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 /* control flags for do_chunk_alloc's force field
  37  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
  38  * if we really need one.
  39  *
  40  * CHUNK_ALLOC_FORCE means it must try to allocate one
  41  *
  42  * CHUNK_ALLOC_LIMITED means to only try and allocate one
  43  * if we have very few chunks already allocated.  This is
  44  * used as part of the clustering code to help make sure
  45  * we have a good pool of storage to cluster in, without
  46  * filling the FS with empty chunks
  47  *
  48  */
  49 enum {
  50         CHUNK_ALLOC_NO_FORCE = 0,
  51         CHUNK_ALLOC_FORCE = 1,
  52         CHUNK_ALLOC_LIMITED = 2,
  53 };
  54
  55 static int update_block_group(struct btrfs_trans_handle *trans,
  56                               struct btrfs_root *root,
  57                               u64 bytenr, u64 num_bytes, int alloc);
  58 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  59                                 struct btrfs_root *root,
  60                                 u64 bytenr, u64 num_bytes, u64 parent,
  61                                 u64 root_objectid, u64 owner_objectid,
  62                                 u64 owner_offset, int refs_to_drop,
  63                                 struct btrfs_delayed_extent_op *extra_op);
  64 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  65                                     struct extent_buffer *leaf,
  66                                     struct btrfs_extent_item *ei);
  67 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  68                                       struct btrfs_root *root,
  69                                       u64 parent, u64 root_objectid,
  70                                       u64 flags, u64 owner, u64 offset,
  71                                       struct btrfs_key *ins, int ref_mod);
  72 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  73                                      struct btrfs_root *root,
  74                                      u64 parent, u64 root_objectid,
  75                                      u64 flags, struct btrfs_disk_key *key,
  76                                      int level, struct btrfs_key *ins);
  77 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  78                           struct btrfs_root *extent_root, u64 alloc_bytes,
  79                           u64 flags, int force);
  80 static int find_next_key(struct btrfs_path *path, int level,
  81                          struct btrfs_key *key);
  82 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  83                             int dump_block_groups);
  84
  85 static noinline int
  86 block_group_cache_done(struct btrfs_block_group_cache *cache)
  87 {
  88         smp_mb();
  89         return cache->cached == BTRFS_CACHE_FINISHED;
  90 }
  91
  92 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  93 {
  94         return (cache->flags & bits) == bits;
  95 }
  96
  97 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
  98 {
  99         atomic_inc(&cache->count);
 100 }
 101
 102 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 103 {
 104         if (atomic_dec_and_test(&cache->count)) {
 105                 WARN_ON(cache->pinned > 0);
 106                 WARN_ON(cache->reserved > 0);
 107                 WARN_ON(cache->reserved_pinned > 0);
 108                 kfree(cache->free_space_ctl);
 109                 kfree(cache);
 110         }
 111 }
 112
 113 /*
 114  * this adds the block group to the fs_info rb tree for the block group
 115  * cache
 116  */
 117 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 118                                 struct btrfs_block_group_cache *block_group)
 119 {
 120         struct rb_node **p;
 121         struct rb_node *parent = NULL;
 122         struct btrfs_block_group_cache *cache;
 123
 124         spin_lock(&info->block_group_cache_lock);
 125         p = &info->block_group_cache_tree.rb_node;
 126
 127         while (*p) {
 128                 parent = *p;
 129                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 130                                  cache_node);
 131                 if (block_group->key.objectid < cache->key.objectid) {
 132                         p = &(*p)->rb_left;
 133                 } else if (block_group->key.objectid > cache->key.objectid) {
 134                         p = &(*p)->rb_right;
 135                 } else {
 136                         spin_unlock(&info->block_group_cache_lock);
 137                         return -EEXIST;
 138                 }
 139         }
 140
 141         rb_link_node(&block_group->cache_node, parent, p);
 142         rb_insert_color(&block_group->cache_node,
 143                         &info->block_group_cache_tree);
 144         spin_unlock(&info->block_group_cache_lock);
 145
 146         return 0;
 147 }
 148
 149 /*
 150  * This will return the block group at or after bytenr if contains is 0, else
 151  * it will return the block group that contains the bytenr
 152  */
 153 static struct btrfs_block_group_cache *
 154 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 155                               int contains)
 156 {
 157         struct btrfs_block_group_cache *cache, *ret = NULL;
 158         struct rb_node *n;
 159         u64 end, start;
 160
 161         spin_lock(&info->block_group_cache_lock);
 162         n = info->block_group_cache_tree.rb_node;
 163
 164         while (n) {
 165                 cache = rb_entry(n, struct btrfs_block_group_cache,
 166                                  cache_node);
 167                 end = cache->key.objectid + cache->key.offset - 1;
 168                 start = cache->key.objectid;
 169
 170                 if (bytenr < start) {
 171                         if (!contains && (!ret || start < ret->key.objectid))
 172                                 ret = cache;
 173                         n = n->rb_left;
 174                 } else if (bytenr > start) {
 175                         if (contains && bytenr <= end) {
 176                                 ret = cache;
 177                                 break;
 178                         }
 179                         n = n->rb_right;
 180                 } else {
 181                         ret = cache;
 182                         break;
 183                 }
 184         }
 185         if (ret)
 186                 btrfs_get_block_group(ret);
 187         spin_unlock(&info->block_group_cache_lock);
 188
 189         return ret;
 190 }
 191
 192 static int add_excluded_extent(struct btrfs_root *root,
 193                                u64 start, u64 num_bytes)
 194 {
 195         u64 end = start + num_bytes - 1;
 196         set_extent_bits(&root->fs_info->freed_extents[0],
 197                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 198         set_extent_bits(&root->fs_info->freed_extents[1],
 199                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 200         return 0;
 201 }
 202
 203 static void free_excluded_extents(struct btrfs_root *root,
 204                                   struct btrfs_block_group_cache *cache)
 205 {
 206         u64 start, end;
 207
 208         start = cache->key.objectid;
 209         end = start + cache->key.offset - 1;
 210
 211         clear_extent_bits(&root->fs_info->freed_extents[0],
 212                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 213         clear_extent_bits(&root->fs_info->freed_extents[1],
 214                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 215 }
 216
 217 static int exclude_super_stripes(struct btrfs_root *root,
 218                                  struct btrfs_block_group_cache *cache)
 219 {
 220         u64 bytenr;
 221         u64 *logical;
 222         int stripe_len;
 223         int i, nr, ret;
 224
 225         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 226                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 227                 cache->bytes_super += stripe_len;
 228                 ret = add_excluded_extent(root, cache->key.objectid,
 229                                           stripe_len);
 230                 BUG_ON(ret);
 231         }
 232
 233         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 234                 bytenr = btrfs_sb_offset(i);
 235                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 236                                        cache->key.objectid, bytenr,
 237                                        0, &logical, &nr, &stripe_len);
 238                 BUG_ON(ret);
 239
 240                 while (nr--) {
 241                         cache->bytes_super += stripe_len;
 242                         ret = add_excluded_extent(root, logical[nr],
 243                                                   stripe_len);
 244                         BUG_ON(ret);
 245                 }
 246
 247                 kfree(logical);
 248         }
 249         return 0;
 250 }
 251
 252 static struct btrfs_caching_control *
 253 get_caching_control(struct btrfs_block_group_cache *cache)
 254 {
 255         struct btrfs_caching_control *ctl;
 256
 257         spin_lock(&cache->lock);
 258         if (cache->cached != BTRFS_CACHE_STARTED) {
 259                 spin_unlock(&cache->lock);
 260                 return NULL;
 261         }
 262
 263         /* We're loading it the fast way, so we don't have a caching_ctl. */
 264         if (!cache->caching_ctl) {
 265                 spin_unlock(&cache->lock);
 266                 return NULL;
 267         }
 268
 269         ctl = cache->caching_ctl;
 270         atomic_inc(&ctl->count);
 271         spin_unlock(&cache->lock);
 272         return ctl;
 273 }
 274
 275 static void put_caching_control(struct btrfs_caching_control *ctl)
 276 {
 277         if (atomic_dec_and_test(&ctl->count))
 278                 kfree(ctl);
 279 }
 280
 281 /*
 282  * this is only called by cache_block_group, since we could have freed extents
 283  * we need to check the pinned_extents for any extents that can't be used yet
 284  * since their free space will be released as soon as the transaction commits.
 285  */
 286 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 287                               struct btrfs_fs_info *info, u64 start, u64 end)
 288 {
 289         u64 extent_start, extent_end, size, total_added = 0;
 290         int ret;
 291
 292         while (start < end) {
 293                 ret = find_first_extent_bit(info->pinned_extents, start,
 294                                             &extent_start, &extent_end,
 295                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 296                 if (ret)
 297                         break;
 298
 299                 if (extent_start <= start) {
 300                         start = extent_end + 1;
 301                 } else if (extent_start > start && extent_start < end) {
 302                         size = extent_start - start;
 303                         total_added += size;
 304                         ret = btrfs_add_free_space(block_group, start,
 305                                                    size);
 306                         BUG_ON(ret);
 307                         start = extent_end + 1;
 308                 } else {
 309                         break;
 310                 }
 311         }
 312
 313         if (start < end) {
 314                 size = end - start;
 315                 total_added += size;
 316                 ret = btrfs_add_free_space(block_group, start, size);
 317                 BUG_ON(ret);
 318         }
 319
 320         return total_added;
 321 }
 322
 323 static noinline void caching_thread(struct btrfs_work *work)
 324 {
 325         struct btrfs_block_group_cache *block_group;
 326         struct btrfs_fs_info *fs_info;
 327         struct btrfs_caching_control *caching_ctl;
 328         struct btrfs_root *extent_root;
 329         struct btrfs_path *path;
 330         struct extent_buffer *leaf;
 331         struct btrfs_key key;
 332         u64 total_found = 0;
 333         u64 last = 0;
 334         u32 nritems;
 335         int ret = 0;
 336
 337         caching_ctl = container_of(work, struct btrfs_caching_control, work);
 338         block_group = caching_ctl->block_group;
 339         fs_info = block_group->fs_info;
 340         extent_root = fs_info->extent_root;
 341
 342         path = btrfs_alloc_path();
 343         if (!path)
 344                 goto out;
 345
 346         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 347
 348         /*
 349          * We don't want to deadlock with somebody trying to allocate a new
 350          * extent for the extent root while also trying to search the extent
 351          * root to add free space.  So we skip locking and search the commit
 352          * root, since its read-only
 353          */
 354         path->skip_locking = 1;
 355         path->search_commit_root = 1;
 356         path->reada = 1;
 357
 358         key.objectid = last;
 359         key.offset = 0;
 360         key.type = BTRFS_EXTENT_ITEM_KEY;
 361 again:
 362         mutex_lock(&caching_ctl->mutex);
 363         /* need to make sure the commit_root doesn't disappear */
 364         down_read(&fs_info->extent_commit_sem);
 365
 366         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 367         if (ret < 0)
 368                 goto err;
 369
 370         leaf = path->nodes[0];
 371         nritems = btrfs_header_nritems(leaf);
 372
 373         while (1) {
 374                 if (btrfs_fs_closing(fs_info) > 1) {
 375                         last = (u64)-1;
 376                         break;
 377                 }
 378
 379                 if (path->slots[0] < nritems) {
 380                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 381                 } else {
 382                         ret = find_next_key(path, 0, &key);
 383                         if (ret)
 384                                 break;
 385
 386                         if (need_resched() ||
 387                             btrfs_next_leaf(extent_root, path)) {
 388                                 caching_ctl->progress = last;
 389                                 btrfs_release_path(path);
 390                                 up_read(&fs_info->extent_commit_sem);
 391                                 mutex_unlock(&caching_ctl->mutex);
 392                                 cond_resched();
 393                                 goto again;
 394                         }
 395                         leaf = path->nodes[0];
 396                         nritems = btrfs_header_nritems(leaf);
 397                         continue;
 398                 }
 399
 400                 if (key.objectid < block_group->key.objectid) {
 401                         path->slots[0]++;
 402                         continue;
 403                 }
 404
 405                 if (key.objectid >= block_group->key.objectid +
 406                     block_group->key.offset)
 407                         break;
 408
 409                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 410                         total_found += add_new_free_space(block_group,
 411                                                           fs_info, last,
 412                                                           key.objectid);
 413                         last = key.objectid + key.offset;
 414
 415                         if (total_found > (1024 * 1024 * 2)) {
 416                                 total_found = 0;
 417                                 wake_up(&caching_ctl->wait);
 418                         }
 419                 }
 420                 path->slots[0]++;
 421         }
 422         ret = 0;
 423
 424         total_found += add_new_free_space(block_group, fs_info, last,
 425                                           block_group->key.objectid +
 426                                           block_group->key.offset);
 427         caching_ctl->progress = (u64)-1;
 428
 429         spin_lock(&block_group->lock);
 430         block_group->caching_ctl = NULL;
 431         block_group->cached = BTRFS_CACHE_FINISHED;
 432         spin_unlock(&block_group->lock);
 433
 434 err:
 435         btrfs_free_path(path);
 436         up_read(&fs_info->extent_commit_sem);
 437
 438         free_excluded_extents(extent_root, block_group);
 439
 440         mutex_unlock(&caching_ctl->mutex);
 441 out:
 442         wake_up(&caching_ctl->wait);
 443
 444         put_caching_control(caching_ctl);
 445         btrfs_put_block_group(block_group);
 446 }
 447
 448 static int cache_block_group(struct btrfs_block_group_cache *cache,
 449                              struct btrfs_trans_handle *trans,
 450                              struct btrfs_root *root,
 451                              int load_cache_only)
 452 {
 453         struct btrfs_fs_info *fs_info = cache->fs_info;
 454         struct btrfs_caching_control *caching_ctl;
 455         int ret = 0;
 456
 457         smp_mb();
 458         if (cache->cached != BTRFS_CACHE_NO)
 459                 return 0;
 460
 461         /*
 462          * We can't do the read from on-disk cache during a commit since we need
 463          * to have the normal tree locking.  Also if we are currently trying to
 464          * allocate blocks for the tree root we can't do the fast caching since
 465          * we likely hold important locks.
