty_cluster = 2 * 1024 * 1024;
4157         int allowed_chunk_alloc = 0;
4158         int done_chunk_alloc = 0;
4159         struct btrfs_space_info *space_info;
4160         int last_ptr_loop = 0;
4161         int loop = 0;
4162         bool found_uncached_bg = false;
4163         bool failed_cluster_refill = false;
4164         bool failed_alloc = false;
4165         u64 ideal_cache_percent = 0;
4166         u64 ideal_cache_offset = 0;
4167
4168         WARN_ON(num_bytes < root->sectorsize);
4169         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4170         ins->objectid = 0;
4171         ins->offset = 0;
4172
4173         space_info = __find_space_info(root->fs_info, data);
4174
4175         if (orig_root->ref_cows || empty_size)
4176                 allowed_chunk_alloc = 1;
4177
4178         if (data & BTRFS_BLOCK_GROUP_METADATA) {
4179                 last_ptr = &root->fs_info->meta_alloc_cluster;
4180                 if (!btrfs_test_opt(root, SSD))
4181                         empty_cluster = 64 * 1024;
4182         }
4183
4184         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4185                 last_ptr = &root->fs_info->data_alloc_cluster;
4186         }
4187
4188         if (last_ptr) {
4189                 spin_lock(&last_ptr->lock);
4190                 if (last_ptr->block_group)
4191                         hint_byte = last_ptr->window_start;
4192                 spin_unlock(&last_ptr->lock);
4193         }
4194
4195         search_start = max(search_start, first_logical_byte(root, 0));
4196         search_start = max(search_start, hint_byte);
4197
4198         if (!last_ptr)
4199                 empty_cluster = 0;
4200
4201         if (search_start == hint_byte) {
4202 ideal_cache:
4203                 block_group = btrfs_lookup_block_group(root->fs_info,
4204                                                        search_start);
4205                 /*
4206                  * we don't want to use the block group if it doesn't match our
4207                  * allocation bits, or if its not cached.
4208                  *
4209                  * However if we are re-searching with an ideal block group
4210                  * picked out then we don't care that the block group is cached.
4211                  */
4212                 if (block_group && block_group_bits(block_group, data) &&
4213                     (block_group->cached != BTRFS_CACHE_NO ||
4214                      search_start == ideal_cache_offset)) {
4215                         down_read(&space_info->groups_sem);
4216                         if (list_empty(&block_group->list) ||
4217                             block_group->ro) {
4218                                 /*
4219                                  * someone is removing this block group,
4220                                  * we can't jump into the have_block_group
4221                                  * target because our list pointers are not
4222                                  * valid
4223                                  */
4224                                 btrfs_put_block_group(block_group);
4225                                 up_read(&space_info->groups_sem);
4226                         } else {
4227                                 goto have_block_group;
4228                         }
4229                 } else if (block_group) {
4230                         btrfs_put_block_group(block_group);
4231                 }
4232         }
4233 search:
4234         down_read(&space_info->groups_sem);
4235         list_for_each_entry(block_group, &space_info->block_groups, list) {
4236                 u64 offset;
4237                 int cached;
4238
4239                 btrfs_get_block_group(block_group);
4240                 search_start = block_group->key.objectid;
4241
4242 have_block_group:
4243                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4244                         u64 free_percent;
4245
4246                         free_percent = btrfs_block_group_used(&block_group->item);
4247                         free_percent *= 100;
4248                         free_percent = div64_u64(free_percent,
4249                                                  block_group->key.offset);
4250                         free_percent = 100 - free_percent;
4251                         if (free_percent > ideal_cache_percent &&
4252                             likely(!block_group->ro)) {
4253                                 ideal_cache_offset = block_group->key.objectid;
4254                                 ideal_cache_percent = free_percent;
4255                         }
4256
4257                         /*
4258                          * We only want to start kthread caching if we are at
4259                          * the point where we will wait for caching to make
4260                          * progress, or if our ideal search is over and we've
4261                          * found somebody to start caching.
4262                          */
4263                         if (loop > LOOP_CACHING_NOWAIT ||
4264                             (loop > LOOP_FIND_IDEAL &&
4265                              atomic_read(&space_info->caching_threads) < 2)) {
4266                                 ret = cache_block_group(block_group);
4267                                 BUG_ON(ret);
4268                         }
4269                         found_uncached_bg = true;
4270
4271                         /*
4272                          * If loop is set for cached only, try the next block
4273                          * group.
4274                          */
4275                         if (loop == LOOP_FIND_IDEAL)
4276                                 goto loop;
4277                 }
4278
4279                 cached = block_group_cache_done(block_group);
4280                 if (unlikely(!cached))
4281                         found_uncached_bg = true;
4282
4283                 if (unlikely(block_group->ro))
4284                         goto loop;
4285
4286                 /*
4287                  * Ok we want to try and use the cluster allocator, so lets look
4288                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4289                  * have tried the cluster allocator plenty of times at this
4290                  * point and not have found anything, so we are likely way too
4291                  * fragmented for the clustering stuff to find anything, so lets
4292                  * just skip it and let the allocator find whatever block it can
4293                  * find
4294                  */
4295                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4296                         /*
4297                          * the refill lock keeps out other
4298                          * people trying to start a new cluster
4299                          */
4300                         spin_lock(&last_ptr->refill_lock);
4301                         if (last_ptr->block_group &&
4302                             (last_ptr->block_group->ro ||
4303                             !block_group_bits(last_ptr->block_group, data))) {
4304                                 offset = 0;
4305                                 goto refill_cluster;
4306                         }
4307
4308                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4309                                                  num_bytes, search_start);
4310                         if (offset) {
4311                                 /* we have a block, we're done */
4312                                 spin_unlock(&last_ptr->refill_lock);
4313                                 goto checks;
4314                         }
4315
4316                         spin_lock(&last_ptr->lock);
4317                         /*
4318                          * whoops, this cluster doesn't actually point to
4319                          * this block group.  Get a ref on the block
4320                          * group is does point to and try again
4321                          */
4322                         if (!last_ptr_loop && last_ptr->block_group &&
4323                             last_ptr->block_group != block_group) {
4324
4325                                 btrfs_put_block_group(block_group);
4326                                 block_group = last_ptr->block_group;
4327                                 btrfs_get_block_group(block_group);
4328                                 spin_unlock(&last_ptr->lock);
4329                                 spin_unlock(&last_ptr->refill_lock);
4330
4331                                 last_ptr_loop = 1;
4332                                 search_start = block_group->key.objectid;
4333                                 /*
4334                                  * we know this block group is properly
4335                                  * in the list because
4336                                  * btrfs_remove_block_group, drops the
4337                                  * cluster before it removes the block
4338                                  * group from the list
4339                                  */
4340                                 goto have_block_group;
4341                         }
4342                         spin_unlock(&last_ptr->lock);
4343 refill_cluster:
4344                         /*
4345                          * this cluster didn't work out, free it and
4346                          * start over
4347                          */
4348                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4349
4350                         last_ptr_loop = 0;
4351
4352                         /* allocate a cluster in this block group */
4353                         ret = btrfs_find_space_cluster(trans, root,
4354                                                block_group, last_ptr,
4355                                                offset, num_bytes,
4356                                                empty_cluster + empty_size);
4357                         if (ret == 0) {
4358                                 /*
4359                                  * now pull our allocation out of this
4360                                  * cluster
4361                                  */
4362                                 offset = btrfs_alloc_from_cluster(block_group,
4363                                                   last_ptr, num_bytes,
4364                                                   search_start);
4365                                 if (offset) {
4366                                         /* we found one, proceed */
4367                                         spin_unlock(&last_ptr->refill_lock);
4368                                         goto checks;
4369                                 }
4370                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
4371                                    && !failed_cluster_refill) {
4372                                 spin_unlock(&last_ptr->refill_lock);
4373
4374                                 failed_cluster_refill = true;
4375                                 wait_block_group_cache_progress(block_group,
4376                                        num_bytes + empty_cluster + empty_size);
4377                                 goto have_block_group;
4378                         }
4379
4380                         /*
4381                          * at this point we either didn't find a cluster
4382                          * or we weren't able to allocate a block from our
4383                          * cluster.  Free the cluster we've been trying
4384                          * to use, and go to the next block group
4385                          */
4386                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4387                         spin_unlock(&last_ptr->refill_lock);
4388                         goto loop;
4389                 }
4390
4391                 offset = btrfs_find_space_for_alloc(block_group, search_start,
4392                                                     num_bytes, empty_size);
4393                 /*
4394                  * If we didn't find a chunk, and we haven't failed on this
4395                  * block group before, and this block group is in the middle of
4396                  * caching and we are ok with waiting, then go ahead and wait
4397                  * for progress to be made, and set failed_alloc to true.
4398                  *
4399                  * If failed_alloc is true then we've already waited on this
4400                  * block group once and should move on to the next block group.
4401                  */
4402                 if (!offset && !failed_alloc && !cached &&
4403                     loop > LOOP_CACHING_NOWAIT) {
4404                         wait_block_group_cache_progress(block_group,
4405                                                 num_bytes + empty_size);
4406                         failed_alloc = true;
4407                         goto have_block_group;
4408                 } else if (!offset) {
4409                         goto loop;
4410                 }
4411 checks:
4412                 search_start = stripe_align(root, offset);
4413                 /* move on to the next group */
4414                 if (search_start + num_bytes >= search_end) {
4415                         btrfs_add_free_space(block_group, offset, num_bytes);
4416                         goto loop;
4417                 }
4418
4419                 /* move on to the next group */
4420                 if (search_start + num_bytes >
4421                     block_group->key.objectid + block_group->key.offset) {
4422                         btrfs_add_free_space(block_group, offset, num_bytes);
4423                         goto loop;
4424                 }
4425
4426                 if (exclude_nr > 0 &&
4427                     (search_start + num_bytes > exclude_start &&
4428                      search_start < exclude_start + exclude_nr)) {
4429                         search_start = exclude_start + exclude_nr;
4430
4431                         btrfs_add_free_space(block_group, offset, num_bytes);
4432                         /*
4433                          * if search_start is still in this block group
4434                          * then we just re-search this block group
4435                          */
4436                         if (search_start >= block_group->key.objectid &&
4437                             search_start < (block_group->key.objectid +
4438                                             block_group->key.offset))
4439                                 goto have_block_group;
4440                         goto loop;
4441                 }
4442
4443                 ins->objectid = search_start;
4444                 ins->offset = num_bytes;
4445
4446                 if (offset < search_start)
4447                         btrfs_add_free_space(block_group, offset,
4448                                              search_start - offset);
4449                 BUG_ON(offset > search_start);
4450
4451                 update_reserved_extents(block_group, num_bytes, 1);
4452
4453                 /* we are all good, lets return */
4454                 break;
4455 loop:
4456                 failed_cluster_refill = false;
4457                 failed_alloc = false;
4458                 btrfs_put_block_group(block_group);
4459         }
4460         up_read(&space_info->groups_sem);
4461
4462         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
4463          *                      for them to make caching progress.  Also
4464          *                      determine the best possible bg to cache
4465          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4466          *                      caching kthreads as we move along
4467          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4468          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4469          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4470          *                      again
4471          */
4472         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4473             (found_uncached_bg || empty_size || empty_cluster ||
4474              allowed_chunk_alloc)) {
4475                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
4476                         found_uncached_bg = false;
4477                         loop++;
4478                         if (!ideal_cache_percent &&
4479                             atomic_read(&space_info->caching_threads))
4480                                 goto search;
4481
4482                         /*
4483                          * 1 of the following 2 things have happened so far
4484                          *
4485                          * 1) We found an ideal block group for caching that
4486                          * is mostly full and will cache quickly, so we might
4487                          * as well wait for it.
4488                          *
4489                          * 2) We searched for cached only and we didn't find
4490                          * anything, and we didn't start any caching kthreads
4491                          * either, so chances are we will loop through and
4492                          * start a couple caching kthreads, and then come back
4493                          * around and just wait for them.  This will be slower
4494                          * because we will have 2 caching kthreads reading at
4495                          * the same time when we could have just started one
4496                          * and waited for it to get far enough to give us an
4497                          * allocation, so go ahead and go to the wait caching
4498                          * loop.
4499                          */
4500                         loop = LOOP_CACHING_WAIT;
4501                         search_start = ideal_cache_offset;
4502                         ideal_cache_percent = 0;
4503                         goto ideal_cache;
4504                 } else if (loop == LOOP_FIND_IDEAL) {
4505                         /*
4506                          * Didn't find a uncached bg, wait on anything we find
4507                          * next.