 466          */
 467         if (trans && (!trans->transaction->in_commit) &&
 468             (root && root != root->fs_info->tree_root)) {
 469                 spin_lock(&cache->lock);
 470                 if (cache->cached != BTRFS_CACHE_NO) {
 471                         spin_unlock(&cache->lock);
 472                         return 0;
 473                 }
 474                 cache->cached = BTRFS_CACHE_STARTED;
 475                 spin_unlock(&cache->lock);
 476
 477                 ret = load_free_space_cache(fs_info, cache);
 478
 479                 spin_lock(&cache->lock);
 480                 if (ret == 1) {
 481                         cache->cached = BTRFS_CACHE_FINISHED;
 482                         cache->last_byte_to_unpin = (u64)-1;
 483                 } else {
 484                         cache->cached = BTRFS_CACHE_NO;
 485                 }
 486                 spin_unlock(&cache->lock);
 487                 if (ret == 1) {
 488                         free_excluded_extents(fs_info->extent_root, cache);
 489                         return 0;
 490                 }
 491         }
 492
 493         if (load_cache_only)
 494                 return 0;
 495
 496         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 497         BUG_ON(!caching_ctl);
 498
 499         INIT_LIST_HEAD(&caching_ctl->list);
 500         mutex_init(&caching_ctl->mutex);
 501         init_waitqueue_head(&caching_ctl->wait);
 502         caching_ctl->block_group = cache;
 503         caching_ctl->progress = cache->key.objectid;
 504         /* one for caching kthread, one for caching block group list */
 505         atomic_set(&caching_ctl->count, 2);
 506         caching_ctl->work.func = caching_thread;
 507
 508         spin_lock(&cache->lock);
 509         if (cache->cached != BTRFS_CACHE_NO) {
 510                 spin_unlock(&cache->lock);
 511                 kfree(caching_ctl);
 512                 return 0;
 513         }
 514         cache->caching_ctl = caching_ctl;
 515         cache->cached = BTRFS_CACHE_STARTED;
 516         spin_unlock(&cache->lock);
 517
 518         down_write(&fs_info->extent_commit_sem);
 519         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 520         up_write(&fs_info->extent_commit_sem);
 521
 522         btrfs_get_block_group(cache);
 523
 524         btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
 525
 526         return ret;
 527 }
 528
 529 /*
 530  * return the block group that starts at or after bytenr
 531  */
 532 static struct btrfs_block_group_cache *
 533 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 534 {
 535         struct btrfs_block_group_cache *cache;
 536
 537         cache = block_group_cache_tree_search(info, bytenr, 0);
 538
 539         return cache;
 540 }
 541
 542 /*
 543  * return the block group that contains the given bytenr
 544  */
 545 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 546                                                  struct btrfs_fs_info *info,
 547                                                  u64 bytenr)
 548 {
 549         struct btrfs_block_group_cache *cache;
 550
 551         cache = block_group_cache_tree_search(info, bytenr, 1);
 552
 553         return cache;
 554 }
 555
 556 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 557                                                   u64 flags)
 558 {
 559         struct list_head *head = &info->space_info;
 560         struct btrfs_space_info *found;
 561
 562         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 563                  BTRFS_BLOCK_GROUP_METADATA;
 564
 565         rcu_read_lock();
 566         list_for_each_entry_rcu(found, head, list) {
 567                 if (found->flags & flags) {
 568                         rcu_read_unlock();
 569                         return found;
 570                 }
 571         }
 572         rcu_read_unlock();
 573         return NULL;
 574 }
 575
 576 /*
 577  * after adding space to the filesystem, we need to clear the full flags
 578  * on all the space infos.
 579  */
 580 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 581 {
 582         struct list_head *head = &info->space_info;
 583         struct btrfs_space_info *found;
 584
 585         rcu_read_lock();
 586         list_for_each_entry_rcu(found, head, list)
 587                 found->full = 0;
 588         rcu_read_unlock();
 589 }
 590
 591 static u64 div_factor(u64 num, int factor)
 592 {
 593         if (factor == 10)
 594                 return num;
 595         num *= factor;
 596         do_div(num, 10);
 597         return num;
 598 }
 599
 600 static u64 div_factor_fine(u64 num, int factor)
 601 {
 602         if (factor == 100)
 603                 return num;
 604         num *= factor;
 605         do_div(num, 100);
 606         return num;
 607 }
 608
 609 u64 btrfs_find_block_group(struct btrfs_root *root,
 610                            u64 search_start, u64 search_hint, int owner)
 611 {
 612         struct btrfs_block_group_cache *cache;
 613         u64 used;
 614         u64 last = max(search_hint, search_start);
 615         u64 group_start = 0;
 616         int full_search = 0;
 617         int factor = 9;
 618         int wrapped = 0;
 619 again:
 620         while (1) {
 621                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 622                 if (!cache)
 623                         break;
 624
 625                 spin_lock(&cache->lock);
 626                 last = cache->key.objectid + cache->key.offset;
 627                 used = btrfs_block_group_used(&cache->item);
 628
 629                 if ((full_search || !cache->ro) &&
 630                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 631                         if (used + cache->pinned + cache->reserved <
 632                             div_factor(cache->key.offset, factor)) {
 633                                 group_start = cache->key.objectid;
 634                                 spin_unlock(&cache->lock);
 635                                 btrfs_put_block_group(cache);
 636                                 goto found;
 637                         }
 638                 }
 639                 spin_unlock(&cache->lock);
 640                 btrfs_put_block_group(cache);
 641                 cond_resched();
 642         }
 643         if (!wrapped) {
 644                 last = search_start;
 645                 wrapped = 1;
 646                 goto again;
 647         }
 648         if (!full_search && factor < 10) {
 649                 last = search_start;
 650                 full_search = 1;
 651                 factor = 10;
 652                 goto again;
 653         }
 654 found:
 655         return group_start;
 656 }
 657
 658 /* simple helper to search for an existing extent at a given offset */
 659 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 660 {
 661         int ret;
 662         struct btrfs_key key;
 663         struct btrfs_path *path;
 664
 665         path = btrfs_alloc_path();
 666         BUG_ON(!path);
 667         key.objectid = start;
 668         key.offset = len;
 669         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 670         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 671                                 0, 0);
 672         btrfs_free_path(path);
 673         return ret;
 674 }
 675
 676 /*
 677  * helper function to lookup reference count and flags of extent.
 678  *
 679  * the head node for delayed ref is used to store the sum of all the
 680  * reference count modifications queued up in the rbtree. the head
 681  * node may also store the extent flags to set. This way you can check
 682  * to see what the reference count and extent flags would be if all of
 683  * the delayed refs are not processed.
 684  */
 685 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 686                              struct btrfs_root *root, u64 bytenr,
 687                              u64 num_bytes, u64 *refs, u64 *flags)
 688 {
 689         struct btrfs_delayed_ref_head *head;
 690         struct btrfs_delayed_ref_root *delayed_refs;
 691         struct btrfs_path *path;
 692         struct btrfs_extent_item *ei;
 693         struct extent_buffer *leaf;
 694         struct btrfs_key key;
 695         u32 item_size;
 696         u64 num_refs;
 697         u64 extent_flags;
 698         int ret;
 699
 700         path = btrfs_alloc_path();
 701         if (!path)
 702                 return -ENOMEM;
 703
 704         key.objectid = bytenr;
 705         key.type = BTRFS_EXTENT_ITEM_KEY;
 706         key.offset = num_bytes;
 707         if (!trans) {
 708                 path->skip_locking = 1;
 709                 path->search_commit_root = 1;
 710         }
 711 again:
 712         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 713                                 &key, path, 0, 0);
 714         if (ret < 0)
 715                 goto out_free;
 716
 717         if (ret == 0) {
 718                 leaf = path->nodes[0];
 719                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 720                 if (item_size >= sizeof(*ei)) {
 721                         ei = btrfs_item_ptr(leaf, path->slots[0],
 722                                             struct btrfs_extent_item);
 723                         num_refs = btrfs_extent_refs(leaf, ei);
 724                         extent_flags = btrfs_extent_flags(leaf, ei);
 725                 } else {
 726 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 727                         struct btrfs_extent_item_v0 *ei0;
 728                         BUG_ON(item_size != sizeof(*ei0));
 729                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 730                                              struct btrfs_extent_item_v0);
 731                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 732                         /* FIXME: this isn't correct for data */
 733                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 734 #else
 735                         BUG();
 736 #endif
 737                 }
 738                 BUG_ON(num_refs == 0);
 739         } else {
 740                 num_refs = 0;
 741                 extent_flags = 0;
 742                 ret = 0;
 743         }
 744
 745         if (!trans)
 746                 goto out;
 747
 748         delayed_refs = &trans->transaction->delayed_refs;
 749         spin_lock(&delayed_refs->lock);
 750         head = btrfs_find_delayed_ref_head(trans, bytenr);
 751         if (head) {
 752                 if (!mutex_trylock(&head->mutex)) {
 753                         atomic_inc(&head->node.refs);
 754                         spin_unlock(&delayed_refs->lock);
 755
 756                         btrfs_release_path(path);
 757
 758                         /*
 759                          * Mutex was contended, block until it's released and try
 760                          * again
 761                          */
 762                         mutex_lock(&head->mutex);
 763                         mutex_unlock(&head->mutex);
 764                         btrfs_put_delayed_ref(&head->node);
 765                         goto again;
 766                 }
 767                 if (head->extent_op && head->extent_op->update_flags)
 768                         extent_flags |= head->extent_op->flags_to_set;
 769                 else
 770                         BUG_ON(num_refs == 0);
 771
 772                 num_refs += head->node.ref_mod;
 773                 mutex_unlock(&head->mutex);
 774         }
 775         spin_unlock(&delayed_refs->lock);
 776 out:
 777         WARN_ON(num_refs == 0);
 778         if (refs)
 779                 *refs = num_refs;
 780         if (flags)
 781                 *flags = extent_flags;
 782 out_free:
 783         btrfs_free_path(path);
 784         return ret;
 785 }
 786
 787 /*
 788  * Back reference rules.  Back refs have three main goals:
 789  *
 790  * 1) differentiate between all holders of references to an extent so that
 791  *    when a reference is dropped we can make sure it was a valid reference
 792  *    before freeing the extent.
 793  *
 794  * 2) Provide enough information to quickly find the holders of an extent
 795  *    if we notice a given block is corrupted or bad.
 796  *
 797  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 798  *    maintenance.  This is actually the same as #2, but with a slightly
 799  *    different use case.
 800  *
 801  * There are two kinds of back refs. The implicit back refs is optimized
 802  * for pointers in non-shared tree blocks. For a given pointer in a block,
 803  * back refs of this kind provide information about the block's owner tree
 804  * and the pointer's key. These information allow us to find the block by
 805  * b-tree searching. The full back refs is for pointers in tree blocks not
 806  * referenced by their owner trees. The location of tree block is recorded
 807  * in the back refs. Actually the full back refs is generic, and can be
 808  * used in all cases the implicit back refs is used. The major shortcoming
 809  * of the full back refs is its overhead. Every time a tree block gets
 810  * COWed, we have to update back refs entry for all pointers in it.
 811  *
 812  * For a newly allocated tree block, we use implicit back refs for
 813  * pointers in it. This means most tree related operations only involve
 814  * implicit back refs. For a tree block created in old transaction, the
 815  * only way to drop a reference to it is COW it. So we can detect the
 816  * event that tree block loses its owner tree's reference and do the
 817  * back refs conversion.
 818  *
 819  * When a tree block is COW'd through a tree, there are four cases:
 820  *
 821  * The reference count of the block is one and the tree is the block's
 822  * owner tree. Nothing to do in this case.
 823  *
 824  * The reference count of the block is one and the tree is not the
 825  * block's owner tree. In this case, full back refs is used for pointers
 826  * in the block. Remove these full back refs, add implicit back refs for
 827  * every pointers in the new block.
 828  *
 829  * The reference count of the block is greater than one and the tree is
 830  * the block's owner tree. In this case, implicit back refs is used for
 831  * pointers in the block. Add full back refs for every pointers in the
 832  * block, increase lower level extents' reference counts. The original
 833  * implicit back refs are entailed to the new block.
 834  *
 835  * The reference count of the block is greater than one and the tree is
 836  * not the block's owner tree. Add implicit back refs for every pointer in
 837  * the new block, increase lower level extents' reference count.
 838  *
 839  * Back Reference Key composing:
 840  *
 841  * The key objectid corresponds to the first byte in the extent,
 842  * The key type is used to differentiate between types of back refs.
 843  * There are different meanings of the key offset for different types
 844  * of back refs.
 845  *
 846  * File extents can be referenced by:
 847  *
 848  * - multiple snapshots, subvolumes, or different generations in one subvol
 849  * - different files inside a single subvolume
 850  * - different offsets inside a file (bookend extents in file.c)
 851  *
 852  * The extent ref structure for the implicit back refs has fields for:
 853  *
 854  * - Objectid of the subvolume root
 855  * - objectid of the file holding the reference
 856  * - original offset in the file
 857  * - how many bookend extents
 858  *
 859  * The key offset for the implicit back refs is hash of the first
 860  * three fields.
 861  *
 862  * The extent ref structure for the full back refs has field for:
 863  *
 864  * - number of pointers in the tree leaf
 865  *
 866  * The key offset for the implicit back refs is the first byte of
 867  * the tree leaf
 868  *
 869  * When a file extent is allocated, The implicit back refs is used.
 870  * the fields are filled in:
 871  *
 872  *     (root_key.objectid, inode objectid, offset in file, 1)
 873  *
 874  * When a file extent is removed file truncation, we find the
 875  * corresponding implicit back refs and check the following fields:
 876  *
 877  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 878  *
 879  * Btree extents can be referenced by:
 880  *
 881  * - Different subvolumes
 882  *
 883  * Both the implicit back refs and the full back refs for tree blocks
 884  * only consist of key. The key offset for the implicit back refs is
 885  * objectid of block's owner tree. The key offset for the full back refs
 886  * is the first byte of parent block.
 887  *
 888  * When implicit back refs is used, information about the lowest key and
 889  * level of the tree block are required. These information are stored in
 890  * tree block info structure.
 891  */
 892
 893 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 894 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 895                                   struct btrfs_root *root,
 896                                   struct btrfs_path *path,
 897                                   u64 owner, u32 extra_size)
 898 {
 899         struct btrfs_extent_item *item;
 900         struct btrfs_extent_item_v0 *ei0;
 901         struct btrfs_extent_ref_v0 *ref0;
 902         struct btrfs_tree_block_info *bi;
 903         struct extent_buffer *leaf;
 904         struct btrfs_key key;
 905         struct btrfs_key found_key;
 906         u32 new_size = sizeof(*item);
 907         u64 refs;
 908         int ret;
 909
 910         leaf = path->nodes[0];
 911         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 912
 913         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 914         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 915                              struct btrfs_extent_item_v0);
 916         refs = btrfs_extent_refs_v0(leaf, ei0);
 917
 918         if (owner == (u64)-1) {
 919                 while (1) {
 920                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 921                                 ret = btrfs_next_leaf(root, path);
 922                                 if (ret < 0)
 923                                         return ret;
 924                                 BUG_ON(ret > 0);
 925                                 leaf = path->nodes[0];
 926                         }
 927                         btrfs_item_key_to_cpu(leaf, &found_key,
 928                                               path->slots[0]);
 929                         BUG_ON(key.objectid != found_key.objectid);
 930                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 931                                 path->slots[0]++;
 932                                 continue;
 933                         }
 934                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 935                                               struct btrfs_extent_ref_v0);
 936                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 937                         break;
 938                 }
 939         }
 940         btrfs_release_path(path);
 941
 942         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 943                 new_size += sizeof(*bi);
 944
 945         new_size -= sizeof(*ei0);
 946         ret = btrfs_search_slot(trans, root, &key, path,
 947                                 new_size + extra_size, 1);
 948         if (ret < 0)
 949                 return ret;
 950         BUG_ON(ret);
 951
 952         ret = btrfs_extend_item(trans, root, path, new_size);
 953
 954         leaf = path->nodes[0];
 955         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 956         btrfs_set_extent_refs(leaf, item, refs);
 957         /* FIXME: get real generation */
 958         btrfs_set_extent_generation(leaf, item, 0);
 959         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 960                 btrfs_set_extent_flags(leaf, item,
 961                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 962                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 963                 bi = (struct btrfs_tree_block_info *)(item + 1);
 964                 /* FIXME: get first key of the block */
 965                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 966                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 967         } else {
 968                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 969         }
 970         btrfs_mark_buffer_dirty(leaf);
 971         return 0;
 972 }
 973 #endif
 974
 975 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 976 {
 977         u32 high_crc = ~(u32)0;
 978         u32 low_crc = ~(u32)0;
 979         __le64 lenum;
 980
 981         lenum = cpu_to_le64(root_objectid);
 982         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 983         lenum = cpu_to_le64(owner);
 984         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 985         lenum = cpu_to_le64(offset);
 986         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 987
 988         return ((u64)high_crc << 31) ^ (u64)low_crc;
 989 }
 990
 991 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 992                                      struct btrfs_extent_data_ref *ref)
 993 {
 994         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 995                                     btrfs_extent_data_ref_objectid(leaf, ref),
 996                                     btrfs_extent_data_ref_offset(leaf, ref));
 997 }
 998
 999 static int match_extent_data_ref(struct extent_buffer *leaf,
1000                                  struct btrfs_extent_data_ref *ref,
1001                                  u64 root_objectid, u64 owner, u64 offset)
1002 {
1003         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1004             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1005             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1006                 return 0;
1007         return 1;
1008 }
1009
1010 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1011                                            struct btrfs_root *root,
1012                                            struct btrfs_path *path,
1013                                            u64 bytenr, u64 parent,
1014                                            u64 root_objectid,
1015                                            u64 owner, u64 offset)
1016 {
1017         struct btrfs_key key;
1018         struct btrfs_extent_data_ref *ref;
1019         struct extent_buffer *leaf;
1020         u32 nritems;
1021         int ret;
1022         int recow;
1023         int err = -ENOENT;
1024
1025         key.objectid = bytenr;
1026         if (parent) {
1027                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1028                 key.offset = parent;
1029         } else {
1030                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1031                 key.offset = hash_extent_data_ref(root_objectid,
1032                                                   owner, offset);
1033         }
1034 again:
1035         recow = 0;
1036         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1037         if (ret < 0) {
1038                 err = ret;
1039                 goto fail;
1040         }
1041
1042         if (parent) {
1043                 if (!ret)
1044                         return 0;
1045 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1046                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1047                 btrfs_release_path(path);
1048                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1049                 if (ret < 0) {
1050                         err = ret;
1051                         goto fail;
1052                 }
1053                 if (!ret)
1054                         return 0;
1055 #endif
1056                 goto fail;
1057         }
1058
1059         leaf = path->nodes[0];
1060         nritems = btrfs_header_nritems(leaf);
1061         while (1) {
1062                 if (path->slots[0] >= nritems) {
1063                         ret = btrfs_next_leaf(root, path);
1064                         if (ret < 0)
1065                                 err = ret;
1066                         if (ret)
1067                                 goto fail;
1068
1069                         leaf = path->nodes[0];
1070                         nritems = btrfs_header_nritems(leaf);
1071                         recow = 1;
1072                 }
1073
1074                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1075                 if (key.objectid != bytenr ||
1076                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1077                         goto fail;
1078
1079                 ref = btrfs_item_ptr(leaf, path->slots[0],
1080                                      struct btrfs_extent_data_ref);
1081
1082                 if (match_extent_data_ref(leaf, ref, root_objectid,
1083                                           owner, offset)) {
1084                         if (recow) {
1085                                 btrfs_release_path(path);
1086                                 goto again;
1087                         }
1088                         err = 0;
1089                         break;
1090                 }
1091                 path->slots[0]++;
1092         }
1093 fail:
1094         return err;
1095 }
1096
1097 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1098                                            struct btrfs_root *root,
1099                                            struct btrfs_path *path,
1100                                            u64 bytenr, u64 parent,
1101                                            u64 root_objectid, u64 owner,
1102                                            u64 offset, int refs_to_add)
1103 {
1104         struct btrfs_key key;
1105         struct extent_buffer *leaf;
1106         u32 size;
1107         u32 num_refs;
1108         int ret;
1109
1110         key.objectid = bytenr;
1111         if (parent) {
1112                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1113                 key.offset = parent;
1114                 size = sizeof(struct btrfs_shared_data_ref);
1115         } else {
1116                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1117                 key.offset = hash_extent_data_ref(root_objectid,
1118                                                   owner, offset);
1119                 size = sizeof(struct btrfs_extent_data_ref);
1120         }
1121
1122         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1123         if (ret && ret != -EEXIST)
1124                 goto fail;
1125
1126         leaf = path->nodes[0];
1127         if (parent) {
1128                 struct btrfs_shared_data_ref *ref;
1129                 ref = btrfs_item_ptr(leaf, path->slots[0],
1130                                      struct btrfs_shared_data_ref);
1131                 if (ret == 0) {
1132                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1133                 } else {
1134                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1135                         num_refs += refs_to_add;
1136                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1137                 }
1138         } else {
1139                 struct btrfs_extent_data_ref *ref;
1140                 while (ret == -EEXIST) {
1141                         ref = btrfs_item_ptr(leaf, path->slots[0],
1142                                              struct btrfs_extent_data_ref);
1143                         if (match_extent_data_ref(leaf, ref, root_objectid,
1144                                                   owner, offset))
1145                                 break;
1146                         btrfs_release_path(path);
1147                         key.offset++;