4508                          */
4509                         loop = LOOP_CACHING_WAIT;
4510                         goto search;
4511                 }
4512
4513                 if (loop < LOOP_CACHING_WAIT) {
4514                         loop++;
4515                         goto search;
4516                 }
4517
4518                 if (loop == LOOP_ALLOC_CHUNK) {
4519                         empty_size = 0;
4520                         empty_cluster = 0;
4521                 }
4522
4523                 if (allowed_chunk_alloc) {
4524                         ret = do_chunk_alloc(trans, root, num_bytes +
4525                                              2 * 1024 * 1024, data, 1);
4526                         allowed_chunk_alloc = 0;
4527                         done_chunk_alloc = 1;
4528                 } else if (!done_chunk_alloc) {
4529                         space_info->force_alloc = 1;
4530                 }
4531
4532                 if (loop < LOOP_NO_EMPTY_SIZE) {
4533                         loop++;
4534                         goto search;
4535                 }
4536                 ret = -ENOSPC;
4537         } else if (!ins->objectid) {
4538                 ret = -ENOSPC;
4539         }
4540
4541         /* we found what we needed */
4542         if (ins->objectid) {
4543                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4544                         trans->block_group = block_group->key.objectid;
4545
4546                 btrfs_put_block_group(block_group);
4547                 ret = 0;
4548         }
4549
4550         return ret;
4551 }
4552
4553 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4554                             int dump_block_groups)
4555 {
4556         struct btrfs_block_group_cache *cache;
4557
4558         spin_lock(&info->lock);
4559         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4560                (unsigned long long)(info->total_bytes - info->bytes_used -
4561                                     info->bytes_pinned - info->bytes_reserved -
4562                                     info->bytes_super),
4563                (info->full) ? "" : "not ");
4564         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4565                " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4566                "\n",
4567                (unsigned long long)info->total_bytes,
4568                (unsigned long long)info->bytes_pinned,
4569                (unsigned long long)info->bytes_delalloc,
4570                (unsigned long long)info->bytes_may_use,
4571                (unsigned long long)info->bytes_used,
4572                (unsigned long long)info->bytes_root,
4573                (unsigned long long)info->bytes_super,
4574                (unsigned long long)info->bytes_reserved);
4575         spin_unlock(&info->lock);
4576
4577         if (!dump_block_groups)
4578                 return;
4579
4580         down_read(&info->groups_sem);
4581         list_for_each_entry(cache, &info->block_groups, list) {
4582                 spin_lock(&cache->lock);
4583                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4584                        "%llu pinned %llu reserved\n",
4585                        (unsigned long long)cache->key.objectid,
4586                        (unsigned long long)cache->key.offset,
4587                        (unsigned long long)btrfs_block_group_used(&cache->item),
4588                        (unsigned long long)cache->pinned,
4589                        (unsigned long long)cache->reserved);
4590                 btrfs_dump_free_space(cache, bytes);
4591                 spin_unlock(&cache->lock);
4592         }
4593         up_read(&info->groups_sem);
4594 }
4595
4596 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4597                          struct btrfs_root *root,
4598                          u64 num_bytes, u64 min_alloc_size,
4599                          u64 empty_size, u64 hint_byte,
4600                          u64 search_end, struct btrfs_key *ins,
4601                          u64 data)
4602 {
4603         int ret;
4604         u64 search_start = 0;
4605
4606         data = btrfs_get_alloc_profile(root, data);
4607 again:
4608         /*
4609          * the only place that sets empty_size is btrfs_realloc_node, which
4610          * is not called recursively on allocations
4611          */
4612         if (empty_size || root->ref_cows)
4613                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4614                                      num_bytes + 2 * 1024 * 1024, data, 0);
4615
4616         WARN_ON(num_bytes < root->sectorsize);
4617         ret = find_free_extent(trans, root, num_bytes, empty_size,
4618                                search_start, search_end, hint_byte, ins,
4619                                trans->alloc_exclude_start,
4620                                trans->alloc_exclude_nr, data);
4621
4622         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4623                 num_bytes = num_bytes >> 1;
4624                 num_bytes = num_bytes & ~(root->sectorsize - 1);
4625                 num_bytes = max(num_bytes, min_alloc_size);
4626                 do_chunk_alloc(trans, root->fs_info->extent_root,
4627                                num_bytes, data, 1);
4628                 goto again;
4629         }
4630         if (ret == -ENOSPC) {
4631                 struct btrfs_space_info *sinfo;
4632
4633                 sinfo = __find_space_info(root->fs_info, data);
4634                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4635                        "wanted %llu\n", (unsigned long long)data,
4636                        (unsigned long long)num_bytes);
4637                 dump_space_info(sinfo, num_bytes, 1);
4638         }
4639
4640         return ret;
4641 }
4642
4643 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4644 {
4645         struct btrfs_block_group_cache *cache;
4646         int ret = 0;
4647
4648         cache = btrfs_lookup_block_group(root->fs_info, start);
4649         if (!cache) {
4650                 printk(KERN_ERR "Unable to find block group for %llu\n",
4651                        (unsigned long long)start);
4652                 return -ENOSPC;
4653         }
4654
4655         ret = btrfs_discard_extent(root, start, len);
4656
4657         btrfs_add_free_space(cache, start, len);
4658         update_reserved_extents(cache, len, 0);
4659         btrfs_put_block_group(cache);
4660
4661         return ret;
4662 }
4663
4664 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4665                                       struct btrfs_root *root,
4666                                       u64 parent, u64 root_objectid,
4667                                       u64 flags, u64 owner, u64 offset,
4668                                       struct btrfs_key *ins, int ref_mod)
4669 {
4670         int ret;
4671         struct btrfs_fs_info *fs_info = root->fs_info;
4672         struct btrfs_extent_item *extent_item;
4673         struct btrfs_extent_inline_ref *iref;
4674         struct btrfs_path *path;
4675         struct extent_buffer *leaf;
4676         int type;
4677         u32 size;
4678
4679         if (parent > 0)
4680                 type = BTRFS_SHARED_DATA_REF_KEY;
4681         else
4682                 type = BTRFS_EXTENT_DATA_REF_KEY;
4683
4684         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4685
4686         path = btrfs_alloc_path();
4687         BUG_ON(!path);
4688
4689         path->leave_spinning = 1;
4690         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4691                                       ins, size);
4692         BUG_ON(ret);
4693
4694         leaf = path->nodes[0];
4695         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4696                                      struct btrfs_extent_item);
4697         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4698         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4699         btrfs_set_extent_flags(leaf, extent_item,
4700                                flags | BTRFS_EXTENT_FLAG_DATA);
4701
4702         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4703         btrfs_set_extent_inline_ref_type(leaf, iref, type);
4704         if (parent > 0) {
4705                 struct btrfs_shared_data_ref *ref;
4706                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4707                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4708                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4709         } else {
4710                 struct btrfs_extent_data_ref *ref;
4711                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4712                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4713                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4714                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4715                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4716         }
4717
4718         btrfs_mark_buffer_dirty(path->nodes[0]);
4719         btrfs_free_path(path);
4720
4721         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4722                                  1, 0);
4723         if (ret) {
4724                 printk(KERN_ERR "btrfs update block group failed for %llu "
4725                        "%llu\n", (unsigned long long)ins->objectid,
4726                        (unsigned long long)ins->offset);
4727                 BUG();
4728         }
4729         return ret;
4730 }
4731
4732 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4733                                      struct btrfs_root *root,
4734                                      u64 parent, u64 root_objectid,
4735                                      u64 flags, struct btrfs_disk_key *key,
4736                                      int level, struct btrfs_key *ins)
4737 {
4738         int ret;
4739         struct btrfs_fs_info *fs_info = root->fs_info;
4740         struct btrfs_extent_item *extent_item;
4741         struct btrfs_tree_block_info *block_info;
4742         struct btrfs_extent_inline_ref *iref;
4743         struct btrfs_path *path;
4744         struct extent_buffer *leaf;
4745         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4746
4747         path = btrfs_alloc_path();
4748         BUG_ON(!path);
4749
4750         path->leave_spinning = 1;
4751         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4752                                       ins, size);
4753         BUG_ON(ret);
4754
4755         leaf = path->nodes[0];
4756         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4757                                      struct btrfs_extent_item);
4758         btrfs_set_extent_refs(leaf, extent_item, 1);
4759         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4760         btrfs_set_extent_flags(leaf, extent_item,
4761                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4762         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4763
4764         btrfs_set_tree_block_key(leaf, block_info, key);
4765         btrfs_set_tree_block_level(leaf, block_info, level);
4766
4767         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4768         if (parent > 0) {
4769                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4770                 btrfs_set_extent_inline_ref_type(leaf, iref,
4771                                                  BTRFS_SHARED_BLOCK_REF_KEY);
4772                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4773         } else {
4774                 btrfs_set_extent_inline_ref_type(leaf, iref,
4775                                                  BTRFS_TREE_BLOCK_REF_KEY);
4776                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4777         }
4778
4779         btrfs_mark_buffer_dirty(leaf);
4780         btrfs_free_path(path);
4781
4782         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4783                                  1, 0);
4784         if (ret) {
4785                 printk(KERN_ERR "btrfs update block group failed for %llu "
4786                        "%llu\n", (unsigned long long)ins->objectid,
4787                        (unsigned long long)ins->offset);
4788                 BUG();
4789         }
4790         return ret;
4791 }
4792
4793 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4794                                      struct btrfs_root *root,
4795                                      u64 root_objectid, u64 owner,
4796                                      u64 offset, struct btrfs_key *ins)
4797 {
4798         int ret;
4799
4800         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4801
4802         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4803                                          0, root_objectid, owner, offset,
4804                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
4805         return ret;
4806 }
4807
4808 /*
4809  * this is used by the tree logging recovery code.  It records that
4810  * an extent has been allocated and makes sure to clear the free
4811  * space cache bits as well
4812  */
4813 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4814                                    struct btrfs_root *root,
4815                                    u64 root_objectid, u64 owner, u64 offset,
4816                                    struct btrfs_key *ins)
4817 {
4818         int ret;
4819         struct btrfs_block_group_cache *block_group;
4820         struct btrfs_caching_control *caching_ctl;
4821         u64 start = ins->objectid;
4822         u64 num_bytes = ins->offset;
4823
4824         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4825         cache_block_group(block_group);
4826         caching_ctl = get_caching_control(block_group);
4827
4828         if (!caching_ctl) {
4829                 BUG_ON(!block_group_cache_done(block_group));
4830                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4831                 BUG_ON(ret);
4832         } else {
4833                 mutex_lock(&caching_ctl->mutex);
4834
4835                 if (start >= caching_ctl->progress) {
4836                         ret = add_excluded_extent(root, start, num_bytes);
4837                         BUG_ON(ret);
4838                 } else if (start + num_bytes <= caching_ctl->progress) {
4839                         ret = btrfs_remove_free_space(block_group,
4840                                                       start, num_bytes);
4841                         BUG_ON(ret);
4842                 } else {
4843                         num_bytes = caching_ctl->progress - start;
4844                         ret = btrfs_remove_free_space(block_group,
4845                                                       start, num_bytes);
4846                         BUG_ON(ret);
4847
4848                         start = caching_ctl->progress;
4849                         num_bytes = ins->objectid + ins->offset -
4850                                     caching_ctl->progress;
4851                         ret = add_excluded_extent(root, start, num_bytes);
4852                         BUG_ON(ret);
4853                 }
4854
4855                 mutex_unlock(&caching_ctl->mutex);
4856                 put_caching_control(caching_ctl);
4857         }
4858
4859         update_reserved_extents(block_group, ins->offset, 1);
4860         btrfs_put_block_group(block_group);
4861         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4862                                          0, owner, offset, ins, 1);
4863         return ret;
4864 }
4865
4866 /*
4867  * finds a free extent and does all the dirty work required for allocation
4868  * returns the key for the extent through ins, and a tree buffer for
4869  * the first block of the extent through buf.
4870  *
4871  * returns 0 if everything worked, non-zero otherwise.
4872  */
4873 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4874                             struct btrfs_root *root,
4875                             u64 num_bytes, u64 parent, u64 root_objectid,
4876                             struct btrfs_disk_key *key, int level,
4877                             u64 empty_size, u64 hint_byte, u64 search_end,
4878                             struct btrfs_key *ins)
4879 {
4880         int ret;
4881         u64 flags = 0;
4882
4883         ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4884                                    empty_size, hint_byte, search_end,
4885                                    ins, 0);
4886         if (ret)
4887                 return ret;
4888
4889         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4890                 if (parent == 0)
4891                         parent = ins->objectid;
4892                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4893         } else
4894                 BUG_ON(parent > 0);
4895
4896         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4897                 struct btrfs_delayed_extent_op *extent_op;
4898                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4899                 BUG_ON(!extent_op);
4900                 if (key)
4901                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
4902                 else
4903                         memset(&extent_op->key, 0, sizeof(extent_op->key));
4904                 extent_op->flags_to_set = flags;
4905                 extent_op->update_key = 1;
4906                 extent_op->update_flags = 1;
4907                 extent_op->is_data = 0;
4908
4909                 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4910                                         ins->offset, parent, root_objectid,
4911                                         level, BTRFS_ADD_DELAYED_EXTENT,
4912                                         extent_op);
4913                 BUG_ON(ret);
4914         }
4915
4916         if (root_objectid == root->root_key.objectid) {
4917                 u64 used;
4918                 spin_lock(&root->node_lock);
4919                 used = btrfs_root_used(&root->root_item) + num_bytes;
4920                 btrfs_set_root_used(&root->root_item, used);
4921                 spin_unlock(&root->node_lock);
4922         }
4923         return ret;
4924 }
4925
4926 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4927                                             struct btrfs_root *root,
4928                                             u64 bytenr, u32 blocksize,
4929                                             int level)
4930 {
4931         struct extent_buffer *buf;
4932
4933         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4934         if (!buf)
4935                 return ERR_PTR(-ENOMEM);
4936         btrfs_set_header_generation(buf, trans->transid);
4937         btrfs_set_buffer_lockdep_class(buf, level);
4938         btrfs_tree_lock(buf);
4939         clean_tree_block(trans, root, buf);
4940
4941         btrfs_set_lock_blocking(buf);
4942         btrfs_set_buffer_uptodate(buf);
4943
4944         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4945                 /*
4946                  * we allow two log transactions at a time, use different
4947                  * EXENT bit to differentiate dirty pages.
4948                  */
4949                 if (root->log_transid % 2 == 0)
4950                         set_extent_dirty(&root->dirty_log_pages, buf->start,
4951                                         buf->start + buf->len - 1, GFP_NOFS);
4952                 else
4953                         set_extent_new(&root->dirty_log_pages, buf->start,
4954                                         buf->start + buf->len - 1, GFP_NOFS);
4955         } else {
4956                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4957                          buf->start + buf->len - 1, GFP_NOFS);
4958         }
4959         trans->blocks_used++;
4960         /* this returns a buffer locked for blocking */
4961         return buf;
4962 }
4963
4964 /*
4965  * helper function to allocate a block for a given tree
4966  * returns the tree buffer or NULL.