1148                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1149                                                       size);
1150                         if (ret && ret != -EEXIST)
1151                                 goto fail;
1152
1153                         leaf = path->nodes[0];
1154                 }
1155                 ref = btrfs_item_ptr(leaf, path->slots[0],
1156                                      struct btrfs_extent_data_ref);
1157                 if (ret == 0) {
1158                         btrfs_set_extent_data_ref_root(leaf, ref,
1159                                                        root_objectid);
1160                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1161                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1162                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1163                 } else {
1164                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1165                         num_refs += refs_to_add;
1166                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1167                 }
1168         }
1169         btrfs_mark_buffer_dirty(leaf);
1170         ret = 0;
1171 fail:
1172         btrfs_release_path(path);
1173         return ret;
1174 }
1175
1176 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1177                                            struct btrfs_root *root,
1178                                            struct btrfs_path *path,
1179                                            int refs_to_drop)
1180 {
1181         struct btrfs_key key;
1182         struct btrfs_extent_data_ref *ref1 = NULL;
1183         struct btrfs_shared_data_ref *ref2 = NULL;
1184         struct extent_buffer *leaf;
1185         u32 num_refs = 0;
1186         int ret = 0;
1187
1188         leaf = path->nodes[0];
1189         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1190
1191         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1192                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1193                                       struct btrfs_extent_data_ref);
1194                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1195         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1196                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1197                                       struct btrfs_shared_data_ref);
1198                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1199 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1200         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1201                 struct btrfs_extent_ref_v0 *ref0;
1202                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1203                                       struct btrfs_extent_ref_v0);
1204                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1205 #endif
1206         } else {
1207                 BUG();
1208         }
1209
1210         BUG_ON(num_refs < refs_to_drop);
1211         num_refs -= refs_to_drop;
1212
1213         if (num_refs == 0) {
1214                 ret = btrfs_del_item(trans, root, path);
1215         } else {
1216                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1217                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1218                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1219                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1220 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1221                 else {
1222                         struct btrfs_extent_ref_v0 *ref0;
1223                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1224                                         struct btrfs_extent_ref_v0);
1225                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1226                 }
1227 #endif
1228                 btrfs_mark_buffer_dirty(leaf);
1229         }
1230         return ret;
1231 }
1232
1233 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1234                                           struct btrfs_path *path,
1235                                           struct btrfs_extent_inline_ref *iref)
1236 {
1237         struct btrfs_key key;
1238         struct extent_buffer *leaf;
1239         struct btrfs_extent_data_ref *ref1;
1240         struct btrfs_shared_data_ref *ref2;
1241         u32 num_refs = 0;
1242
1243         leaf = path->nodes[0];
1244         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1245         if (iref) {
1246                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1247                     BTRFS_EXTENT_DATA_REF_KEY) {
1248                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1249                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1250                 } else {
1251                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1252                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1253                 }
1254         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1255                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1256                                       struct btrfs_extent_data_ref);
1257                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1258         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1259                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1260                                       struct btrfs_shared_data_ref);
1261                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1262 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1263         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1264                 struct btrfs_extent_ref_v0 *ref0;
1265                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1266                                       struct btrfs_extent_ref_v0);
1267                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1268 #endif
1269         } else {
1270                 WARN_ON(1);
1271         }
1272         return num_refs;
1273 }
1274
1275 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1276                                           struct btrfs_root *root,
1277                                           struct btrfs_path *path,
1278                                           u64 bytenr, u64 parent,
1279                                           u64 root_objectid)
1280 {
1281         struct btrfs_key key;
1282         int ret;
1283
1284         key.objectid = bytenr;
1285         if (parent) {
1286                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1287                 key.offset = parent;
1288         } else {
1289                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1290                 key.offset = root_objectid;
1291         }
1292
1293         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1294         if (ret > 0)
1295                 ret = -ENOENT;
1296 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1297         if (ret == -ENOENT && parent) {
1298                 btrfs_release_path(path);
1299                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1300                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1301                 if (ret > 0)
1302                         ret = -ENOENT;
1303         }
1304 #endif
1305         return ret;
1306 }
1307
1308 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1309                                           struct btrfs_root *root,
1310                                           struct btrfs_path *path,
1311                                           u64 bytenr, u64 parent,
1312                                           u64 root_objectid)
1313 {
1314         struct btrfs_key key;
1315         int ret;
1316
1317         key.objectid = bytenr;
1318         if (parent) {
1319                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1320                 key.offset = parent;
1321         } else {
1322                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1323                 key.offset = root_objectid;
1324         }
1325
1326         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1327         btrfs_release_path(path);
1328         return ret;
1329 }
1330
1331 static inline int extent_ref_type(u64 parent, u64 owner)
1332 {
1333         int type;
1334         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1335                 if (parent > 0)
1336                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1337                 else
1338                         type = BTRFS_TREE_BLOCK_REF_KEY;
1339         } else {
1340                 if (parent > 0)
1341                         type = BTRFS_SHARED_DATA_REF_KEY;
1342                 else
1343                         type = BTRFS_EXTENT_DATA_REF_KEY;
1344         }
1345         return type;
1346 }
1347
1348 static int find_next_key(struct btrfs_path *path, int level,
1349                          struct btrfs_key *key)
1350
1351 {
1352         for (; level < BTRFS_MAX_LEVEL; level++) {
1353                 if (!path->nodes[level])
1354                         break;
1355                 if (path->slots[level] + 1 >=
1356                     btrfs_header_nritems(path->nodes[level]))
1357                         continue;
1358                 if (level == 0)
1359                         btrfs_item_key_to_cpu(path->nodes[level], key,
1360                                               path->slots[level] + 1);
1361                 else
1362                         btrfs_node_key_to_cpu(path->nodes[level], key,
1363                                               path->slots[level] + 1);
1364                 return 0;
1365         }
1366         return 1;
1367 }
1368
1369 /*
1370  * look for inline back ref. if back ref is found, *ref_ret is set
1371  * to the address of inline back ref, and 0 is returned.
1372  *
1373  * if back ref isn't found, *ref_ret is set to the address where it
1374  * should be inserted, and -ENOENT is returned.
1375  *
1376  * if insert is true and there are too many inline back refs, the path
1377  * points to the extent item, and -EAGAIN is returned.
1378  *
1379  * NOTE: inline back refs are ordered in the same way that back ref
1380  *       items in the tree are ordered.
1381  */
1382 static noinline_for_stack
1383 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1384                                  struct btrfs_root *root,
1385                                  struct btrfs_path *path,
1386                                  struct btrfs_extent_inline_ref **ref_ret,
1387                                  u64 bytenr, u64 num_bytes,
1388                                  u64 parent, u64 root_objectid,
1389                                  u64 owner, u64 offset, int insert)
1390 {
1391         struct btrfs_key key;
1392         struct extent_buffer *leaf;
1393         struct btrfs_extent_item *ei;
1394         struct btrfs_extent_inline_ref *iref;
1395         u64 flags;
1396         u64 item_size;
1397         unsigned long ptr;
1398         unsigned long end;
1399         int extra_size;
1400         int type;
1401         int want;
1402         int ret;
1403         int err = 0;
1404
1405         key.objectid = bytenr;
1406         key.type = BTRFS_EXTENT_ITEM_KEY;
1407         key.offset = num_bytes;
1408
1409         want = extent_ref_type(parent, owner);
1410         if (insert) {
1411                 extra_size = btrfs_extent_inline_ref_size(want);
1412                 path->keep_locks = 1;
1413         } else
1414                 extra_size = -1;
1415         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1416         if (ret < 0) {
1417                 err = ret;
1418                 goto out;
1419         }
1420         BUG_ON(ret);
1421
1422         leaf = path->nodes[0];
1423         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1424 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1425         if (item_size < sizeof(*ei)) {
1426                 if (!insert) {
1427                         err = -ENOENT;
1428                         goto out;
1429                 }
1430                 ret = convert_extent_item_v0(trans, root, path, owner,
1431                                              extra_size);
1432                 if (ret < 0) {
1433                         err = ret;
1434                         goto out;
1435                 }
1436                 leaf = path->nodes[0];
1437                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1438         }
1439 #endif
1440         BUG_ON(item_size < sizeof(*ei));
1441
1442         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1443         flags = btrfs_extent_flags(leaf, ei);
1444
1445         ptr = (unsigned long)(ei + 1);
1446         end = (unsigned long)ei + item_size;
1447
1448         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1449                 ptr += sizeof(struct btrfs_tree_block_info);
1450                 BUG_ON(ptr > end);
1451         } else {
1452                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1453         }
1454
1455         err = -ENOENT;
1456         while (1) {
1457                 if (ptr >= end) {
1458                         WARN_ON(ptr > end);
1459                         break;
1460                 }
1461                 iref = (struct btrfs_extent_inline_ref *)ptr;
1462                 type = btrfs_extent_inline_ref_type(leaf, iref);
1463                 if (want < type)
1464                         break;
1465                 if (want > type) {
1466                         ptr += btrfs_extent_inline_ref_size(type);
1467                         continue;
1468                 }
1469
1470                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1471                         struct btrfs_extent_data_ref *dref;
1472                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1473                         if (match_extent_data_ref(leaf, dref, root_objectid,
1474                                                   owner, offset)) {
1475                                 err = 0;
1476                                 break;
1477                         }
1478                         if (hash_extent_data_ref_item(leaf, dref) <
1479                             hash_extent_data_ref(root_objectid, owner, offset))
1480                                 break;
1481                 } else {
1482                         u64 ref_offset;
1483                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1484                         if (parent > 0) {
1485                                 if (parent == ref_offset) {
1486                                         err = 0;
1487                                         break;
1488                                 }
1489                                 if (ref_offset < parent)
1490                                         break;
1491                         } else {
1492                                 if (root_objectid == ref_offset) {
1493                                         err = 0;
1494                                         break;
1495                                 }
1496                                 if (ref_offset < root_objectid)
1497                                         break;
1498                         }
1499                 }
1500                 ptr += btrfs_extent_inline_ref_size(type);
1501         }
1502         if (err == -ENOENT && insert) {
1503                 if (item_size + extra_size >=
1504                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1505                         err = -EAGAIN;
1506                         goto out;
1507                 }
1508                 /*
1509                  * To add new inline back ref, we have to make sure
1510                  * there is no corresponding back ref item.
1511                  * For simplicity, we just do not add new inline back
1512                  * ref if there is any kind of item for this block
1513                  */
1514                 if (find_next_key(path, 0, &key) == 0 &&
1515                     key.objectid == bytenr &&
1516                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1517                         err = -EAGAIN;
1518                         goto out;
1519                 }
1520         }
1521         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1522 out:
1523         if (insert) {
1524                 path->keep_locks = 0;
1525                 btrfs_unlock_up_safe(path, 1);
1526         }
1527         return err;
1528 }
1529
1530 /*
1531  * helper to add new inline back ref
1532  */
1533 static noinline_for_stack
1534 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1535                                 struct btrfs_root *root,
1536                                 struct btrfs_path *path,
1537                                 struct btrfs_extent_inline_ref *iref,
1538                                 u64 parent, u64 root_objectid,
1539                                 u64 owner, u64 offset, int refs_to_add,
1540                                 struct btrfs_delayed_extent_op *extent_op)
1541 {
1542         struct extent_buffer *leaf;
1543         struct btrfs_extent_item *ei;
1544         unsigned long ptr;
1545         unsigned long end;
1546         unsigned long item_offset;
1547         u64 refs;
1548         int size;
1549         int type;
1550         int ret;
1551
1552         leaf = path->nodes[0];
1553         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1554         item_offset = (unsigned long)iref - (unsigned long)ei;
1555
1556         type = extent_ref_type(parent, owner);
1557         size = btrfs_extent_inline_ref_size(type);
1558
1559         ret = btrfs_extend_item(trans, root, path, size);
1560
1561         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1562         refs = btrfs_extent_refs(leaf, ei);
1563         refs += refs_to_add;
1564         btrfs_set_extent_refs(leaf, ei, refs);
1565         if (extent_op)
1566                 __run_delayed_extent_op(extent_op, leaf, ei);
1567
1568         ptr = (unsigned long)ei + item_offset;
1569         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1570         if (ptr < end - size)
1571                 memmove_extent_buffer(leaf, ptr + size, ptr,
1572                                       end - size - ptr);
1573
1574         iref = (struct btrfs_extent_inline_ref *)ptr;
1575         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1576         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1577                 struct btrfs_extent_data_ref *dref;
1578                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1579                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1580                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1581                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1582                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1583         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1584                 struct btrfs_shared_data_ref *sref;
1585                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1586                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1587                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1588         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1589                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1590         } else {
1591                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1592         }
1593         btrfs_mark_buffer_dirty(leaf);
1594         return 0;
1595 }
1596
1597 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1598                                  struct btrfs_root *root,
1599                                  struct btrfs_path *path,
1600                                  struct btrfs_extent_inline_ref **ref_ret,
1601                                  u64 bytenr, u64 num_bytes, u64 parent,
1602                                  u64 root_objectid, u64 owner, u64 offset)
1603 {
1604         int ret;
1605
1606         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1607                                            bytenr, num_bytes, parent,
1608                                            root_objectid, owner, offset, 0);
1609         if (ret != -ENOENT)
1610                 return ret;
1611
1612         btrfs_release_path(path);
1613         *ref_ret = NULL;
1614
1615         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1616                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1617                                             root_objectid);
1618         } else {
1619                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1620                                              root_objectid, owner, offset);
1621         }
1622         return ret;
1623 }
1624
1625 /*
1626  * helper to update/remove inline back ref
1627  */
1628 static noinline_for_stack
1629 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1630                                  struct btrfs_root *root,
1631                                  struct btrfs_path *path,
1632                                  struct btrfs_extent_inline_ref *iref,
1633                                  int refs_to_mod,
1634                                  struct btrfs_delayed_extent_op *extent_op)
1635 {
1636         struct extent_buffer *leaf;
1637         struct btrfs_extent_item *ei;
1638         struct btrfs_extent_data_ref *dref = NULL;
1639         struct btrfs_shared_data_ref *sref = NULL;
1640         unsigned long ptr;
1641         unsigned long end;
1642         u32 item_size;
1643         int size;
1644         int type;
1645         int ret;
1646         u64 refs;
1647
1648         leaf = path->nodes[0];
1649         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1650         refs = btrfs_extent_refs(leaf, ei);
1651         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1652         refs += refs_to_mod;
1653         btrfs_set_extent_refs(leaf, ei, refs);
1654         if (extent_op)
1655                 __run_delayed_extent_op(extent_op, leaf, ei);
1656
1657         type = btrfs_extent_inline_ref_type(leaf, iref);
1658
1659         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1660                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1661                 refs = btrfs_extent_data_ref_count(leaf, dref);
1662         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1663                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1664                 refs = btrfs_shared_data_ref_count(leaf, sref);
1665         } else {
1666                 refs = 1;
1667                 BUG_ON(refs_to_mod != -1);
1668         }
1669
1670         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1671         refs += refs_to_mod;
1672
1673         if (refs > 0) {
1674                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1675                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1676                 else
1677                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1678         } else {
1679                 size =  btrfs_extent_inline_ref_size(type);
1680                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1681                 ptr = (unsigned long)iref;
1682                 end = (unsigned long)ei + item_size;
1683                 if (ptr + size < end)
1684                         memmove_extent_buffer(leaf, ptr, ptr + size,
1685                                               end - ptr - size);
1686                 item_size -= size;
1687                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1688         }
1689         btrfs_mark_buffer_dirty(leaf);
1690         return 0;
1691 }
1692
1693 static noinline_for_stack
1694 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1695                                  struct btrfs_root *root,
1696                                  struct btrfs_path *path,
1697                                  u64 bytenr, u64 num_bytes, u64 parent,
1698                                  u64 root_objectid, u64 owner,
1699                                  u64 offset, int refs_to_add,
1700                                  struct btrfs_delayed_extent_op *extent_op)
1701 {
1702         struct btrfs_extent_inline_ref *iref;
1703         int ret;
1704
1705         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1706                                            bytenr, num_bytes, parent,
1707                                            root_objectid, owner, offset, 1);
1708         if (ret == 0) {
1709                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1710                 ret = update_inline_extent_backref(trans, root, path, iref,
1711                                                    refs_to_add, extent_op);
1712         } else if (ret == -ENOENT) {
1713                 ret = setup_inline_extent_backref(trans, root, path, iref,
1714                                                   parent, root_objectid,
1715                                                   owner, offset, refs_to_add,
1716                                                   extent_op);
1717         }
1718         return ret;
1719 }
1720
1721 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1722                                  struct btrfs_root *root,
1723                                  struct btrfs_path *path,
1724                                  u64 bytenr, u64 parent, u64 root_objectid,
1725                                  u64 owner, u64 offset, int refs_to_add)
1726 {
1727         int ret;
1728         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1729                 BUG_ON(refs_to_add != 1);
1730                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1731                                             parent, root_objectid);
1732         } else {
1733                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1734                                              parent, root_objectid,
1735                                              owner, offset, refs_to_add);
1736         }
1737         return ret;
1738 }
1739
1740 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1741                                  struct btrfs_root *root,
1742                                  struct btrfs_path *path,
1743                                  struct btrfs_extent_inline_ref *iref,
1744                                  int refs_to_drop, int is_data)
1745 {
1746         int ret;
1747
1748         BUG_ON(!is_data && refs_to_drop != 1);
1749         if (iref) {
1750                 ret = update_inline_extent_backref(trans, root, path, iref,
1751                                                    -refs_to_drop, NULL);
1752         } else if (is_data) {
1753                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1754         } else {
1755                 ret = btrfs_del_item(trans, root, path);
1756         }
1757         return ret;
1758 }
1759
1760 static int btrfs_issue_discard(struct block_device *bdev,
1761                                 u64 start, u64 len)
1762 {
1763         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1764 }
1765
1766 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1767                                 u64 num_bytes, u64 *actual_bytes)
1768 {
1769         int ret;
1770         u64 discarded_bytes = 0;
1771         struct btrfs_multi_bio *multi = NULL;
1772
1773
1774         /* Tell the block device(s) that the sectors can be discarded */
1775         ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1776                               bytenr, &num_bytes, &multi, 0);
1777         if (!ret) {
1778                 struct btrfs_bio_stripe *stripe = multi->stripes;
1779                 int i;
1780
1781
1782                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1783                         ret = btrfs_issue_discard(stripe->dev->bdev,
1784                                                   stripe->physical,
1785                                                   stripe->length);
1786                         if (!ret)
1787                                 discarded_bytes += stripe->length;
1788                         else if (ret != -EOPNOTSUPP)
1789                                 break;
1790                 }
1791                 kfree(multi);
1792         }
1793         if (discarded_bytes && ret == -EOPNOTSUPP)
1794                 ret = 0;
1795
1796         if (actual_bytes)
1797                 *actual_bytes = discarded_bytes;
1798
1799
1800         return ret;
1801 }
1802
1803 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1804                          struct btrfs_root *root,
1805                          u64 bytenr, u64 num_bytes, u64 parent,
1806                          u64 root_objectid, u64 owner, u64 offset)
1807 {
1808         int ret;
1809         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1810                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1811
1812         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1813                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1814                                         parent, root_objectid, (int)owner,
1815                                         BTRFS_ADD_DELAYED_REF, NULL);
1816         } else {
1817                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1818                                         parent, root_objectid, owner, offset,
1819                                         BTRFS_ADD_DELAYED_REF, NULL);
1820         }
1821         return ret;
1822 }
1823
1824 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1825                                   struct btrfs_root *root,
1826                                   u64 bytenr, u64 num_bytes,
1827                                   u64 parent, u64 root_objectid,
1828                                   u64 owner, u64 offset, int refs_to_add,
1829                                   struct btrfs_delayed_extent_op *extent_op)
1830 {
1831         struct btrfs_path *path;
1832         struct extent_buffer *leaf;
1833         struct btrfs_extent_item *item;
1834         u64 refs;
1835         int ret;
1836         int err = 0;
1837
1838         path = btrfs_alloc_path();
1839         if (!path)
1840                 return -ENOMEM;
1841
1842         path->reada = 1;
1843         path->leave_spinning = 1;
1844         /* this will setup the path even if it fails to insert the back ref */
1845         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1846                                            path, bytenr, num_bytes, parent,
1847                                            root_objectid, owner, offset,
1848                                            refs_to_add, extent_op);
1849         if (ret == 0)
1850                 goto out;
1851
1852         if (ret != -EAGAIN) {
1853                 err = ret;
1854                 goto out;
1855         }
1856
1857         leaf = path->nodes[0];
1858         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1859         refs = btrfs_extent_refs(leaf, item);
1860         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1861         if (extent_op)
1862                 __run_delayed_extent_op(extent_op, leaf, item);
1863
1864         btrfs_mark_buffer_dirty(leaf);
1865         btrfs_release_path(path);
1866
1867         path->reada = 1;
1868         path->leave_spinning = 1;
1869
1870         /* now insert the actual backref */
1871         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1872                                     path, bytenr, parent, root_objectid,