4967  */
4968 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4969                                         struct btrfs_root *root, u32 blocksize,
4970                                         u64 parent, u64 root_objectid,
4971                                         struct btrfs_disk_key *key, int level,
4972                                         u64 hint, u64 empty_size)
4973 {
4974         struct btrfs_key ins;
4975         int ret;
4976         struct extent_buffer *buf;
4977
4978         ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4979                                key, level, empty_size, hint, (u64)-1, &ins);
4980         if (ret) {
4981                 BUG_ON(ret > 0);
4982                 return ERR_PTR(ret);
4983         }
4984
4985         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4986                                     blocksize, level);
4987         return buf;
4988 }
4989
4990 struct walk_control {
4991         u64 refs[BTRFS_MAX_LEVEL];
4992         u64 flags[BTRFS_MAX_LEVEL];
4993         struct btrfs_key update_progress;
4994         int stage;
4995         int level;
4996         int shared_level;
4997         int update_ref;
4998         int keep_locks;
4999         int reada_slot;
5000         int reada_count;
5001 };
5002
5003 #define DROP_REFERENCE  1
5004 #define UPDATE_BACKREF  2
5005
5006 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5007                                      struct btrfs_root *root,
5008                                      struct walk_control *wc,
5009                                      struct btrfs_path *path)
5010 {
5011         u64 bytenr;
5012         u64 generation;
5013         u64 refs;
5014         u64 flags;
5015         u64 last = 0;
5016         u32 nritems;
5017         u32 blocksize;
5018         struct btrfs_key key;
5019         struct extent_buffer *eb;
5020         int ret;
5021         int slot;
5022         int nread = 0;
5023
5024         if (path->slots[wc->level] < wc->reada_slot) {
5025                 wc->reada_count = wc->reada_count * 2 / 3;
5026                 wc->reada_count = max(wc->reada_count, 2);
5027         } else {
5028                 wc->reada_count = wc->reada_count * 3 / 2;
5029                 wc->reada_count = min_t(int, wc->reada_count,
5030                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5031         }
5032
5033         eb = path->nodes[wc->level];
5034         nritems = btrfs_header_nritems(eb);
5035         blocksize = btrfs_level_size(root, wc->level - 1);
5036
5037         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5038                 if (nread >= wc->reada_count)
5039                         break;
5040
5041                 cond_resched();
5042                 bytenr = btrfs_node_blockptr(eb, slot);
5043                 generation = btrfs_node_ptr_generation(eb, slot);
5044
5045                 if (slot == path->slots[wc->level])
5046                         goto reada;
5047
5048                 if (wc->stage == UPDATE_BACKREF &&
5049                     generation <= root->root_key.offset)
5050                         continue;
5051
5052                 /* We don't lock the tree block, it's OK to be racy here */
5053                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5054                                                &refs, &flags);
5055                 BUG_ON(ret);
5056                 BUG_ON(refs == 0);
5057
5058                 if (wc->stage == DROP_REFERENCE) {
5059                         if (refs == 1)
5060                                 goto reada;
5061
5062                         if (wc->level == 1 &&
5063                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5064                                 continue;
5065                         if (!wc->update_ref ||
5066                             generation <= root->root_key.offset)
5067                                 continue;
5068                         btrfs_node_key_to_cpu(eb, &key, slot);
5069                         ret = btrfs_comp_cpu_keys(&key,
5070                                                   &wc->update_progress);
5071                         if (ret < 0)
5072                                 continue;
5073                 } else {
5074                         if (wc->level == 1 &&
5075                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5076                                 continue;
5077                 }
5078 reada:
5079                 ret = readahead_tree_block(root, bytenr, blocksize,
5080                                            generation);
5081                 if (ret)
5082                         break;
5083                 last = bytenr + blocksize;
5084                 nread++;
5085         }
5086         wc->reada_slot = slot;
5087 }
5088
5089 /*
5090  * hepler to process tree block while walking down the tree.
5091  *
5092  * when wc->stage == UPDATE_BACKREF, this function updates
5093  * back refs for pointers in the block.
5094  *
5095  * NOTE: return value 1 means we should stop walking down.
5096  */
5097 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5098                                    struct btrfs_root *root,
5099                                    struct btrfs_path *path,
5100                                    struct walk_control *wc, int lookup_info)
5101 {
5102         int level = wc->level;
5103         struct extent_buffer *eb = path->nodes[level];
5104         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5105         int ret;
5106
5107         if (wc->stage == UPDATE_BACKREF &&
5108             btrfs_header_owner(eb) != root->root_key.objectid)
5109                 return 1;
5110
5111         /*
5112          * when reference count of tree block is 1, it won't increase
5113          * again. once full backref flag is set, we never clear it.
5114          */
5115         if (lookup_info &&
5116             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5117              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5118                 BUG_ON(!path->locks[level]);
5119                 ret = btrfs_lookup_extent_info(trans, root,
5120                                                eb->start, eb->len,
5121                                                &wc->refs[level],
5122                                                &wc->flags[level]);
5123                 BUG_ON(ret);
5124                 BUG_ON(wc->refs[level] == 0);
5125         }
5126
5127         if (wc->stage == DROP_REFERENCE) {
5128                 if (wc->refs[level] > 1)
5129                         return 1;
5130
5131                 if (path->locks[level] && !wc->keep_locks) {
5132                         btrfs_tree_unlock(eb);
5133                         path->locks[level] = 0;
5134                 }
5135                 return 0;
5136         }
5137
5138         /* wc->stage == UPDATE_BACKREF */
5139         if (!(wc->flags[level] & flag)) {
5140                 BUG_ON(!path->locks[level]);
5141                 ret = btrfs_inc_ref(trans, root, eb, 1);
5142                 BUG_ON(ret);
5143                 ret = btrfs_dec_ref(trans, root, eb, 0);
5144                 BUG_ON(ret);
5145                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5146                                                   eb->len, flag, 0);
5147                 BUG_ON(ret);
5148                 wc->flags[level] |= flag;
5149         }
5150
5151         /*
5152          * the block is shared by multiple trees, so it's not good to
5153          * keep the tree lock
5154          */
5155         if (path->locks[level] && level > 0) {
5156                 btrfs_tree_unlock(eb);
5157                 path->locks[level] = 0;
5158         }
5159         return 0;
5160 }
5161
5162 /*
5163  * hepler to process tree block pointer.
5164  *
5165  * when wc->stage == DROP_REFERENCE, this function checks
5166  * reference count of the block pointed to. if the block
5167  * is shared and we need update back refs for the subtree
5168  * rooted at the block, this function changes wc->stage to
5169  * UPDATE_BACKREF. if the block is shared and there is no
5170  * need to update back, this function drops the reference
5171  * to the block.
5172  *
5173  * NOTE: return value 1 means we should stop walking down.
5174  */
5175 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5176                                  struct btrfs_root *root,
5177                                  struct btrfs_path *path,
5178                                  struct walk_control *wc, int *lookup_info)
5179 {
5180         u64 bytenr;
5181         u64 generation;
5182         u64 parent;
5183         u32 blocksize;
5184         struct btrfs_key key;
5185         struct extent_buffer *next;
5186         int level = wc->level;
5187         int reada = 0;
5188         int ret = 0;
5189
5190         generation = btrfs_node_ptr_generation(path->nodes[level],
5191                                                path->slots[level]);
5192         /*
5193          * if the lower level block was created before the snapshot
5194          * was created, we know there is no need to update back refs
5195          * for the subtree
5196          */
5197         if (wc->stage == UPDATE_BACKREF &&
5198             generation <= root->root_key.offset) {
5199                 *lookup_info = 1;
5200                 return 1;
5201         }
5202
5203         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5204         blocksize = btrfs_level_size(root, level - 1);
5205
5206         next = btrfs_find_tree_block(root, bytenr, blocksize);
5207         if (!next) {
5208                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5209                 reada = 1;
5210         }
5211         btrfs_tree_lock(next);
5212         btrfs_set_lock_blocking(next);
5213
5214         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5215                                        &wc->refs[level - 1],
5216                                        &wc->flags[level - 1]);
5217         BUG_ON(ret);
5218         BUG_ON(wc->refs[level - 1] == 0);
5219         *lookup_info = 0;
5220
5221         if (wc->stage == DROP_REFERENCE) {
5222                 if (wc->refs[level - 1] > 1) {
5223                         if (level == 1 &&
5224                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5225                                 goto skip;
5226
5227                         if (!wc->update_ref ||
5228                             generation <= root->root_key.offset)
5229                                 goto skip;
5230
5231                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5232                                               path->slots[level]);
5233                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5234                         if (ret < 0)
5235                                 goto skip;
5236
5237                         wc->stage = UPDATE_BACKREF;
5238                         wc->shared_level = level - 1;
5239                 }
5240         } else {
5241                 if (level == 1 &&
5242                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5243                         goto skip;
5244         }
5245
5246         if (!btrfs_buffer_uptodate(next, generation)) {
5247                 btrfs_tree_unlock(next);
5248                 free_extent_buffer(next);
5249                 next = NULL;
5250                 *lookup_info = 1;
5251         }
5252
5253         if (!next) {
5254                 if (reada && level == 1)
5255                         reada_walk_down(trans, root, wc, path);
5256                 next = read_tree_block(root, bytenr, blocksize, generation);
5257                 btrfs_tree_lock(next);
5258                 btrfs_set_lock_blocking(next);
5259         }
5260
5261         level--;
5262         BUG_ON(level != btrfs_header_level(next));
5263         path->nodes[level] = next;
5264         path->slots[level] = 0;
5265         path->locks[level] = 1;
5266         wc->level = level;
5267         if (wc->level == 1)
5268                 wc->reada_slot = 0;
5269         return 0;
5270 skip:
5271         wc->refs[level - 1] = 0;
5272         wc->flags[level - 1] = 0;
5273         if (wc->stage == DROP_REFERENCE) {
5274                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5275                         parent = path->nodes[level]->start;
5276                 } else {
5277                         BUG_ON(root->root_key.objectid !=
5278                                btrfs_header_owner(path->nodes[level]));
5279                         parent = 0;
5280                 }
5281
5282                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5283                                         root->root_key.objectid, level - 1, 0);
5284                 BUG_ON(ret);
5285         }
5286         btrfs_tree_unlock(next);
5287         free_extent_buffer(next);
5288         *lookup_info = 1;
5289         return 1;
5290 }
5291
5292 /*
5293  * hepler to process tree block while walking up the tree.
5294  *
5295  * when wc->stage == DROP_REFERENCE, this function drops
5296  * reference count on the block.
5297  *
5298  * when wc->stage == UPDATE_BACKREF, this function changes
5299  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5300  * to UPDATE_BACKREF previously while processing the block.
5301  *
5302  * NOTE: return value 1 means we should stop walking up.
5303  */
5304 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5305                                  struct btrfs_root *root,
5306                                  struct btrfs_path *path,
5307                                  struct walk_control *wc)
5308 {
5309         int ret = 0;
5310         int level = wc->level;
5311         struct extent_buffer *eb = path->nodes[level];
5312         u64 parent = 0;
5313
5314         if (wc->stage == UPDATE_BACKREF) {
5315                 BUG_ON(wc->shared_level < level);
5316                 if (level < wc->shared_level)
5317                         goto out;
5318
5319                 ret = find_next_key(path, level + 1, &wc->update_progress);
5320                 if (ret > 0)
5321                         wc->update_ref = 0;
5322
5323                 wc->stage = DROP_REFERENCE;
5324                 wc->shared_level = -1;
5325                 path->slots[level] = 0;
5326
5327                 /*
5328                  * check reference count again if the block isn't locked.
5329                  * we should start walking down the tree again if reference
5330                  * count is one.
5331                  */
5332                 if (!path->locks[level]) {
5333                         BUG_ON(level == 0);
5334                         btrfs_tree_lock(eb);
5335                         btrfs_set_lock_blocking(eb);
5336                         path->locks[level] = 1;
5337
5338                         ret = btrfs_lookup_extent_info(trans, root,
5339                                                        eb->start, eb->len,
5340                                                        &wc->refs[level],
5341                                                        &wc->flags[level]);
5342                         BUG_ON(ret);
5343                         BUG_ON(wc->refs[level] == 0);
5344                         if (wc->refs[level] == 1) {
5345                                 btrfs_tree_unlock(eb);
5346                                 path->locks[level] = 0;
5347                                 return 1;
5348                         }
5349                 }
5350         }
5351
5352         /* wc->stage == DROP_REFERENCE */
5353         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5354
5355         if (wc->refs[level] == 1) {
5356                 if (level == 0) {
5357                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5358                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5359                         else
5360                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5361                         BUG_ON(ret);
5362                 }
5363                 /* make block locked assertion in clean_tree_block happy */
5364                 if (!path->locks[level] &&
5365                     btrfs_header_generation(eb) == trans->transid) {
5366                         btrfs_tree_lock(eb);
5367                         btrfs_set_lock_blocking(eb);
5368                         path->locks[level] = 1;
5369                 }
5370                 clean_tree_block(trans, root, eb);
5371         }
5372
5373         if (eb == root->node) {
5374                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5375                         parent = eb->start;
5376                 else
5377                         BUG_ON(root->root_key.objectid !=
5378                                btrfs_header_owner(eb));
5379         } else {
5380                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5381                         parent = path->nodes[level + 1]->start;
5382                 else
5383                         BUG_ON(root->root_key.objectid !=
5384                                btrfs_header_owner(path->nodes[level + 1]));
5385         }
5386
5387         ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
5388                                 root->root_key.objectid, level, 0);
5389         BUG_ON(ret);
5390 out:
5391         wc->refs[level] = 0;
5392         wc->flags[level] = 0;
5393         return ret;
5394 }
5395
5396 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5397                                    struct btrfs_root *root,
5398                                    struct btrfs_path *path,
5399                                    struct walk_control *wc)
5400 {
5401         int level = wc->level;
5402         int lookup_info = 1;
5403         int ret;
5404
5405         while (level >= 0) {
5406                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5407                 if (ret > 0)
5408                         break;
5409
5410                 if (level == 0)
5411                         break;
5412
5413                 if (path->slots[level] >=
5414                     btrfs_header_nritems(path->nodes[level]))
5415                         break;
5416
5417                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5418                 if (ret > 0) {
5419                         path->slots[level]++;
5420                         continue;
5421                 }
5422                 level = wc->level;
5423         }
5424         return 0;
5425 }
5426
5427 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5428                                  struct btrfs_root *root,
5429                                  struct btrfs_path *path,
5430                                  struct walk_control *wc, int max_level)
5431 {
5432         int level = wc->level;
5433         int ret;
5434
5435         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5436         while (level < max_level && path->nodes[level]) {
5437                 wc->level = level;
5438                 if (path->slots[level] + 1 <
5439                     btrfs_header_nritems(path->nodes[level])) {
5440                         path->slots[level]++;
5441                         return 0;
5442                 } else {
5443                         ret = walk_up_proc(trans, root, path, wc);
5444                         if (ret > 0)
5445                                 return 0;
5446
5447                         if (path->locks[level]) {
5448                                 btrfs_tree_unlock(path->nodes[level]);
5449                                 path->locks[level] = 0;
5450                         }
5451                         free_extent_buffer(path->nodes[level]);
5452                         path->nodes[level] = NULL;
5453                         level++;
5454                 }
5455         }
5456         return 1;
5457 }
5458
5459 /*
5460  * drop a subvolume tree.
5461  *
5462  * this function traverses the tree freeing any blocks that only
5463  * referenced by the tree.
5464  *
5465  * when a shared tree block is found. this function decreases its
5466  * reference count by one. if update_ref is true, this function
5467  * also make sure backrefs for the shared block and all lower level
5468  * blocks are properly updated.
5469  */
5470 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5471 {
5472         struct btrfs_path *path;
5473         struct btrfs_trans_handle *trans;
5474         struct btrfs_root *tree_root = root->fs_info->tree_root;
5475         struct btrfs_root_item *root_item = &root->root_item;
5476         struct walk_control *wc;
5477         struct btrfs_key key;
5478         int err = 0;
5479         int ret;
5480         int level;
5481
5482         path = btrfs_alloc_path();
5483         BUG_ON(!path);
5484
5485         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5486         BUG_ON(!wc);
5487
5488         trans = btrfs_start_transaction(tree_root, 1);
5489
5490         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5491                 level = btrfs_header_level(root->node);
5492                 path->nodes[level] = btrfs_lock_root_node(root);
5493                 btrfs_set_lock_blocking(path->nodes[level]);
5494                 path->slots[level] = 0;
5495                 path->locks[level] = 1;
5496                 memset(&wc->update_progress, 0,
5497                        sizeof(wc->update_progress));
5498         } else {
5499                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5500                 memcpy(&wc->update_progress, &key,
5501                        sizeof(wc->update_progress));
5502
5503                 level = root_item->drop_level;
5504                 BUG_ON(level == 0);
5505                 path->lowest_level = level;
5506                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5507                 path->lowest_level = 0;
5508                 if (ret < 0) {
5509                         err = ret;
5510                         goto out;
5511                 }
5512                 WARN_ON(ret > 0);
5513
5514                 /*
5515                  * unlock our path, this is safe because only this
5516                  * function is allowed to delete this snapshot
5517                  */
5518                 btrfs_unlock_up_safe(path, 0);
5519
5520                 level = btrfs_header_level(root->node);
5521                 while (1) {
5522                         btrfs_tree_lock(path->nodes[level]);
5523                         btrfs_set_lock_blocking(path->nodes[level]);
5524
5525                         ret = btrfs_lookup_extent_info(trans, root,
5526                                                 path->nodes[level]->start,
5527                                                 path->nodes[level]->len,
5528                                                 &wc->refs[level],
5529                                                 &wc->flags[level]);
5530                         BUG_ON(ret);
5531                         BUG_ON(wc->refs[level] == 0);
5532
5533                         if (level == root_item->drop_level)
5534                                 break;
5535
5536                         btrfs_tree_unlock(path->nodes[level]);
5537                         WARN_ON(wc->refs[level] != 1);
5538                         level--;
5539                 }
5540         }
5541
5542         wc->level = level;
5543         wc->shared_level = -1;
5544         wc->stage = DROP_REFERENCE;
5545         wc->update_ref = update_ref;
5546         wc->keep_locks = 0;
5547         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5548
5549         while (1) {
5550                 ret = walk_down_tree(trans, root, path, wc);
5551                 if (ret < 0) {
5552                         err = ret;
5553                         break;
5554                 }
5555
5556                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5557                 if (ret < 0) {
5558                         err = ret;
5559                         break;
5560                 }
5561
5562                 if (ret > 0) {
5563                         BUG_ON(wc->stage != DROP_REFERENCE);
5564                         break;
5565                 }
5566
5567                 if (wc->stage == DROP_REFERENCE) {
5568                         level = wc->level;
5569                         btrfs_node_key(path->nodes[level],
5570                                        &root_item->drop_progress,
5571                                        path->slots[level]);
5572                         root_item->drop_level = level;
5573                 }
5574
5575                 BUG_ON(wc->level == 0);
5576                 if (trans->transaction->in_commit ||
5577                     trans->transaction->delayed_refs.flushing) {
5578                         ret = btrfs_update_root(trans, tree_root,
5579                                                 &root->root_key,
5580                                                 root_item);
5581                         BUG_ON(ret);
5582
5583                         btrfs_end_transaction(trans, tree_root);
5584                         trans = btrfs_start_transaction(tree_root, 1);
5585                 } else {
5586                         unsigned long update;
5587                         update = trans->delayed_ref_updates;
5588                         trans->delayed_ref_updates = 0;
5589                         if (update)
5590                                 btrfs_run_delayed_refs(trans, tree_root,
5591                                                        update);
5592                 }
5593         }
5594         btrfs_release_path(root, path);
5595         BUG_ON(err);
5596
5597         ret = btrfs_del_root(trans, tree_root, &root->root_key);
5598         BUG_ON(ret);
5599
5600         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5601                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5602                                            NULL, NULL);
5603                 BUG_ON(ret < 0);
5604                 if (ret > 0) {
5605                         ret = btrfs_del_orphan_item(trans, tree_root,
5606                                                     root->root_key.objectid);
5607                         BUG_ON(ret);
5608                 }
5609         }
5610
5611         if (root->in_radix) {
5612                 btrfs_free_fs_root(tree_root->fs_info, root);
5613         } else {
5614                 free_extent_buffer(root->node);
5615                 free_extent_buffer(root->commit_root);
5616                 kfree(root);
5617         }
5618 out:
5619         btrfs_end_transaction(trans, tree_root);
5620         kfree(wc);
5621         btrfs_free_path(path);
5622         return err;
5623 }
5624
5625 /*
5626  * drop subtree rooted at tree block 'node'.
5627  *
5628  * NOTE: this function will unlock and release tree block 'node'
5629  */
5630 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5631                         struct btrfs_root *root,
5632                         struct extent_buffer *node,
5633                         struct extent_buffer *parent)
5634 {
5635         struct btrfs_path *path;
5636         struct walk_control *wc;
5637         int level;
5638         int parent_level;
5639         int ret = 0;
5640         int wret;
5641
5642         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5643
5644         path = btrfs_alloc_path();
5645         BUG_ON(!path);
5646
5647         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5648         BUG_ON(!wc);
5649
5650         btrfs_assert_tree_locked(parent);
5651         parent_level = btrfs_header_level(parent);
5652         extent_buffer_get(parent);
5653         path->nodes[parent_level] = parent;
5654         path->slots[parent_level] = btrfs_header_nritems(parent);
5655
5656         btrfs_assert_tree_locked(node);
5657         level = btrfs_header_level(node);
5658         path->nodes[level] = node;
5659         path->slots[level] = 0;
5660         path->locks[level] = 1;
5661
5662         wc->refs[parent_level] = 1;
5663         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5664         wc->level = level;
5665         wc->shared_level = -1;
5666         wc->stage = DROP_REFERENCE;
5667         wc->update_ref = 0;
5668         wc->keep_locks = 1;
5669         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5670
5671         while (1) {
5672                 wret = walk_down_tree(trans, root, path, wc);
5673                 if (wret < 0) {
5674                         ret = wret;
5675                         break;
5676                 }
5677
5678                 wret = walk_up_tree(trans, root, path, wc, parent_level);
5679                 if (wret < 0)
5680                         ret = wret;
5681                 if (wret != 0)
5682                         break;
5683         }
5684
5685         kfree(wc);
5686         btrfs_free_path(path);
5687         return ret;
5688 }
5689
5690 #if 0
5691 static unsigned long calc_ra(unsigned long start, unsigned long last,
5692                              unsigned long nr)
5693 {
5694         return min(last, start + nr - 1);
5695 }
5696
5697 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5698                                          u64 len)
5699 {
5700         u64 page_start;
5701         u64 page_end;
5702         unsigned long first_index;
5703         unsigned long last_index;
5704         unsigned long i;
5705         struct page *page;
5706         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5707         struct file_ra_state *ra;
5708         struct btrfs_ordered_extent *ordered;
5709         unsigned int total_read = 0;
5710         unsigned int total_dirty = 0;
5711         int ret = 0;
5712
5713         ra = kzalloc(sizeof(*ra), GFP_NOFS);
5714
5715         mutex_lock(&inode->i_mutex);
5716         first_index = start >> PAGE_CACHE_SHIFT;
5717         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5718
5719         /* make sure the dirty trick played by the caller work */
5720         ret = invalidate_inode_pages2_range(inode->i_mapping,
5721                                             first_index, last_index);
5722         if (ret)
5723                 goto out_unlock;
5724
5725         file_ra_state_init(ra, inode->i_mapping);
5726
5727         for (i = first_index ; i <= last_index; i++) {
5728                 if (total_read % ra->ra_pages == 0) {
5729                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5730                                        calc_ra(i, last_index, ra->ra_pages));
5731                 }
5732                 total_read++;
5733 again:
5734                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5735                         BUG_ON(1);
5736                 page = grab_cache_page(inode->i_mapping, i);
5737                 if (!page) {
5738                         ret = -ENOMEM;
5739                         goto out_unlock;
5740                 }
5741                 if (!PageUptodate(page)) {
5742                         btrfs_readpage(NULL, page);
5743                         lock_page(page);
5744                         if (!PageUptodate(page)) {
5745                                 unlock_page(page);
5746                                 page_cache_release(page);
5747                                 ret = -EIO;
5748                                 goto out_unlock;
5749                         }
5750                 }
5751                 wait_on_page_writeback(page);
5752
5753                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5754                 page_end = page_start + PAGE_CACHE_SIZE - 1;
5755                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5756
5757                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5758                 if (ordered) {
5759                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5760                         unlock_page(page);
5761                         page_cache_release(page);
5762                         btrfs_start_ordered_extent(inode, ordered, 1);
5763                         btrfs_put_ordered_extent(ordered);
5764                         goto again;
5765                 }
5766                 set_page_extent_mapped(page);
5767
5768                 if (i == first_index)
5769                         set_extent_bits(io_tree, page_start, page_end,
5770                                         EXTENT_BOUNDARY, GFP_NOFS);
5771                 btrfs_set_extent_delalloc(inode, page_start, page_end);
5772
5773                 set_page_dirty(page);
5774                 total_dirty++;
5775
5776                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5777                 unlock_page(page);
5778                 page_cache_release(page);
5779         }
5780
5781 out_unlock:
5782         kfree(ra);
5783         mutex_unlock(&inode->i_mutex);
5784         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5785         return ret;
5786 }
5787
5788 static noinline int relocate_data_extent(struct inode *reloc_inode,
5789                                          struct btrfs_key *extent_key,
5790                                          u64 offset)
5791 {
5792         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5793         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5794         struct extent_map *em;
5795         u64 start = extent_key->objectid - offset;
5796         u64 end = start + extent_key->offset - 1;
5797
5798         em = alloc_extent_map(GFP_NOFS);
5799         BUG_ON(!em || IS_ERR(em));
5800
5801         em->start = start;
5802         em->len = extent_key->offset;
5803         em->block_len = extent_key->offset;
5804         em->block_start = extent_key->objectid;
5805         em->bdev = root->fs_info->fs_devices->latest_bdev;
5806         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5807
5808         /* setup extent map to cheat btrfs_readpage */
5809         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5810         while (1) {
5811                 int ret;
5812                 write_lock(&em_tree->lock);
5813                 ret = add_extent_mapping(em_tree, em);
5814                 write_unlock(&em_tree->lock);
5815                 if (ret != -EEXIST) {
5816                         free_extent_map(em);
5817                         break;
5818                 }
5819                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5820         }
5821         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5822
5823         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5824 }
5825
5826 struct btrfs_ref_path {
5827         u64 extent_start;
5828         u64 nodes[BTRFS_MAX_LEVEL];
5829         u64 root_objectid;
5830         u64 root_generation;
5831         u64 owner_objectid;
5832         u32 num_refs;
5833         int lowest_level;
5834         int current_level;
5835         int shared_level;
5836
5837         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5838         u64 new_nodes[BTRFS_MAX_LEVEL];
5839 };
5840
5841 struct disk_extent {
5842         u64 ram_bytes;
5843         u64 disk_bytenr;
5844         u64 disk_num_bytes;
5845         u64 offset;
5846         u64 num_bytes;
5847         u8 compression;
5848         u8 encryption;
5849         u16 other_encoding;
5850 };
5851
5852 static int is_cowonly_root(u64 root_objectid)
5853 {
5854         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5855             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5856             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5857             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5858             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5859             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5860                 return 1;
5861         return 0;
5862 }
5863
5864 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5865                                     struct btrfs_root *extent_root,
5866                                     struct btrfs_ref_path *ref_path,
5867                                     int first_time)
5868 {
5869         struct extent_buffer *leaf;
5870         struct btrfs_path *path;
5871         struct btrfs_extent_ref *ref;
5872         struct btrfs_key key;
5873         struct btrfs_key found_key;
5874         u64 bytenr;
5875         u32 nritems;
5876         int level;
5877         int ret = 1;
5878
5879         path = btrfs_alloc_path();
5880         if (!path)
5881                 return -ENOMEM;
5882
5883         if (first_time) {
5884                 ref_path->lowest_level = -1;
5885                 ref_path->current_level = -1;
5886                 ref_path->shared_level = -1;
5887                 goto walk_up;
5888         }
5889 walk_down:
5890         level = ref_path->current_level - 1;
5891         while (level >= -1) {
5892                 u64 parent;
5893                 if (level < ref_path->lowest_level)
5894                         break;
5895
5896                 if (level >= 0)
5897                         bytenr = ref_path->nodes[level];
5898                 else
5899                         bytenr = ref_path->extent_start;
5900                 BUG_ON(bytenr == 0);
5901
5902                 parent = ref_path->nodes[level + 1];
5903                 ref_path->nodes[level + 1] = 0;
5904                 ref_path->current_level = level;
5905                 BUG_ON(parent == 0);
5906
5907                 key.objectid = bytenr;
5908                 key.offset = parent + 1;
5909                 key.type = BTRFS_EXTENT_REF_KEY;
5910
5911                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5912                 if (ret < 0)
5913                         goto out;
5914                 BUG_ON(ret == 0);
5915
5916                 leaf = path->nodes[0];
5917                 nritems = btrfs_header_nritems(leaf);
5918                 if (path->slots[0] >= nritems) {
5919                         ret = btrfs_next_leaf(extent_root, path);
5920                         if (ret < 0)
5921                                 goto out;
5922                         if (ret > 0)
5923                                 goto next;
5924                         leaf = path->nodes[0];
5925                 }
5926
5927                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5928                 if (found_key.objectid == bytenr &&
5929                     found_key.type == BTRFS_EXTENT_REF_KEY) {
5930                         if (level < ref_path->shared_level)
5931                                 ref_path->shared_level = level;
5932                         goto found;
5933                 }
5934 next:
5935                 level--;
5936                 btrfs_release_path(extent_root, path);
5937                 cond_resched();
5938         }
5939         /* reached lowest level */
5940         ret = 1;
5941         goto out;
5942 walk_up:
5943         level = ref_path->current_level;
5944         while (level < BTRFS_MAX_LEVEL - 1) {
5945                 u64 ref_objectid;
5946
5947                 if (level >= 0)
5948                         bytenr = ref_path->nodes[level];
5949                 else
5950                         bytenr = ref_path->extent_start;
5951
5952                 BUG_ON(bytenr == 0);
5953
5954                 key.objectid = bytenr;
5955                 key.offset = 0;
5956                 key.type = BTRFS_EXTENT_REF_KEY;
5957
5958                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5959                 if (ret < 0)
5960                         goto out;
5961
5962                 leaf = path->nodes[0];
5963                 nritems = btrfs_header_nritems(leaf);
5964                 if (path->slots[0] >= nritems) {
5965                         ret = btrfs_next_leaf(extent_root, path);
5966                         if (ret < 0)
5967                                 goto out;
5968                         if (ret > 0) {
5969                                 /* the extent was freed by someone */
5970                                 if (ref_path->lowest_level == level)
5971                                         goto out;
5972                                 btrfs_release_path(extent_root, path);
5973                                 goto walk_down;
5974                         }
5975                         leaf = path->nodes[0];
5976                 }
5977
5978                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5979                 if (found_key.objectid != bytenr ||
5980                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
5981                         /* the extent was freed by someone */
5982                         if (ref_path->lowest_level == level) {
5983                                 ret = 1;
5984                                 goto out;
5985                         }
5986                         btrfs_release_path(extent_root, path);
5987                         goto walk_down;
5988                 }
5989 found:
5990                 ref = btrfs_item_ptr(leaf, path->slots[0],
5991                                 struct btrfs_extent_ref);
5992                 ref_objectid = btrfs_ref_objectid(leaf, ref);
5993                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5994                         if (first_time) {
5995                                 level = (int)ref_objectid;
5996                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
5997                                 ref_path->lowest_level = level;
5998                                 ref_path->current_level = level;
5999                                 ref_path->nodes[level] = bytenr;
6000                         } else {
6001                                 WARN_ON(ref_objectid != level);
6002                         }
6003                 } else {
6004                         WARN_ON(level != -1);
6005                 }
6006                 first_time = 0;
6007
6008                 if (ref_path->lowest_level == level) {
6009                         ref_path->owner_objectid = ref_objectid;
6010                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6011                 }
6012
6013                 /*
6014                  * the block is tree root or the block isn't in reference
6015                  * counted tree.