1873                                     owner, offset, refs_to_add);
1874         BUG_ON(ret);
1875 out:
1876         btrfs_free_path(path);
1877         return err;
1878 }
1879
1880 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1881                                 struct btrfs_root *root,
1882                                 struct btrfs_delayed_ref_node *node,
1883                                 struct btrfs_delayed_extent_op *extent_op,
1884                                 int insert_reserved)
1885 {
1886         int ret = 0;
1887         struct btrfs_delayed_data_ref *ref;
1888         struct btrfs_key ins;
1889         u64 parent = 0;
1890         u64 ref_root = 0;
1891         u64 flags = 0;
1892
1893         ins.objectid = node->bytenr;
1894         ins.offset = node->num_bytes;
1895         ins.type = BTRFS_EXTENT_ITEM_KEY;
1896
1897         ref = btrfs_delayed_node_to_data_ref(node);
1898         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1899                 parent = ref->parent;
1900         else
1901                 ref_root = ref->root;
1902
1903         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1904                 if (extent_op) {
1905                         BUG_ON(extent_op->update_key);
1906                         flags |= extent_op->flags_to_set;
1907                 }
1908                 ret = alloc_reserved_file_extent(trans, root,
1909                                                  parent, ref_root, flags,
1910                                                  ref->objectid, ref->offset,
1911                                                  &ins, node->ref_mod);
1912         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1913                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1914                                              node->num_bytes, parent,
1915                                              ref_root, ref->objectid,
1916                                              ref->offset, node->ref_mod,
1917                                              extent_op);
1918         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1919                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1920                                           node->num_bytes, parent,
1921                                           ref_root, ref->objectid,
1922                                           ref->offset, node->ref_mod,
1923                                           extent_op);
1924         } else {
1925                 BUG();
1926         }
1927         return ret;
1928 }
1929
1930 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1931                                     struct extent_buffer *leaf,
1932                                     struct btrfs_extent_item *ei)
1933 {
1934         u64 flags = btrfs_extent_flags(leaf, ei);
1935         if (extent_op->update_flags) {
1936                 flags |= extent_op->flags_to_set;
1937                 btrfs_set_extent_flags(leaf, ei, flags);
1938         }
1939
1940         if (extent_op->update_key) {
1941                 struct btrfs_tree_block_info *bi;
1942                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1943                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1944                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1945         }
1946 }
1947
1948 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1949                                  struct btrfs_root *root,
1950                                  struct btrfs_delayed_ref_node *node,
1951                                  struct btrfs_delayed_extent_op *extent_op)
1952 {
1953         struct btrfs_key key;
1954         struct btrfs_path *path;
1955         struct btrfs_extent_item *ei;
1956         struct extent_buffer *leaf;
1957         u32 item_size;
1958         int ret;
1959         int err = 0;
1960
1961         path = btrfs_alloc_path();
1962         if (!path)
1963                 return -ENOMEM;
1964
1965         key.objectid = node->bytenr;
1966         key.type = BTRFS_EXTENT_ITEM_KEY;
1967         key.offset = node->num_bytes;
1968
1969         path->reada = 1;
1970         path->leave_spinning = 1;
1971         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1972                                 path, 0, 1);
1973         if (ret < 0) {
1974                 err = ret;
1975                 goto out;
1976         }
1977         if (ret > 0) {
1978                 err = -EIO;
1979                 goto out;
1980         }
1981
1982         leaf = path->nodes[0];
1983         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1984 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1985         if (item_size < sizeof(*ei)) {
1986                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1987                                              path, (u64)-1, 0);
1988                 if (ret < 0) {
1989                         err = ret;
1990                         goto out;
1991                 }
1992                 leaf = path->nodes[0];
1993                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1994         }
1995 #endif
1996         BUG_ON(item_size < sizeof(*ei));
1997         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1998         __run_delayed_extent_op(extent_op, leaf, ei);
1999
2000         btrfs_mark_buffer_dirty(leaf);
2001 out:
2002         btrfs_free_path(path);
2003         return err;
2004 }
2005
2006 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2007                                 struct btrfs_root *root,
2008                                 struct btrfs_delayed_ref_node *node,
2009                                 struct btrfs_delayed_extent_op *extent_op,
2010                                 int insert_reserved)
2011 {
2012         int ret = 0;
2013         struct btrfs_delayed_tree_ref *ref;
2014         struct btrfs_key ins;
2015         u64 parent = 0;
2016         u64 ref_root = 0;
2017
2018         ins.objectid = node->bytenr;
2019         ins.offset = node->num_bytes;
2020         ins.type = BTRFS_EXTENT_ITEM_KEY;
2021
2022         ref = btrfs_delayed_node_to_tree_ref(node);
2023         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2024                 parent = ref->parent;
2025         else
2026                 ref_root = ref->root;
2027
2028         BUG_ON(node->ref_mod != 1);
2029         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2030                 BUG_ON(!extent_op || !extent_op->update_flags ||
2031                        !extent_op->update_key);
2032                 ret = alloc_reserved_tree_block(trans, root,
2033                                                 parent, ref_root,
2034                                                 extent_op->flags_to_set,
2035                                                 &extent_op->key,
2036                                                 ref->level, &ins);
2037         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2038                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2039                                              node->num_bytes, parent, ref_root,
2040                                              ref->level, 0, 1, extent_op);
2041         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2042                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2043                                           node->num_bytes, parent, ref_root,
2044                                           ref->level, 0, 1, extent_op);
2045         } else {
2046                 BUG();
2047         }
2048         return ret;
2049 }
2050
2051 /* helper function to actually process a single delayed ref entry */
2052 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2053                                struct btrfs_root *root,
2054                                struct btrfs_delayed_ref_node *node,
2055                                struct btrfs_delayed_extent_op *extent_op,
2056                                int insert_reserved)
2057 {
2058         int ret;
2059         if (btrfs_delayed_ref_is_head(node)) {
2060                 struct btrfs_delayed_ref_head *head;
2061                 /*
2062                  * we've hit the end of the chain and we were supposed
2063                  * to insert this extent into the tree.  But, it got
2064                  * deleted before we ever needed to insert it, so all
2065                  * we have to do is clean up the accounting
2066                  */
2067                 BUG_ON(extent_op);
2068                 head = btrfs_delayed_node_to_head(node);
2069                 if (insert_reserved) {
2070                         btrfs_pin_extent(root, node->bytenr,
2071                                          node->num_bytes, 1);
2072                         if (head->is_data) {
2073                                 ret = btrfs_del_csums(trans, root,
2074                                                       node->bytenr,
2075                                                       node->num_bytes);
2076                                 BUG_ON(ret);
2077                         }
2078                 }
2079                 mutex_unlock(&head->mutex);
2080                 return 0;
2081         }
2082
2083         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2084             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2085                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2086                                            insert_reserved);
2087         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2088                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2089                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2090                                            insert_reserved);
2091         else
2092                 BUG();
2093         return ret;
2094 }
2095
2096 static noinline struct btrfs_delayed_ref_node *
2097 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2098 {
2099         struct rb_node *node;
2100         struct btrfs_delayed_ref_node *ref;
2101         int action = BTRFS_ADD_DELAYED_REF;
2102 again:
2103         /*
2104          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2105          * this prevents ref count from going down to zero when
2106          * there still are pending delayed ref.
2107          */
2108         node = rb_prev(&head->node.rb_node);
2109         while (1) {
2110                 if (!node)
2111                         break;
2112                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2113                                 rb_node);
2114                 if (ref->bytenr != head->node.bytenr)
2115                         break;
2116                 if (ref->action == action)
2117                         return ref;
2118                 node = rb_prev(node);
2119         }
2120         if (action == BTRFS_ADD_DELAYED_REF) {
2121                 action = BTRFS_DROP_DELAYED_REF;
2122                 goto again;
2123         }
2124         return NULL;
2125 }
2126
2127 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2128                                        struct btrfs_root *root,
2129                                        struct list_head *cluster)
2130 {
2131         struct btrfs_delayed_ref_root *delayed_refs;
2132         struct btrfs_delayed_ref_node *ref;
2133         struct btrfs_delayed_ref_head *locked_ref = NULL;
2134         struct btrfs_delayed_extent_op *extent_op;
2135         int ret;
2136         int count = 0;
2137         int must_insert_reserved = 0;
2138
2139         delayed_refs = &trans->transaction->delayed_refs;
2140         while (1) {
2141                 if (!locked_ref) {
2142                         /* pick a new head ref from the cluster list */
2143                         if (list_empty(cluster))
2144                                 break;
2145
2146                         locked_ref = list_entry(cluster->next,
2147                                      struct btrfs_delayed_ref_head, cluster);
2148
2149                         /* grab the lock that says we are going to process
2150                          * all the refs for this head */
2151                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2152
2153                         /*
2154                          * we may have dropped the spin lock to get the head
2155                          * mutex lock, and that might have given someone else
2156                          * time to free the head.  If that's true, it has been
2157                          * removed from our list and we can move on.
2158                          */
2159                         if (ret == -EAGAIN) {
2160                                 locked_ref = NULL;
2161                                 count++;
2162                                 continue;
2163                         }
2164                 }
2165
2166                 /*
2167                  * record the must insert reserved flag before we
2168                  * drop the spin lock.
2169                  */
2170                 must_insert_reserved = locked_ref->must_insert_reserved;
2171                 locked_ref->must_insert_reserved = 0;
2172
2173                 extent_op = locked_ref->extent_op;
2174                 locked_ref->extent_op = NULL;
2175
2176                 /*
2177                  * locked_ref is the head node, so we have to go one
2178                  * node back for any delayed ref updates
2179                  */
2180                 ref = select_delayed_ref(locked_ref);
2181                 if (!ref) {
2182                         /* All delayed refs have been processed, Go ahead
2183                          * and send the head node to run_one_delayed_ref,
2184                          * so that any accounting fixes can happen
2185                          */
2186                         ref = &locked_ref->node;
2187
2188                         if (extent_op && must_insert_reserved) {
2189                                 kfree(extent_op);
2190                                 extent_op = NULL;
2191                         }
2192
2193                         if (extent_op) {
2194                                 spin_unlock(&delayed_refs->lock);
2195
2196                                 ret = run_delayed_extent_op(trans, root,
2197                                                             ref, extent_op);
2198                                 BUG_ON(ret);
2199                                 kfree(extent_op);
2200
2201                                 cond_resched();
2202                                 spin_lock(&delayed_refs->lock);
2203                                 continue;
2204                         }
2205
2206                         list_del_init(&locked_ref->cluster);
2207                         locked_ref = NULL;
2208                 }
2209
2210                 ref->in_tree = 0;
2211                 rb_erase(&ref->rb_node, &delayed_refs->root);
2212                 delayed_refs->num_entries--;
2213
2214                 spin_unlock(&delayed_refs->lock);
2215
2216                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2217                                           must_insert_reserved);
2218                 BUG_ON(ret);
2219
2220                 btrfs_put_delayed_ref(ref);
2221                 kfree(extent_op);
2222                 count++;
2223
2224                 cond_resched();
2225                 spin_lock(&delayed_refs->lock);
2226         }
2227         return count;
2228 }
2229
2230 /*
2231  * this starts processing the delayed reference count updates and
2232  * extent insertions we have queued up so far.  count can be
2233  * 0, which means to process everything in the tree at the start
2234  * of the run (but not newly added entries), or it can be some target
2235  * number you'd like to process.