6016                  */
6017                 if (found_key.objectid == found_key.offset ||
6018                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6019                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6020                         ref_path->root_generation =
6021                                 btrfs_ref_generation(leaf, ref);
6022                         if (level < 0) {
6023                                 /* special reference from the tree log */
6024                                 ref_path->nodes[0] = found_key.offset;
6025                                 ref_path->current_level = 0;
6026                         }
6027                         ret = 0;
6028                         goto out;
6029                 }
6030
6031                 level++;
6032                 BUG_ON(ref_path->nodes[level] != 0);
6033                 ref_path->nodes[level] = found_key.offset;
6034                 ref_path->current_level = level;
6035
6036                 /*
6037                  * the reference was created in the running transaction,
6038                  * no need to continue walking up.
6039                  */
6040                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6041                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6042                         ref_path->root_generation =
6043                                 btrfs_ref_generation(leaf, ref);
6044                         ret = 0;
6045                         goto out;
6046                 }
6047
6048                 btrfs_release_path(extent_root, path);
6049                 cond_resched();
6050         }
6051         /* reached max tree level, but no tree root found. */
6052         BUG();
6053 out:
6054         btrfs_free_path(path);
6055         return ret;
6056 }
6057
6058 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6059                                 struct btrfs_root *extent_root,
6060                                 struct btrfs_ref_path *ref_path,
6061                                 u64 extent_start)
6062 {
6063         memset(ref_path, 0, sizeof(*ref_path));
6064         ref_path->extent_start = extent_start;
6065
6066         return __next_ref_path(trans, extent_root, ref_path, 1);
6067 }
6068
6069 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6070                                struct btrfs_root *extent_root,
6071                                struct btrfs_ref_path *ref_path)
6072 {
6073         return __next_ref_path(trans, extent_root, ref_path, 0);
6074 }
6075
6076 static noinline int get_new_locations(struct inode *reloc_inode,
6077                                       struct btrfs_key *extent_key,
6078                                       u64 offset, int no_fragment,
6079                                       struct disk_extent **extents,
6080                                       int *nr_extents)
6081 {
6082         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6083         struct btrfs_path *path;
6084         struct btrfs_file_extent_item *fi;
6085         struct extent_buffer *leaf;
6086         struct disk_extent *exts = *extents;
6087         struct btrfs_key found_key;
6088         u64 cur_pos;
6089         u64 last_byte;
6090         u32 nritems;
6091         int nr = 0;
6092         int max = *nr_extents;
6093         int ret;
6094
6095         WARN_ON(!no_fragment && *extents);
6096         if (!exts) {
6097                 max = 1;
6098                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6099                 if (!exts)
6100                         return -ENOMEM;
6101         }
6102
6103         path = btrfs_alloc_path();
6104         BUG_ON(!path);
6105
6106         cur_pos = extent_key->objectid - offset;
6107         last_byte = extent_key->objectid + extent_key->offset;
6108         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6109                                        cur_pos, 0);
6110         if (ret < 0)
6111                 goto out;
6112         if (ret > 0) {
6113                 ret = -ENOENT;
6114                 goto out;
6115         }
6116
6117         while (1) {
6118                 leaf = path->nodes[0];
6119                 nritems = btrfs_header_nritems(leaf);
6120                 if (path->slots[0] >= nritems) {
6121                         ret = btrfs_next_leaf(root, path);
6122                         if (ret < 0)
6123                                 goto out;
6124                         if (ret > 0)
6125                                 break;
6126                         leaf = path->nodes[0];
6127                 }
6128
6129                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6130                 if (found_key.offset != cur_pos ||
6131                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6132                     found_key.objectid != reloc_inode->i_ino)
6133                         break;
6134
6135                 fi = btrfs_item_ptr(leaf, path->slots[0],
6136                                     struct btrfs_file_extent_item);
6137                 if (btrfs_file_extent_type(leaf, fi) !=
6138                     BTRFS_FILE_EXTENT_REG ||
6139                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6140                         break;
6141
6142                 if (nr == max) {
6143                         struct disk_extent *old = exts;
6144                         max *= 2;
6145                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6146                         memcpy(exts, old, sizeof(*exts) * nr);
6147                         if (old != *extents)
6148                                 kfree(old);
6149                 }
6150
6151                 exts[nr].disk_bytenr =
6152                         btrfs_file_extent_disk_bytenr(leaf, fi);
6153                 exts[nr].disk_num_bytes =
6154                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6155                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6156                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6157                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6158                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6159                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6160                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6161                                                                            fi);
6162                 BUG_ON(exts[nr].offset > 0);
6163                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6164                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6165
6166                 cur_pos += exts[nr].num_bytes;
6167                 nr++;
6168
6169                 if (cur_pos + offset >= last_byte)
6170                         break;
6171
6172                 if (no_fragment) {
6173                         ret = 1;
6174                         goto out;
6175                 }
6176                 path->slots[0]++;
6177         }
6178
6179         BUG_ON(cur_pos + offset > last_byte);
6180         if (cur_pos + offset < last_byte) {
6181                 ret = -ENOENT;
6182                 goto out;
6183         }
6184         ret = 0;
6185 out:
6186         btrfs_free_path(path);
6187         if (ret) {
6188                 if (exts != *extents)
6189                         kfree(exts);
6190         } else {
6191                 *extents = exts;
6192                 *nr_extents = nr;
6193         }
6194         return ret;
6195 }
6196
6197 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6198                                         struct btrfs_root *root,
6199                                         struct btrfs_path *path,
6200                                         struct btrfs_key *extent_key,
6201                                         struct btrfs_key *leaf_key,
6202                                         struct btrfs_ref_path *ref_path,
6203                                         struct disk_extent *new_extents,
6204                                         int nr_extents)
6205 {
6206         struct extent_buffer *leaf;
6207         struct btrfs_file_extent_item *fi;
6208         struct inode *inode = NULL;
6209         struct btrfs_key key;
6210         u64 lock_start = 0;
6211         u64 lock_end = 0;
6212         u64 num_bytes;
6213         u64 ext_offset;
6214         u64 search_end = (u64)-1;
6215         u32 nritems;
6216         int nr_scaned = 0;
6217         int extent_locked = 0;
6218         int extent_type;
6219         int ret;
6220
6221         memcpy(&key, leaf_key, sizeof(key));
6222         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6223                 if (key.objectid < ref_path->owner_objectid ||
6224                     (key.objectid == ref_path->owner_objectid &&
6225                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6226                         key.objectid = ref_path->owner_objectid;
6227                         key.type = BTRFS_EXTENT_DATA_KEY;
6228                         key.offset = 0;
6229                 }
6230         }
6231
6232         while (1) {
6233                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6234                 if (ret < 0)
6235                         goto out;
6236
6237                 leaf = path->nodes[0];
6238                 nritems = btrfs_header_nritems(leaf);
6239 next:
6240                 if (extent_locked && ret > 0) {
6241                         /*
6242                          * the file extent item was modified by someone
6243                          * before the extent got locked.
6244                          */
6245                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6246                                       lock_end, GFP_NOFS);
6247                         extent_locked = 0;
6248                 }
6249
6250                 if (path->slots[0] >= nritems) {
6251                         if (++nr_scaned > 2)
6252                                 break;
6253
6254                         BUG_ON(extent_locked);
6255                         ret = btrfs_next_leaf(root, path);
6256                         if (ret < 0)
6257                                 goto out;
6258                         if (ret > 0)
6259                                 break;
6260                         leaf = path->nodes[0];
6261                         nritems = btrfs_header_nritems(leaf);
6262                 }
6263
6264                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6265
6266                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6267                         if ((key.objectid > ref_path->owner_objectid) ||
6268                             (key.objectid == ref_path->owner_objectid &&
6269                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6270                             key.offset >= search_end)
6271                                 break;
6272                 }
6273
6274                 if (inode && key.objectid != inode->i_ino) {
6275                         BUG_ON(extent_locked);
6276                         btrfs_release_path(root, path);
6277                         mutex_unlock(&inode->i_mutex);
6278                         iput(inode);
6279                         inode = NULL;
6280                         continue;
6281                 }
6282
6283                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6284                         path->slots[0]++;
6285                         ret = 1;
6286                         goto next;
6287                 }
6288                 fi = btrfs_item_ptr(leaf, path->slots[0],
6289                                     struct btrfs_file_extent_item);
6290                 extent_type = btrfs_file_extent_type(leaf, fi);
6291                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6292                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6293                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6294                      extent_key->objectid)) {
6295                         path->slots[0]++;
6296                         ret = 1;
6297                         goto next;
6298                 }
6299
6300                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6301                 ext_offset = btrfs_file_extent_offset(leaf, fi);
6302
6303                 if (search_end == (u64)-1) {
6304                         search_end = key.offset - ext_offset +
6305                                 btrfs_file_extent_ram_bytes(leaf, fi);
6306                 }
6307
6308                 if (!extent_locked) {
6309                         lock_start = key.offset;
6310                         lock_end = lock_start + num_bytes - 1;
6311                 } else {
6312                         if (lock_start > key.offset ||
6313                             lock_end + 1 < key.offset + num_bytes) {
6314                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6315                                               lock_start, lock_end, GFP_NOFS);
6316                                 extent_locked = 0;
6317                         }
6318                 }
6319
6320                 if (!inode) {
6321                         btrfs_release_path(root, path);
6322
6323                         inode = btrfs_iget_locked(root->fs_info->sb,
6324                                                   key.objectid, root);
6325                         if (inode->i_state & I_NEW) {
6326                                 BTRFS_I(inode)->root = root;
6327                                 BTRFS_I(inode)->location.objectid =
6328                                         key.objectid;
6329                                 BTRFS_I(inode)->location.type =
6330                                         BTRFS_INODE_ITEM_KEY;
6331                                 BTRFS_I(inode)->location.offset = 0;
6332                                 btrfs_read_locked_inode(inode);
6333                                 unlock_new_inode(inode);
6334                         }
6335                         /*
6336                          * some code call btrfs_commit_transaction while
6337                          * holding the i_mutex, so we can't use mutex_lock
6338                          * here.
6339                          */
6340                         if (is_bad_inode(inode) ||
6341                             !mutex_trylock(&inode->i_mutex)) {
6342                                 iput(inode);
6343                                 inode = NULL;
6344                                 key.offset = (u64)-1;
6345                                 goto skip;
6346                         }
6347                 }
6348
6349                 if (!extent_locked) {
6350                         struct btrfs_ordered_extent *ordered;
6351
6352                         btrfs_release_path(root, path);
6353
6354                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6355                                     lock_end, GFP_NOFS);
6356                         ordered = btrfs_lookup_first_ordered_extent(inode,
6357                                                                     lock_end);
6358                         if (ordered &&
6359                             ordered->file_offset <= lock_end &&
6360                             ordered->file_offset + ordered->len > lock_start) {
6361                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6362                                               lock_start, lock_end, GFP_NOFS);
6363                                 btrfs_start_ordered_extent(inode, ordered, 1);
6364                                 btrfs_put_ordered_extent(ordered);
6365                                 key.offset += num_bytes;
6366                                 goto skip;
6367                         }
6368                         if (ordered)
6369                                 btrfs_put_ordered_extent(ordered);
6370
6371                         extent_locked = 1;
6372                         continue;
6373                 }
6374
6375                 if (nr_extents == 1) {
6376                         /* update extent pointer in place */
6377                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
6378                                                 new_extents[0].disk_bytenr);
6379                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6380                                                 new_extents[0].disk_num_bytes);
6381                         btrfs_mark_buffer_dirty(leaf);
6382
6383                         btrfs_drop_extent_cache(inode, key.offset,
6384                                                 key.offset + num_bytes - 1, 0);
6385
6386                         ret = btrfs_inc_extent_ref(trans, root,
6387                                                 new_extents[0].disk_bytenr,
6388                                                 new_extents[0].disk_num_bytes,
6389                                                 leaf->start,
6390                                                 root->root_key.objectid,
6391                                                 trans->transid,
6392                                                 key.objectid);
6393                         BUG_ON(ret);
6394
6395                         ret = btrfs_free_extent(trans, root,
6396                                                 extent_key->objectid,
6397                                                 extent_key->offset,
6398                                                 leaf->start,
6399                                                 btrfs_header_owner(leaf),
6400                                                 btrfs_header_generation(leaf),
6401                                                 key.objectid, 0);
6402                         BUG_ON(ret);
6403
6404                         btrfs_release_path(root, path);
6405                         key.offset += num_bytes;
6406                 } else {
6407                         BUG_ON(1);
6408 #if 0
6409                         u64 alloc_hint;
6410                         u64 extent_len;
6411                         int i;
6412                         /*
6413                          * drop old extent pointer at first, then insert the
6414                          * new pointers one bye one
6415                          */
6416                         btrfs_release_path(root, path);
6417                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
6418                                                  key.offset + num_bytes,
6419                                                  key.offset, &alloc_hint);
6420                         BUG_ON(ret);
6421
6422                         for (i = 0; i < nr_extents; i++) {
6423                                 if (ext_offset >= new_extents[i].num_bytes) {
6424                                         ext_offset -= new_extents[i].num_bytes;
6425                                         continue;
6426                                 }
6427                                 extent_len = min(new_extents[i].num_bytes -
6428                                                  ext_offset, num_bytes);
6429
6430                                 ret = btrfs_insert_empty_item(trans, root,
6431                                                               path, &key,
6432                                                               sizeof(*fi));
6433                                 BUG_ON(ret);
6434
6435                                 leaf = path->nodes[0];
6436                                 fi = btrfs_item_ptr(leaf, path->slots[0],
6437                                                 struct btrfs_file_extent_item);
6438                                 btrfs_set_file_extent_generation(leaf, fi,
6439                                                         trans->transid);
6440                                 btrfs_set_file_extent_type(leaf, fi,
6441                                                         BTRFS_FILE_EXTENT_REG);
6442                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6443                                                 new_extents[i].disk_bytenr);
6444                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6445                                                 new_extents[i].disk_num_bytes);
6446                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
6447                                                 new_extents[i].ram_bytes);
6448
6449                                 btrfs_set_file_extent_compression(leaf, fi,
6450                                                 new_extents[i].compression);
6451                                 btrfs_set_file_extent_encryption(leaf, fi,
6452                                                 new_extents[i].encryption);
6453                                 btrfs_set_file_extent_other_encoding(leaf, fi,
6454                                                 new_extents[i].other_encoding);
6455
6456                                 btrfs_set_file_extent_num_bytes(leaf, fi,
6457                                                         extent_len);
6458                                 ext_offset += new_extents[i].offset;
6459                                 btrfs_set_file_extent_offset(leaf, fi,
6460                                                         ext_offset);
6461                                 btrfs_mark_buffer_dirty(leaf);
6462
6463                                 btrfs_drop_extent_cache(inode, key.offset,
6464                                                 key.offset + extent_len - 1, 0);
6465
6466                                 ret = btrfs_inc_extent_ref(trans, root,
6467                                                 new_extents[i].disk_bytenr,
6468                                                 new_extents[i].disk_num_bytes,
6469                                                 leaf->start,
6470                                                 root->root_key.objectid,
6471                                                 trans->transid, key.objectid);
6472                                 BUG_ON(ret);
6473                                 btrfs_release_path(root, path);
6474
6475                                 inode_add_bytes(inode, extent_len);
6476
6477                                 ext_offset = 0;
6478                                 num_bytes -= extent_len;
6479                                 key.offset += extent_len;
6480
6481                                 if (num_bytes == 0)
6482                                         break;
6483                         }
6484                         BUG_ON(i >= nr_extents);
6485 #endif
6486                 }
6487
6488                 if (extent_locked) {
6489                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6490                                       lock_end, GFP_NOFS);
6491                         extent_locked = 0;
6492                 }
6493 skip:
6494                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6495                     key.offset >= search_end)
6496                         break;
6497
6498                 cond_resched();
6499         }
6500         ret = 0;
6501 out:
6502         btrfs_release_path(root, path);
6503         if (inode) {
6504                 mutex_unlock(&inode->i_mutex);
6505                 if (extent_locked) {
6506                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6507                                       lock_end, GFP_NOFS);
6508                 }
6509                 iput(inode);
6510         }
6511         return ret;
6512 }
6513
6514 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6515                                struct btrfs_root *root,
6516                                struct extent_buffer *buf, u64 orig_start)
6517 {
6518         int level;
6519         int ret;
6520
6521         BUG_ON(btrfs_header_generation(buf) != trans->transid);
6522         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6523
6524         level = btrfs_header_level(buf);
6525         if (level == 0) {
6526                 struct btrfs_leaf_ref *ref;
6527                 struct btrfs_leaf_ref *orig_ref;
6528
6529                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6530                 if (!orig_ref)
6531                         return -ENOENT;
6532
6533                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6534                 if (!ref) {
6535                         btrfs_free_leaf_ref(root, orig_ref);
6536                         return -ENOMEM;
6537                 }
6538
6539                 ref->nritems = orig_ref->nritems;
6540                 memcpy(ref->extents, orig_ref->extents,
6541                         sizeof(ref->extents[0]) * ref->nritems);
6542
6543                 btrfs_free_leaf_ref(root, orig_ref);
6544
6545                 ref->root_gen = trans->transid;
6546                 ref->bytenr = buf->start;
6547                 ref->owner = btrfs_header_owner(buf);
6548                 ref->generation = btrfs_header_generation(buf);
6549
6550                 ret = btrfs_add_leaf_ref(root, ref, 0);
6551                 WARN_ON(ret);
6552                 btrfs_free_leaf_ref(root, ref);
6553         }
6554         return 0;
6555 }
6556
6557 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6558                                         struct extent_buffer *leaf,
6559                                         struct btrfs_block_group_cache *group,
6560                                         struct btrfs_root *target_root)
6561 {
6562         struct btrfs_key key;
6563         struct inode *inode = NULL;
6564         struct btrfs_file_extent_item *fi;
6565         struct extent_state *cached_state = NULL;
6566         u64 num_bytes;
6567         u64 skip_objectid = 0;
6568         u32 nritems;
6569         u32 i;
6570
6571         nritems = btrfs_header_nritems(leaf);
6572         for (i = 0; i < nritems; i++) {
6573                 btrfs_item_key_to_cpu(leaf, &key, i);
6574                 if (key.objectid == skip_objectid ||
6575                     key.type != BTRFS_EXTENT_DATA_KEY)
6576                         continue;
6577                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6578                 if (btrfs_file_extent_type(leaf, fi) ==
6579                     BTRFS_FILE_EXTENT_INLINE)
6580                         continue;
6581                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6582                         continue;
6583                 if (!inode || inode->i_ino != key.objectid) {
6584                         iput(inode);
6585                         inode = btrfs_ilookup(target_root->fs_info->sb,
6586                                               key.objectid, target_root, 1);
6587                 }
6588                 if (!inode) {
6589                         skip_objectid = key.objectid;
6590                         continue;
6591                 }
6592                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6593
6594                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
6595                                  key.offset + num_bytes - 1, 0, &cached_state,
6596                                  GFP_NOFS);
6597                 btrfs_drop_extent_cache(inode, key.offset,
6598                                         key.offset + num_bytes - 1, 1);
6599                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
6600                                      key.offset + num_bytes - 1, &cached_state,
6601                                      GFP_NOFS);
6602                 cond_resched();
6603         }
6604         iput(inode);
6605         return 0;
6606 }
6607
6608 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6609                                         struct btrfs_root *root,
6610                                         struct extent_buffer *leaf,
6611                                         struct btrfs_block_group_cache *group,
6612                                         struct inode *reloc_inode)
6613 {
6614         struct btrfs_key key;
6615         struct btrfs_key extent_key;
6616         struct btrfs_file_extent_item *fi;
6617         struct btrfs_leaf_ref *ref;
6618         struct disk_extent *new_extent;
6619         u64 bytenr;
6620         u64 num_bytes;
6621         u32 nritems;
6622         u32 i;
6623         int ext_index;
6624         int nr_extent;
6625         int ret;
6626
6627         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6628         BUG_ON(!new_extent);
6629
6630         ref = btrfs_lookup_leaf_ref(root, leaf->start);
6631         BUG_ON(!ref);
6632
6633         ext_index = -1;
6634         nritems = btrfs_header_nritems(leaf);
6635         for (i = 0; i < nritems; i++) {
6636                 btrfs_item_key_to_cpu(leaf, &key, i);
6637                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6638                         continue;
6639                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6640                 if (btrfs_file_extent_type(leaf, fi) ==
6641                     BTRFS_FILE_EXTENT_INLINE)
6642                         continue;
6643                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6644                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6645                 if (bytenr == 0)
6646                         continue;
6647
6648                 ext_index++;
6649                 if (bytenr >= group->key.objectid + group->key.offset ||
6650                     bytenr + num_bytes <= group->key.objectid)
6651                         continue;
6652
6653                 extent_key.objectid = bytenr;
6654                 extent_key.offset = num_bytes;
6655                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6656                 nr_extent = 1;
6657                 ret = get_new_locations(reloc_inode, &extent_key,
6658                                         group->key.objectid, 1,
6659                                         &new_extent, &nr_extent);
6660                 if (ret > 0)
6661                         continue;
6662                 BUG_ON(ret < 0);
6663
6664                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6665                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6666                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6667                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6668
6669                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6670                                                 new_extent->disk_bytenr);
6671                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6672                                                 new_extent->disk_num_bytes);
6673                 btrfs_mark_buffer_dirty(leaf);
6674
6675                 ret = btrfs_inc_extent_ref(trans, root,
6676                                         new_extent->disk_bytenr,
6677                                         new_extent->disk_num_bytes,
6678                                         leaf->start,
6679                                         root->root_key.objectid,
6680                                         trans->transid, key.objectid);
6681                 BUG_ON(ret);
6682
6683                 ret = btrfs_free_extent(trans, root,
6684                                         bytenr, num_bytes, leaf->start,
6685                                         btrfs_header_owner(leaf),
6686                                         btrfs_header_generation(leaf),
6687                                         key.objectid, 0);
6688                 BUG_ON(ret);
6689                 cond_resched();
6690         }
6691         kfree(new_extent);
6692         BUG_ON(ext_index + 1 != ref->nritems);
6693         btrfs_free_leaf_ref(root, ref);
6694         return 0;
6695 }
6696
6697 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6698                           struct btrfs_root *root)
6699 {
6700         struct btrfs_root *reloc_root;
6701         int ret;
6702
6703         if (root->reloc_root) {
6704                 reloc_root = root->reloc_root;
6705                 root->reloc_root = NULL;
6706                 list_add(&reloc_root->dead_list,
6707                          &root->fs_info->dead_reloc_roots);
6708
6709                 btrfs_set_root_bytenr(&reloc_root->root_item,
6710                                       reloc_root->node->start);
6711                 btrfs_set_root_level(&root->root_item,
6712                                      btrfs_header_level(reloc_root->node));
6713                 memset(&reloc_root->root_item.drop_progress, 0,
6714                         sizeof(struct btrfs_disk_key));
6715                 reloc_root->root_item.drop_level = 0;
6716
6717                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6718                                         &reloc_root->root_key,
6719                                         &reloc_root->root_item);
6720                 BUG_ON(ret);
6721         }
6722         return 0;
6723 }
6724
6725 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6726 {
6727         struct btrfs_trans_handle *trans;
6728         struct btrfs_root *reloc_root;
6729         struct btrfs_root *prev_root = NULL;
6730         struct list_head dead_roots;
6731         int ret;
6732         unsigned long nr;
6733
6734         INIT_LIST_HEAD(&dead_roots);
6735         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6736
6737         while (!list_empty(&dead_roots)) {
6738                 reloc_root = list_entry(dead_roots.prev,
6739                                         struct btrfs_root, dead_list);
6740                 list_del_init(&reloc_root->dead_list);
6741
6742                 BUG_ON(reloc_root->commit_root != NULL);
6743                 while (1) {
6744                         trans = btrfs_join_transaction(root, 1);
6745                         BUG_ON(!trans);
6746
6747                         mutex_lock(&root->fs_info->drop_mutex);
6748                         ret = btrfs_drop_snapshot(trans, reloc_root);
6749                         if (ret != -EAGAIN)
6750                                 break;
6751                         mutex_unlock(&root->fs_info->drop_mutex);
6752
6753                         nr = trans->blocks_used;
6754                         ret = btrfs_end_transaction(trans, root);
6755                         BUG_ON(ret);
6756                         btrfs_btree_balance_dirty(root, nr);
6757                 }
6758
6759                 free_extent_buffer(reloc_root->node);
6760
6761                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6762                                      &reloc_root->root_key);
6763                 BUG_ON(ret);
6764                 mutex_unlock(&root->fs_info->drop_mutex);
6765
6766                 nr = trans->blocks_used;
6767                 ret = btrfs_end_transaction(trans, root);
6768                 BUG_ON(ret);
6769                 btrfs_btree_balance_dirty(root, nr);
6770
6771                 kfree(prev_root);
6772                 prev_root = reloc_root;
6773         }
6774         if (prev_root) {
6775                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6776                 kfree(prev_root);
6777         }
6778         return 0;
6779 }
6780
6781 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6782 {
6783         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6784         return 0;
6785 }
6786
6787 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6788 {
6789         struct btrfs_root *reloc_root;
6790         struct btrfs_trans_handle *trans;
6791         struct btrfs_key location;
6792         int found;
6793         int ret;
6794
6795         mutex_lock(&root->fs_info->tree_reloc_mutex);
6796         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6797         BUG_ON(ret);
6798         found = !list_empty(&root->fs_info->dead_reloc_roots);
6799         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6800
6801         if (found) {
6802                 trans = btrfs_start_transaction(root, 1);
6803                 BUG_ON(!trans);
6804                 ret = btrfs_commit_transaction(trans, root);
6805                 BUG_ON(ret);
6806         }
6807
6808         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6809         location.offset = (u64)-1;
6810         location.type = BTRFS_ROOT_ITEM_KEY;
6811
6812         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6813         BUG_ON(!reloc_root);
6814         btrfs_orphan_cleanup(reloc_root);
6815         return 0;
6816 }
6817
6818 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6819                                     struct btrfs_root *root)
6820 {
6821         struct btrfs_root *reloc_root;
6822         struct extent_buffer *eb;
6823         struct btrfs_root_item *root_item;
6824         struct btrfs_key root_key;
6825         int ret;
6826
6827         BUG_ON(!root->ref_cows);
6828         if (root->reloc_root)
6829                 return 0;
6830
6831         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6832         BUG_ON(!root_item);
6833
6834         ret = btrfs_copy_root(trans, root, root->commit_root,
6835                               &eb, BTRFS_TREE_RELOC_OBJECTID);
6836         BUG_ON(ret);
6837
6838         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6839         root_key.offset = root->root_key.objectid;
6840         root_key.type = BTRFS_ROOT_ITEM_KEY;
6841
6842         memcpy(root_item, &root->root_item, sizeof(root_item));
6843         btrfs_set_root_refs(root_item, 0);
6844         btrfs_set_root_bytenr(root_item, eb->start);
6845         btrfs_set_root_level(root_item, btrfs_header_level(eb));
6846         btrfs_set_root_generation(root_item, trans->transid);
6847
6848         btrfs_tree_unlock(eb);
6849         free_extent_buffer(eb);
6850
6851         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6852                                 &root_key, root_item);
6853         BUG_ON(ret);
6854         kfree(root_item);
6855
6856         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6857                                                  &root_key);
6858         BUG_ON(!reloc_root);
6859         reloc_root->last_trans = trans->transid;
6860         reloc_root->commit_root = NULL;
6861         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6862
6863         root->reloc_root = reloc_root;
6864         return 0;
6865 }
6866
6867 /*
6868  * Core function of space balance.