2236  */
2237 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2238                            struct btrfs_root *root, unsigned long count)
2239 {
2240         struct rb_node *node;
2241         struct btrfs_delayed_ref_root *delayed_refs;
2242         struct btrfs_delayed_ref_node *ref;
2243         struct list_head cluster;
2244         int ret;
2245         int run_all = count == (unsigned long)-1;
2246         int run_most = 0;
2247
2248         if (root == root->fs_info->extent_root)
2249                 root = root->fs_info->tree_root;
2250
2251         delayed_refs = &trans->transaction->delayed_refs;
2252         INIT_LIST_HEAD(&cluster);
2253 again:
2254         spin_lock(&delayed_refs->lock);
2255         if (count == 0) {
2256                 count = delayed_refs->num_entries * 2;
2257                 run_most = 1;
2258         }
2259         while (1) {
2260                 if (!(run_all || run_most) &&
2261                     delayed_refs->num_heads_ready < 64)
2262                         break;
2263
2264                 /*
2265                  * go find something we can process in the rbtree.  We start at
2266                  * the beginning of the tree, and then build a cluster
2267                  * of refs to process starting at the first one we are able to
2268                  * lock
2269                  */
2270                 ret = btrfs_find_ref_cluster(trans, &cluster,
2271                                              delayed_refs->run_delayed_start);
2272                 if (ret)
2273                         break;
2274
2275                 ret = run_clustered_refs(trans, root, &cluster);
2276                 BUG_ON(ret < 0);
2277
2278                 count -= min_t(unsigned long, ret, count);
2279
2280                 if (count == 0)
2281                         break;
2282         }
2283
2284         if (run_all) {
2285                 node = rb_first(&delayed_refs->root);
2286                 if (!node)
2287                         goto out;
2288                 count = (unsigned long)-1;
2289
2290                 while (node) {
2291                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2292                                        rb_node);
2293                         if (btrfs_delayed_ref_is_head(ref)) {
2294                                 struct btrfs_delayed_ref_head *head;
2295
2296                                 head = btrfs_delayed_node_to_head(ref);
2297                                 atomic_inc(&ref->refs);
2298
2299                                 spin_unlock(&delayed_refs->lock);
2300                                 /*
2301                                  * Mutex was contended, block until it's
2302                                  * released and try again
2303                                  */
2304                                 mutex_lock(&head->mutex);
2305                                 mutex_unlock(&head->mutex);
2306
2307                                 btrfs_put_delayed_ref(ref);
2308                                 cond_resched();
2309                                 goto again;
2310                         }
2311                         node = rb_next(node);
2312                 }
2313                 spin_unlock(&delayed_refs->lock);
2314                 schedule_timeout(1);
2315                 goto again;
2316         }
2317 out:
2318         spin_unlock(&delayed_refs->lock);
2319         return 0;
2320 }
2321
2322 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2323                                 struct btrfs_root *root,
2324                                 u64 bytenr, u64 num_bytes, u64 flags,
2325                                 int is_data)
2326 {
2327         struct btrfs_delayed_extent_op *extent_op;
2328         int ret;
2329
2330         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2331         if (!extent_op)
2332                 return -ENOMEM;
2333
2334         extent_op->flags_to_set = flags;
2335         extent_op->update_flags = 1;
2336         extent_op->update_key = 0;
2337         extent_op->is_data = is_data ? 1 : 0;
2338
2339         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2340         if (ret)
2341                 kfree(extent_op);
2342         return ret;
2343 }
2344
2345 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2346                                       struct btrfs_root *root,
2347                                       struct btrfs_path *path,
2348                                       u64 objectid, u64 offset, u64 bytenr)
2349 {
2350         struct btrfs_delayed_ref_head *head;
2351         struct btrfs_delayed_ref_node *ref;
2352         struct btrfs_delayed_data_ref *data_ref;
2353         struct btrfs_delayed_ref_root *delayed_refs;
2354         struct rb_node *node;
2355         int ret = 0;
2356
2357         ret = -ENOENT;
2358         delayed_refs = &trans->transaction->delayed_refs;
2359         spin_lock(&delayed_refs->lock);
2360         head = btrfs_find_delayed_ref_head(trans, bytenr);
2361         if (!head)
2362                 goto out;
2363
2364         if (!mutex_trylock(&head->mutex)) {
2365                 atomic_inc(&head->node.refs);
2366                 spin_unlock(&delayed_refs->lock);
2367
2368                 btrfs_release_path(path);
2369
2370                 /*
2371                  * Mutex was contended, block until it's released and let
2372                  * caller try again
2373                  */
2374                 mutex_lock(&head->mutex);
2375                 mutex_unlock(&head->mutex);
2376                 btrfs_put_delayed_ref(&head->node);
2377                 return -EAGAIN;
2378         }
2379
2380         node = rb_prev(&head->node.rb_node);
2381         if (!node)
2382                 goto out_unlock;
2383
2384         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2385
2386         if (ref->bytenr != bytenr)
2387                 goto out_unlock;
2388
2389         ret = 1;
2390         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2391                 goto out_unlock;
2392
2393         data_ref = btrfs_delayed_node_to_data_ref(ref);
2394
2395         node = rb_prev(node);
2396         if (node) {
2397                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2398                 if (ref->bytenr == bytenr)
2399                         goto out_unlock;
2400         }
2401
2402         if (data_ref->root != root->root_key.objectid ||
2403             data_ref->objectid != objectid || data_ref->offset != offset)
2404                 goto out_unlock;
2405
2406         ret = 0;
2407 out_unlock:
2408         mutex_unlock(&head->mutex);
2409 out:
2410         spin_unlock(&delayed_refs->lock);
2411         return ret;
2412 }
2413
2414 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2415                                         struct btrfs_root *root,
2416                                         struct btrfs_path *path,
2417                                         u64 objectid, u64 offset, u64 bytenr)
2418 {
2419         struct btrfs_root *extent_root = root->fs_info->extent_root;
2420         struct extent_buffer *leaf;
2421         struct btrfs_extent_data_ref *ref;
2422         struct btrfs_extent_inline_ref *iref;
2423         struct btrfs_extent_item *ei;
2424         struct btrfs_key key;
2425         u32 item_size;
2426         int ret;
2427
2428         key.objectid = bytenr;
2429         key.offset = (u64)-1;
2430         key.type = BTRFS_EXTENT_ITEM_KEY;
2431
2432         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2433         if (ret < 0)
2434                 goto out;
2435         BUG_ON(ret == 0);
2436
2437         ret = -ENOENT;
2438         if (path->slots[0] == 0)
2439                 goto out;
2440
2441         path->slots[0]--;
2442         leaf = path->nodes[0];
2443         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2444
2445         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2446                 goto out;
2447
2448         ret = 1;
2449         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2450 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2451         if (item_size < sizeof(*ei)) {
2452                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2453                 goto out;
2454         }
2455 #endif
2456         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2457
2458         if (item_size != sizeof(*ei) +
2459             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2460                 goto out;
2461
2462         if (btrfs_extent_generation(leaf, ei) <=
2463             btrfs_root_last_snapshot(&root->root_item))
2464                 goto out;
2465
2466         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2467         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2468             BTRFS_EXTENT_DATA_REF_KEY)
2469                 goto out;
2470
2471         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2472         if (btrfs_extent_refs(leaf, ei) !=
2473             btrfs_extent_data_ref_count(leaf, ref) ||
2474             btrfs_extent_data_ref_root(leaf, ref) !=
2475             root->root_key.objectid ||
2476             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2477             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2478                 goto out;
2479
2480         ret = 0;
2481 out:
2482         return ret;
2483 }
2484
2485 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2486                           struct btrfs_root *root,
2487                           u64 objectid, u64 offset, u64 bytenr)
2488 {
2489         struct btrfs_path *path;
2490         int ret;
2491         int ret2;
2492
2493         path = btrfs_alloc_path();
2494         if (!path)
2495                 return -ENOENT;
2496
2497         do {
2498                 ret = check_committed_ref(trans, root, path, objectid,
2499                                           offset, bytenr);
2500                 if (ret && ret != -ENOENT)
2501                         goto out;
2502
2503                 ret2 = check_delayed_ref(trans, root, path, objectid,
2504                                          offset, bytenr);
2505         } while (ret2 == -EAGAIN);
2506
2507         if (ret2 && ret2 != -ENOENT) {
2508                 ret = ret2;
2509                 goto out;
2510         }
2511
2512         if (ret != -ENOENT || ret2 != -ENOENT)
2513                 ret = 0;
2514 out:
2515         btrfs_free_path(path);
2516         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2517                 WARN_ON(ret > 0);
2518         return ret;
2519 }
2520
2521 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2522                            struct btrfs_root *root,
2523                            struct extent_buffer *buf,
2524                            int full_backref, int inc)
2525 {
2526         u64 bytenr;
2527         u64 num_bytes;
2528         u64 parent;
2529         u64 ref_root;
2530         u32 nritems;
2531         struct btrfs_key key;
2532         struct btrfs_file_extent_item *fi;
2533         int i;
2534         int level;
2535         int ret = 0;
2536         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2537                             u64, u64, u64, u64, u64, u64);
2538
2539         ref_root = btrfs_header_owner(buf);
2540         nritems = btrfs_header_nritems(buf);
2541         level = btrfs_header_level(buf);
2542
2543         if (!root->ref_cows && level == 0)
2544                 return 0;
2545
2546         if (inc)
2547                 process_func = btrfs_inc_extent_ref;
2548         else
2549                 process_func = btrfs_free_extent;
2550
2551         if (full_backref)
2552                 parent = buf->start;
2553         else
2554                 parent = 0;
2555
2556         for (i = 0; i < nritems; i++) {
2557                 if (level == 0) {
2558                         btrfs_item_key_to_cpu(buf, &key, i);
2559                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2560                                 continue;
2561                         fi = btrfs_item_ptr(buf, i,
2562                                             struct btrfs_file_extent_item);
2563                         if (btrfs_file_extent_type(buf, fi) ==
2564                             BTRFS_FILE_EXTENT_INLINE)
2565                                 continue;
2566                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2567                         if (bytenr == 0)
2568                                 continue;
2569
2570                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2571                         key.offset -= btrfs_file_extent_offset(buf, fi);
2572                         ret = process_func(trans, root, bytenr, num_bytes,
2573                                            parent, ref_root, key.objectid,
2574                                            key.offset);
2575                         if (ret)
2576                                 goto fail;
2577                 } else {
2578                         bytenr = btrfs_node_blockptr(buf, i);
2579                         num_bytes = btrfs_level_size(root, level - 1);
2580                         ret = process_func(trans, root, bytenr, num_bytes,
2581                                            parent, ref_root, level - 1, 0);
2582                         if (ret)
2583                                 goto fail;
2584                 }
2585         }
2586         return 0;
2587 fail:
2588         BUG();
2589         return ret;
2590 }
2591
2592 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2593                   struct extent_buffer *buf, int full_backref)
2594 {
2595         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2596 }
2597
2598 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2599                   struct extent_buffer *buf, int full_backref)
2600 {
2601         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2602 }
2603
2604 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2605                                  struct btrfs_root *root,
2606                                  struct btrfs_path *path,
2607                                  struct btrfs_block_group_cache *cache)
2608 {
2609         int ret;
2610         struct btrfs_root *extent_root = root->fs_info->extent_root;
2611         unsigned long bi;
2612         struct extent_buffer *leaf;
2613
2614         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2615         if (ret < 0)
2616                 goto fail;
2617         BUG_ON(ret);
2618
2619         leaf = path->nodes[0];
2620         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2621         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2622         btrfs_mark_buffer_dirty(leaf);
2623         btrfs_release_path(path);
2624 fail:
2625         if (ret)
2626                 return ret;
2627         return 0;
2628
2629 }
2630
2631 static struct btrfs_block_group_cache *
2632 next_block_group(struct btrfs_root *root,
2633                  struct btrfs_block_group_cache *cache)
2634 {
2635         struct rb_node *node;
2636         spin_lock(&root->fs_info->block_group_cache_lock);
2637         node = rb_next(&cache->cache_node);
2638         btrfs_put_block_group(cache);
2639         if (node) {
2640                 cache = rb_entry(node, struct btrfs_block_group_cache,
2641                                  cache_node);
2642                 btrfs_get_block_group(cache);
2643         } else
2644                 cache = NULL;
2645         spin_unlock(&root->fs_info->block_group_cache_lock);
2646         return cache;
2647 }
2648
2649 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2650                             struct btrfs_trans_handle *trans,
2651                             struct btrfs_path *path)
2652 {
2653         struct btrfs_root *root = block_group->fs_info->tree_root;
2654         struct inode *inode = NULL;
2655         u64 alloc_hint = 0;
2656         int dcs = BTRFS_DC_ERROR;
2657         int num_pages = 0;
2658         int retries = 0;
2659         int ret = 0;
2660
2661         /*
2662          * If this block group is smaller than 100 megs don't bother caching the
2663          * block group.