6869  *
6870  * The idea is using reloc trees to relocate tree blocks in reference
6871  * counted roots. There is one reloc tree for each subvol, and all
6872  * reloc trees share same root key objectid. Reloc trees are snapshots
6873  * of the latest committed roots of subvols (root->commit_root).
6874  *
6875  * To relocate a tree block referenced by a subvol, there are two steps.
6876  * COW the block through subvol's reloc tree, then update block pointer
6877  * in the subvol to point to the new block. Since all reloc trees share
6878  * same root key objectid, doing special handing for tree blocks owned
6879  * by them is easy. Once a tree block has been COWed in one reloc tree,
6880  * we can use the resulting new block directly when the same block is
6881  * required to COW again through other reloc trees. By this way, relocated
6882  * tree blocks are shared between reloc trees, so they are also shared
6883  * between subvols.
6884  */
6885 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6886                                       struct btrfs_root *root,
6887                                       struct btrfs_path *path,
6888                                       struct btrfs_key *first_key,
6889                                       struct btrfs_ref_path *ref_path,
6890                                       struct btrfs_block_group_cache *group,
6891                                       struct inode *reloc_inode)
6892 {
6893         struct btrfs_root *reloc_root;
6894         struct extent_buffer *eb = NULL;
6895         struct btrfs_key *keys;
6896         u64 *nodes;
6897         int level;
6898         int shared_level;
6899         int lowest_level = 0;
6900         int ret;
6901
6902         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6903                 lowest_level = ref_path->owner_objectid;
6904
6905         if (!root->ref_cows) {
6906                 path->lowest_level = lowest_level;
6907                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6908                 BUG_ON(ret < 0);
6909                 path->lowest_level = 0;
6910                 btrfs_release_path(root, path);
6911                 return 0;
6912         }
6913
6914         mutex_lock(&root->fs_info->tree_reloc_mutex);
6915         ret = init_reloc_tree(trans, root);
6916         BUG_ON(ret);
6917         reloc_root = root->reloc_root;
6918
6919         shared_level = ref_path->shared_level;
6920         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6921
6922         keys = ref_path->node_keys;
6923         nodes = ref_path->new_nodes;
6924         memset(&keys[shared_level + 1], 0,
6925                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6926         memset(&nodes[shared_level + 1], 0,
6927                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6928
6929         if (nodes[lowest_level] == 0) {
6930                 path->lowest_level = lowest_level;
6931                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6932                                         0, 1);
6933                 BUG_ON(ret);
6934                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6935                         eb = path->nodes[level];
6936                         if (!eb || eb == reloc_root->node)
6937                                 break;
6938                         nodes[level] = eb->start;
6939                         if (level == 0)
6940                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6941                         else
6942                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6943                 }
6944                 if (nodes[0] &&
6945                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6946                         eb = path->nodes[0];
6947                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
6948                                                       group, reloc_inode);
6949                         BUG_ON(ret);
6950                 }
6951                 btrfs_release_path(reloc_root, path);
6952         } else {
6953                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6954                                        lowest_level);
6955                 BUG_ON(ret);
6956         }
6957
6958         /*
6959          * replace tree blocks in the fs tree with tree blocks in
6960          * the reloc tree.
6961          */
6962         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6963         BUG_ON(ret < 0);
6964
6965         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6966                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6967                                         0, 0);
6968                 BUG_ON(ret);
6969                 extent_buffer_get(path->nodes[0]);
6970                 eb = path->nodes[0];
6971                 btrfs_release_path(reloc_root, path);
6972                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6973                 BUG_ON(ret);
6974                 free_extent_buffer(eb);
6975         }
6976
6977         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6978         path->lowest_level = 0;
6979         return 0;
6980 }
6981
6982 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6983                                         struct btrfs_root *root,
6984                                         struct btrfs_path *path,
6985                                         struct btrfs_key *first_key,
6986                                         struct btrfs_ref_path *ref_path)
6987 {
6988         int ret;
6989
6990         ret = relocate_one_path(trans, root, path, first_key,
6991                                 ref_path, NULL, NULL);
6992         BUG_ON(ret);
6993
6994         return 0;
6995 }
6996
6997 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6998                                     struct btrfs_root *extent_root,
6999                                     struct btrfs_path *path,
7000                                     struct btrfs_key *extent_key)
7001 {
7002         int ret;
7003
7004         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7005         if (ret)
7006                 goto out;
7007         ret = btrfs_del_item(trans, extent_root, path);
7008 out:
7009         btrfs_release_path(extent_root, path);
7010         return ret;
7011 }
7012
7013 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7014                                                 struct btrfs_ref_path *ref_path)
7015 {
7016         struct btrfs_key root_key;
7017
7018         root_key.objectid = ref_path->root_objectid;
7019         root_key.type = BTRFS_ROOT_ITEM_KEY;
7020         if (is_cowonly_root(ref_path->root_objectid))
7021                 root_key.offset = 0;
7022         else
7023                 root_key.offset = (u64)-1;
7024
7025         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7026 }
7027
7028 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7029                                         struct btrfs_path *path,
7030                                         struct btrfs_key *extent_key,
7031                                         struct btrfs_block_group_cache *group,
7032                                         struct inode *reloc_inode, int pass)
7033 {
7034         struct btrfs_trans_handle *trans;
7035         struct btrfs_root *found_root;
7036         struct btrfs_ref_path *ref_path = NULL;
7037         struct disk_extent *new_extents = NULL;
7038         int nr_extents = 0;
7039         int loops;
7040         int ret;
7041         int level;
7042         struct btrfs_key first_key;
7043         u64 prev_block = 0;
7044
7045
7046         trans = btrfs_start_transaction(extent_root, 1);
7047         BUG_ON(!trans);
7048
7049         if (extent_key->objectid == 0) {
7050                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7051                 goto out;
7052         }
7053
7054         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7055         if (!ref_path) {
7056                 ret = -ENOMEM;
7057                 goto out;
7058         }
7059
7060         for (loops = 0; ; loops++) {
7061                 if (loops == 0) {
7062                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7063                                                    extent_key->objectid);
7064                 } else {
7065                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7066                 }
7067                 if (ret < 0)
7068                         goto out;
7069                 if (ret > 0)
7070                         break;
7071
7072                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7073                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7074                         continue;
7075
7076                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7077                 BUG_ON(!found_root);
7078                 /*
7079                  * for reference counted tree, only process reference paths
7080                  * rooted at the latest committed root.
7081                  */
7082                 if (found_root->ref_cows &&
7083                     ref_path->root_generation != found_root->root_key.offset)
7084                         continue;
7085
7086                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7087                         if (pass == 0) {
7088                                 /*
7089                                  * copy data extents to new locations
7090                                  */
7091                                 u64 group_start = group->key.objectid;
7092                                 ret = relocate_data_extent(reloc_inode,
7093                                                            extent_key,
7094                                                            group_start);
7095                                 if (ret < 0)
7096                                         goto out;
7097                                 break;
7098                         }
7099                         level = 0;
7100                 } else {
7101                         level = ref_path->owner_objectid;
7102                 }
7103
7104                 if (prev_block != ref_path->nodes[level]) {
7105                         struct extent_buffer *eb;
7106                         u64 block_start = ref_path->nodes[level];
7107                         u64 block_size = btrfs_level_size(found_root, level);
7108
7109                         eb = read_tree_block(found_root, block_start,
7110                                              block_size, 0);
7111                         btrfs_tree_lock(eb);
7112                         BUG_ON(level != btrfs_header_level(eb));
7113
7114                         if (level == 0)
7115                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7116                         else
7117                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7118
7119                         btrfs_tree_unlock(eb);
7120                         free_extent_buffer(eb);
7121                         prev_block = block_start;
7122                 }
7123
7124                 mutex_lock(&extent_root->fs_info->trans_mutex);
7125                 btrfs_record_root_in_trans(found_root);
7126                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7127                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7128                         /*
7129                          * try to update data extent references while
7130                          * keeping metadata shared between snapshots.
7131                          */
7132                         if (pass == 1) {
7133                                 ret = relocate_one_path(trans, found_root,
7134                                                 path, &first_key, ref_path,
7135                                                 group, reloc_inode);
7136                                 if (ret < 0)
7137                                         goto out;
7138                                 continue;
7139                         }
7140                         /*
7141                          * use fallback method to process the remaining
7142                          * references.
7143                          */
7144                         if (!new_extents) {
7145                                 u64 group_start = group->key.objectid;
7146                                 new_extents = kmalloc(sizeof(*new_extents),
7147                                                       GFP_NOFS);
7148                                 nr_extents = 1;
7149                                 ret = get_new_locations(reloc_inode,
7150                                                         extent_key,
7151                                                         group_start, 1,
7152                                                         &new_extents,
7153                                                         &nr_extents);
7154                                 if (ret)
7155                                         goto out;
7156                         }
7157                         ret = replace_one_extent(trans, found_root,
7158                                                 path, extent_key,
7159                                                 &first_key, ref_path,
7160                                                 new_extents, nr_extents);
7161                 } else {
7162                         ret = relocate_tree_block(trans, found_root, path,
7163                                                   &first_key, ref_path);
7164                 }
7165                 if (ret < 0)
7166                         goto out;
7167         }
7168         ret = 0;
7169 out:
7170         btrfs_end_transaction(trans, extent_root);
7171         kfree(new_extents);
7172         kfree(ref_path);
7173         return ret;
7174 }
7175 #endif
7176
7177 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7178 {
7179         u64 num_devices;
7180         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7181                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7182
7183         num_devices = root->fs_info->fs_devices->rw_devices;
7184         if (num_devices == 1) {
7185                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7186                 stripped = flags & ~stripped;
7187
7188                 /* turn raid0 into single device chunks */
7189                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7190                         return stripped;
7191
7192                 /* turn mirroring into duplication */
7193                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7194                              BTRFS_BLOCK_GROUP_RAID10))
7195                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7196                 return flags;
7197         } else {
7198                 /* they already had raid on here, just return */
7199                 if (flags & stripped)
7200                         return flags;
7201
7202                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7203                 stripped = flags & ~stripped;
7204
7205                 /* switch duplicated blocks with raid1 */
7206                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7207                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7208
7209                 /* turn single device chunks into raid0 */
7210                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7211         }
7212         return flags;
7213 }
7214
7215 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
7216                      struct btrfs_block_group_cache *shrink_block_group,
7217                      int force)
7218 {
7219         struct btrfs_trans_handle *trans;
7220         u64 new_alloc_flags;
7221         u64 calc;
7222
7223         spin_lock(&shrink_block_group->lock);
7224         if (btrfs_block_group_used(&shrink_block_group->item) +
7225             shrink_block_group->reserved > 0) {
7226                 spin_unlock(&shrink_block_group->lock);
7227
7228                 trans = btrfs_start_transaction(root, 1);
7229                 spin_lock(&shrink_block_group->lock);
7230
7231                 new_alloc_flags = update_block_group_flags(root,
7232                                                    shrink_block_group->flags);
7233                 if (new_alloc_flags != shrink_block_group->flags) {
7234                         calc =
7235                              btrfs_block_group_used(&shrink_block_group->item);
7236                 } else {
7237                         calc = shrink_block_group->key.offset;
7238                 }
7239                 spin_unlock(&shrink_block_group->lock);
7240
7241                 do_chunk_alloc(trans, root->fs_info->extent_root,
7242                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
7243
7244                 btrfs_end_transaction(trans, root);
7245         } else
7246                 spin_unlock(&shrink_block_group->lock);
7247         return 0;
7248 }
7249
7250
7251 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
7252                                          struct btrfs_block_group_cache *group)
7253
7254 {
7255         __alloc_chunk_for_shrink(root, group, 1);
7256         set_block_group_readonly(group);
7257         return 0;
7258 }
7259
7260 /*
7261  * checks to see if its even possible to relocate this block group.
7262  *
7263  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7264  * ok to go ahead and try.
7265  */
7266 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7267 {
7268         struct btrfs_block_group_cache *block_group;
7269         struct btrfs_space_info *space_info;
7270         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7271         struct btrfs_device *device;
7272         int full = 0;
7273         int ret = 0;
7274
7275         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7276
7277         /* odd, couldn't find the block group, leave it alone */
7278         if (!block_group)
7279                 return -1;
7280
7281         /* no bytes used, we're good */
7282         if (!btrfs_block_group_used(&block_group->item))
7283                 goto out;
7284
7285         space_info = block_group->space_info;
7286         spin_lock(&space_info->lock);
7287
7288         full = space_info->full;
7289
7290         /*
7291          * if this is the last block group we have in this space, we can't
7292          * relocate it unless we're able to allocate a new chunk below.
7293          *
7294          * Otherwise, we need to make sure we have room in the space to handle
7295          * all of the extents from this block group.  If we can, we're good
7296          */
7297         if ((space_info->total_bytes != block_group->key.offset) &&
7298            (space_info->bytes_used + space_info->bytes_reserved +
7299             space_info->bytes_pinned + space_info->bytes_readonly +
7300             btrfs_block_group_used(&block_group->item) <
7301             space_info->total_bytes)) {
7302                 spin_unlock(&space_info->lock);
7303                 goto out;
7304         }
7305         spin_unlock(&space_info->lock);
7306
7307         /*
7308          * ok we don't have enough space, but maybe we have free space on our
7309          * devices to allocate new chunks for relocation, so loop through our
7310          * alloc devices and guess if we have enough space.  However, if we
7311          * were marked as full, then we know there aren't enough chunks, and we
7312          * can just return.
7313          */
7314         ret = -1;
7315         if (full)
7316                 goto out;
7317
7318         mutex_lock(&root->fs_info->chunk_mutex);
7319         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7320                 u64 min_free = btrfs_block_group_used(&block_group->item);
7321                 u64 dev_offset, max_avail;
7322
7323                 /*
7324                  * check to make sure we can actually find a chunk with enough
7325                  * space to fit our block group in.
7326                  */
7327                 if (device->total_bytes > device->bytes_used + min_free) {
7328                         ret = find_free_dev_extent(NULL, device, min_free,
7329                                                    &dev_offset, &max_avail);
7330                         if (!ret)
7331                                 break;
7332                         ret = -1;
7333                 }
7334         }
7335         mutex_unlock(&root->fs_info->chunk_mutex);
7336 out:
7337         btrfs_put_block_group(block_group);
7338         return ret;
7339 }
7340
7341 static int find_first_block_group(struct btrfs_root *root,
7342                 struct btrfs_path *path, struct btrfs_key *key)
7343 {
7344         int ret = 0;
7345         struct btrfs_key found_key;
7346         struct extent_buffer *leaf;
7347         int slot;
7348
7349         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7350         if (ret < 0)
7351                 goto out;
7352
7353         while (1) {
7354                 slot = path->slots[0];
7355                 leaf = path->nodes[0];
7356                 if (slot >= btrfs_header_nritems(leaf)) {
7357                         ret = btrfs_next_leaf(root, path);
7358                         if (ret == 0)
7359                                 continue;
7360                         if (ret < 0)
7361                                 goto out;
7362                         break;
7363                 }
7364                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7365
7366                 if (found_key.objectid >= key->objectid &&
7367                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7368                         ret = 0;
7369                         goto out;
7370                 }
7371                 path->slots[0]++;
7372         }
7373         ret = -ENOENT;
7374 out:
7375         return ret;
7376 }
7377
7378 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7379 {
7380         struct btrfs_block_group_cache *block_group;
7381         struct btrfs_space_info *space_info;
7382         struct btrfs_caching_control *caching_ctl;
7383         struct rb_node *n;
7384
7385         down_write(&info->extent_commit_sem);
7386         while (!list_empty(&info->caching_block_groups)) {
7387                 caching_ctl = list_entry(info->caching_block_groups.next,
7388                                          struct btrfs_caching_control, list);
7389                 list_del(&caching_ctl->list);
7390                 put_caching_control(caching_ctl);
7391         }
7392         up_write(&info->extent_commit_sem);
7393
7394         spin_lock(&info->block_group_cache_lock);
7395         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7396                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7397                                        cache_node);
7398                 rb_erase(&block_group->cache_node,
7399                          &info->block_group_cache_tree);
7400                 spin_unlock(&info->block_group_cache_lock);
7401
7402                 down_write(&block_group->space_info->groups_sem);
7403                 list_del(&block_group->list);
7404                 up_write(&block_group->space_info->groups_sem);
7405
7406                 if (block_group->cached == BTRFS_CACHE_STARTED)
7407                         wait_block_group_cache_done(block_group);
7408
7409                 btrfs_remove_free_space_cache(block_group);
7410                 btrfs_put_block_group(block_group);
7411
7412                 spin_lock(&info->block_group_cache_lock);
7413         }
7414         spin_unlock(&info->block_group_cache_lock);
7415
7416         /* now that all the block groups are freed, go through and
7417          * free all the space_info structs.  This is only called during
7418          * the final stages of unmount, and so we know nobody is
7419          * using them.  We call synchronize_rcu() once before we start,
7420          * just to be on the safe side.
7421          */
7422         synchronize_rcu();
7423
7424         while(!list_empty(&info->space_info)) {
7425                 space_info = list_entry(info->space_info.next,
7426                                         struct btrfs_space_info,
7427                                         list);
7428
7429                 list_del(&space_info->list);
7430                 kfree(space_info);
7431         }
7432         return 0;
7433 }
7434
7435 int btrfs_read_block_groups(struct btrfs_root *root)
7436 {
7437         struct btrfs_path *path;
7438         int ret;
7439         struct btrfs_block_group_cache *cache;
7440         struct btrfs_fs_info *info = root->fs_info;
7441         struct btrfs_space_info *space_info;
7442         struct btrfs_key key;
7443         struct btrfs_key found_key;
7444         struct extent_buffer *leaf;
7445
7446         root = info->extent_root;
7447         key.objectid = 0;
7448         key.offset = 0;
7449         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7450         path = btrfs_alloc_path();
7451         if (!path)
7452                 return -ENOMEM;
7453
7454         while (1) {
7455                 ret = find_first_block_group(root, path, &key);
7456                 if (ret > 0) {
7457                         ret = 0;
7458                         goto error;
7459                 }
7460                 if (ret != 0)
7461                         goto error;
7462
7463                 leaf = path->nodes[0];
7464                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7465                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7466                 if (!cache) {
7467                         ret = -ENOMEM;
7468                         break;
7469                 }
7470
7471                 atomic_set(&cache->count, 1);
7472                 spin_lock_init(&cache->lock);
7473                 spin_lock_init(&cache->tree_lock);
7474                 cache->fs_info = info;
7475                 INIT_LIST_HEAD(&cache->list);
7476                 INIT_LIST_HEAD(&cache->cluster_list);
7477
7478                 /*
7479                  * we only want to have 32k of ram per block group for keeping
7480                  * track of free space, and if we pass 1/2 of that we want to
7481                  * start converting things over to using bitmaps
7482                  */
7483                 cache->extents_thresh = ((1024 * 32) / 2) /
7484                         sizeof(struct btrfs_free_space);
7485
7486                 read_extent_buffer(leaf, &cache->item,
7487                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7488                                    sizeof(cache->item));
7489                 memcpy(&cache->key, &found_key, sizeof(found_key));
7490
7491                 key.objectid = found_key.objectid + found_key.offset;
7492                 btrfs_release_path(root, path);
7493                 cache->flags = btrfs_block_group_flags(&cache->item);
7494                 cache->sectorsize = root->sectorsize;
7495
7496                 /*
7497                  * check for two cases, either we are full, and therefore
7498                  * don't need to bother with the caching work since we won't
7499                  * find any space, or we are empty, and we can just add all
7500                  * the space in and be done with it.  This saves us _alot_ of
7501                  * time, particularly in the full case.
7502                  */
7503                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7504                         exclude_super_stripes(root, cache);
7505                         cache->last_byte_to_unpin = (u64)-1;
7506                         cache->cached = BTRFS_CACHE_FINISHED;
7507                         free_excluded_extents(root, cache);
7508                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7509                         exclude_super_stripes(root, cache);
7510                         cache->last_byte_to_unpin = (u64)-1;
7511                         cache->cached = BTRFS_CACHE_FINISHED;
7512                         add_new_free_space(cache, root->fs_info,
7513                                            found_key.objectid,
7514                                            found_key.objectid +
7515                                            found_key.offset);
7516                         free_excluded_extents(root, cache);
7517                 }
7518
7519                 ret = update_space_info(info, cache->flags, found_key.offset,
7520                                         btrfs_block_group_used(&cache->item),
7521                                         &space_info);
7522                 BUG_ON(ret);
7523                 cache->space_info = space_info;
7524                 spin_lock(&cache->space_info->lock);
7525                 cache->space_info->bytes_super += cache->bytes_super;
7526                 spin_unlock(&cache->space_info->lock);
7527
7528                 down_write(&space_info->groups_sem);
7529                 list_add_tail(&cache->list, &space_info->block_groups);
7530                 up_write(&space_info->groups_sem);
7531
7532                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7533                 BUG_ON(ret);
7534
7535                 set_avail_alloc_bits(root->fs_info, cache->flags);
7536                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7537                         set_block_group_readonly(cache);
7538         }
7539         ret = 0;
7540 error:
7541         btrfs_free_path(path);
7542         return ret;
7543 }
7544
7545 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7546                            struct btrfs_root *root, u64 bytes_used,
7547                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7548                            u64 size)
7549 {
7550         int ret;
7551         struct btrfs_root *extent_root;
7552         struct btrfs_block_group_cache *cache;
7553
7554         extent_root = root->fs_info->extent_root;
7555
7556         root->fs_info->last_trans_log_full_commit = trans->transid;
7557
7558         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7559         if (!cache)
7560                 return -ENOMEM;
7561
7562         cache->key.objectid = chunk_offset;
7563         cache->key.offset = size;
7564         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7565         cache->sectorsize = root->sectorsize;
7566
7567         /*
7568          * we only want to have 32k of ram per block group for keeping track
7569          * of free space, and if we pass 1/2 of that we want to start
7570          * converting things over to using bitmaps
7571          */
7572         cache->extents_thresh = ((1024 * 32) / 2) /
7573                 sizeof(struct btrfs_free_space);
7574         atomic_set(&cache->count, 1);
7575         spin_lock_init(&cache->lock);
7576         spin_lock_init(&cache->tree_lock);
7577         INIT_LIST_HEAD(&cache->list);
7578         INIT_LIST_HEAD(&cache->cluster_list);
7579
7580         btrfs_set_block_group_used(&cache->item, bytes_used);
7581         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7582         cache->flags = type;
7583         btrfs_set_block_group_flags(&cache->item, type);
7584
7585         cache->last_byte_to_unpin = (u64)-1;
7586         cache->cached = BTRFS_CACHE_FINISHED;
7587         exclude_super_stripes(root, cache);
7588
7589         add_new_free_space(cache, root->fs_info, chunk_offset,
7590                            chunk_offset + size);
7591
7592         free_excluded_extents(root, cache);
7593
7594         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7595                                 &cache->space_info);
7596         BUG_ON(ret);
7597
7598         spin_lock(&cache->space_info->lock);
7599         cache->space_info->bytes_super += cache->bytes_super;
7600         spin_unlock(&cache->space_info->lock);
7601
7602         down_write(&cache->space_info->groups_sem);
7603         list_add_tail(&cache->list, &cache->space_info->block_groups);
7604         up_write(&cache->space_info->groups_sem);
7605
7606         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7607         BUG_ON(ret);
7608
7609         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7610                                 sizeof(cache->item));
7611         BUG_ON(ret);
7612
7613         set_avail_alloc_bits(extent_root->fs_info, type);
7614
7615         return 0;
7616 }
7617
7618 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7619                              struct btrfs_root *root, u64 group_start)
7620 {
7621         struct btrfs_path *path;
7622         struct btrfs_block_group_cache *block_group;
7623         struct btrfs_free_cluster *cluster;
7624         struct btrfs_key key;
7625         int ret;
7626
7627         root = root->fs_info->extent_root;
7628
7629         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7630         BUG_ON(!block_group);
7631         BUG_ON(!block_group->ro);
7632
7633         memcpy(&key, &block_group->key, sizeof(key));
7634
7635         /* make sure this block group isn't part of an allocation cluster */
7636         cluster = &root->fs_info->data_alloc_cluster;
7637         spin_lock(&cluster->refill_lock);
7638         btrfs_return_cluster_to_free_space(block_group, cluster);
7639         spin_unlock(&cluster->refill_lock);
7640
7641         /*
7642          * make sure this block group isn't part of a metadata
7643          * allocation cluster
7644          */
7645         cluster = &root->fs_info->meta_alloc_cluster;
7646         spin_lock(&cluster->refill_lock);
7647         btrfs_return_cluster_to_free_space(block_group, cluster);
7648         spin_unlock(&cluster->refill_lock);
7649
7650         path = btrfs_alloc_path();
7651         BUG_ON(!path);
7652
7653         spin_lock(&root->fs_info->block_group_cache_lock);
7654         rb_erase(&block_group->cache_node,
7655                  &root->fs_info->block_group_cache_tree);
7656         spin_unlock(&root->fs_info->block_group_cache_lock);
7657
7658         down_write(&block_group->space_info->groups_sem);
7659         /*
7660          * we must use list_del_init so people can check to see if they
7661          * are still on the list after taking the semaphore
7662          */
7663         list_del_init(&block_group->list);
7664         up_write(&block_group->space_info->groups_sem);
7665
7666         if (block_group->cached == BTRFS_CACHE_STARTED)
7667                 wait_block_group_cache_done(block_group);
7668
7669         btrfs_remove_free_space_cache(block_group);
7670
7671         spin_lock(&block_group->space_info->lock);
7672         block_group->space_info->total_bytes -= block_group->key.offset;
7673         block_group->space_info->bytes_readonly -= block_group->key.offset;
7674         spin_unlock(&block_group->space_info->lock);
7675
7676         btrfs_clear_space_info_full(root->fs_info);
7677
7678         btrfs_put_block_group(block_group);
7679         btrfs_put_block_group(block_group);
7680
7681         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7682         if (ret > 0)
7683                 ret = -EIO;
7684         if (ret < 0)
7685                 goto out;
7686
7687         ret = btrfs_del_item(trans, root, path);
7688 out:
7689         btrfs_free_path(path);
7690         return ret;
7691 }