2664          */
2665         if (block_group->key.offset < (100 * 1024 * 1024)) {
2666                 spin_lock(&block_group->lock);
2667                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2668                 spin_unlock(&block_group->lock);
2669                 return 0;
2670         }
2671
2672 again:
2673         inode = lookup_free_space_inode(root, block_group, path);
2674         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2675                 ret = PTR_ERR(inode);
2676                 btrfs_release_path(path);
2677                 goto out;
2678         }
2679
2680         if (IS_ERR(inode)) {
2681                 BUG_ON(retries);
2682                 retries++;
2683
2684                 if (block_group->ro)
2685                         goto out_free;
2686
2687                 ret = create_free_space_inode(root, trans, block_group, path);
2688                 if (ret)
2689                         goto out_free;
2690                 goto again;
2691         }
2692
2693         /*
2694          * We want to set the generation to 0, that way if anything goes wrong
2695          * from here on out we know not to trust this cache when we load up next
2696          * time.
2697          */
2698         BTRFS_I(inode)->generation = 0;
2699         ret = btrfs_update_inode(trans, root, inode);
2700         WARN_ON(ret);
2701
2702         if (i_size_read(inode) > 0) {
2703                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2704                                                       inode);
2705                 if (ret)
2706                         goto out_put;
2707         }
2708
2709         spin_lock(&block_group->lock);
2710         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2711                 /* We're not cached, don't bother trying to write stuff out */
2712                 dcs = BTRFS_DC_WRITTEN;
2713                 spin_unlock(&block_group->lock);
2714                 goto out_put;
2715         }
2716         spin_unlock(&block_group->lock);
2717
2718         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2719         if (!num_pages)
2720                 num_pages = 1;
2721
2722         /*
2723          * Just to make absolutely sure we have enough space, we're going to
2724          * preallocate 12 pages worth of space for each block group.  In
2725          * practice we ought to use at most 8, but we need extra space so we can
2726          * add our header and have a terminator between the extents and the
2727          * bitmaps.
2728          */
2729         num_pages *= 16;
2730         num_pages *= PAGE_CACHE_SIZE;
2731
2732         ret = btrfs_check_data_free_space(inode, num_pages);
2733         if (ret)
2734                 goto out_put;
2735
2736         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2737                                               num_pages, num_pages,
2738                                               &alloc_hint);
2739         if (!ret)
2740                 dcs = BTRFS_DC_SETUP;
2741         btrfs_free_reserved_data_space(inode, num_pages);
2742 out_put:
2743         iput(inode);
2744 out_free:
2745         btrfs_release_path(path);
2746 out:
2747         spin_lock(&block_group->lock);
2748         block_group->disk_cache_state = dcs;
2749         spin_unlock(&block_group->lock);
2750
2751         return ret;
2752 }
2753
2754 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2755                                    struct btrfs_root *root)
2756 {
2757         struct btrfs_block_group_cache *cache;
2758         int err = 0;
2759         struct btrfs_path *path;
2760         u64 last = 0;
2761
2762         path = btrfs_alloc_path();
2763         if (!path)
2764                 return -ENOMEM;
2765
2766 again:
2767         while (1) {
2768                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2769                 while (cache) {
2770                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2771                                 break;
2772                         cache = next_block_group(root, cache);
2773                 }
2774                 if (!cache) {
2775                         if (last == 0)
2776                                 break;
2777                         last = 0;
2778                         continue;
2779                 }
2780                 err = cache_save_setup(cache, trans, path);
2781                 last = cache->key.objectid + cache->key.offset;
2782                 btrfs_put_block_group(cache);
2783         }
2784
2785         while (1) {
2786                 if (last == 0) {
2787                         err = btrfs_run_delayed_refs(trans, root,
2788                                                      (unsigned long)-1);
2789                         BUG_ON(err);
2790                 }
2791
2792                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2793                 while (cache) {
2794                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2795                                 btrfs_put_block_group(cache);
2796                                 goto again;
2797                         }
2798
2799                         if (cache->dirty)
2800                                 break;
2801                         cache = next_block_group(root, cache);
2802                 }
2803                 if (!cache) {
2804                         if (last == 0)
2805                                 break;
2806                         last = 0;
2807                         continue;
2808                 }
2809
2810                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2811                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2812                 cache->dirty = 0;
2813                 last = cache->key.objectid + cache->key.offset;
2814
2815                 err = write_one_cache_group(trans, root, path, cache);
2816                 BUG_ON(err);
2817                 btrfs_put_block_group(cache);
2818         }
2819
2820         while (1) {
2821                 /*
2822                  * I don't think this is needed since we're just marking our
2823                  * preallocated extent as written, but just in case it can't
2824                  * hurt.
2825                  */
2826                 if (last == 0) {
2827                         err = btrfs_run_delayed_refs(trans, root,
2828                                                      (unsigned long)-1);
2829                         BUG_ON(err);
2830                 }
2831
2832                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2833                 while (cache) {
2834                         /*
2835                          * Really this shouldn't happen, but it could if we
2836                          * couldn't write the entire preallocated extent and
2837                          * splitting the extent resulted in a new block.
2838                          */
2839                         if (cache->dirty) {
2840                                 btrfs_put_block_group(cache);
2841                                 goto again;
2842                         }
2843                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2844                                 break;
2845                         cache = next_block_group(root, cache);
2846                 }
2847                 if (!cache) {
2848                         if (last == 0)
2849                                 break;
2850                         last = 0;
2851                         continue;
2852                 }
2853
2854                 btrfs_write_out_cache(root, trans, cache, path);
2855
2856                 /*
2857                  * If we didn't have an error then the cache state is still
2858                  * NEED_WRITE, so we can set it to WRITTEN.
2859                  */
2860                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2861                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2862                 last = cache->key.objectid + cache->key.offset;
2863                 btrfs_put_block_group(cache);
2864         }
2865
2866         btrfs_free_path(path);
2867         return 0;
2868 }
2869
2870 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2871 {
2872         struct btrfs_block_group_cache *block_group;
2873         int readonly = 0;
2874
2875         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2876         if (!block_group || block_group->ro)
2877                 readonly = 1;
2878         if (block_group)
2879                 btrfs_put_block_group(block_group);
2880         return readonly;
2881 }
2882
2883 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2884                              u64 total_bytes, u64 bytes_used,
2885                              struct btrfs_space_info **space_info)
2886 {
2887         struct btrfs_space_info *found;
2888         int i;
2889         int factor;
2890
2891         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2892                      BTRFS_BLOCK_GROUP_RAID10))
2893                 factor = 2;
2894         else
2895                 factor = 1;
2896
2897         found = __find_space_info(info, flags);
2898         if (found) {
2899                 spin_lock(&found->lock);
2900                 found->total_bytes += total_bytes;
2901                 found->disk_total += total_bytes * factor;
2902                 found->bytes_used += bytes_used;
2903                 found->disk_used += bytes_used * factor;
2904                 found->full = 0;
2905                 spin_unlock(&found->lock);
2906                 *space_info = found;
2907                 return 0;
2908         }
2909         found = kzalloc(sizeof(*found), GFP_NOFS);
2910         if (!found)
2911                 return -ENOMEM;
2912
2913         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2914                 INIT_LIST_HEAD(&found->block_groups[i]);
2915         init_rwsem(&found->groups_sem);
2916         spin_lock_init(&found->lock);
2917         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2918                                 BTRFS_BLOCK_GROUP_SYSTEM |
2919                                 BTRFS_BLOCK_GROUP_METADATA);
2920         found->total_bytes = total_bytes;
2921         found->disk_total = total_bytes * factor;
2922         found->bytes_used = bytes_used;
2923         found->disk_used = bytes_used * factor;
2924         found->bytes_pinned = 0;
2925         found->bytes_reserved = 0;
2926         found->bytes_readonly = 0;
2927         found->bytes_may_use = 0;
2928         found->full = 0;
2929         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
2930         found->chunk_alloc = 0;
2931         found->flush = 0;
2932         init_waitqueue_head(&found->wait);
2933         *space_info = found;
2934         list_add_rcu(&found->list, &info->space_info);
2935         return 0;
2936 }
2937
2938 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2939 {
2940         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2941                                    BTRFS_BLOCK_GROUP_RAID1 |
2942                                    BTRFS_BLOCK_GROUP_RAID10 |
2943                                    BTRFS_BLOCK_GROUP_DUP);
2944         if (extra_flags) {
2945                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2946                         fs_info->avail_data_alloc_bits |= extra_flags;
2947                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2948                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2949                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2950                         fs_info->avail_system_alloc_bits |= extra_flags;
2951         }
2952 }
2953
2954 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2955 {
2956         /*
2957          * we add in the count of missing devices because we want
2958          * to make sure that any RAID levels on a degraded FS
2959          * continue to be honored.
2960          */
2961         u64 num_devices = root->fs_info->fs_devices->rw_devices +
2962                 root->fs_info->fs_devices->missing_devices;
2963
2964         if (num_devices == 1)
2965                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2966         if (num_devices < 4)
2967                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2968
2969         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2970             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2971                       BTRFS_BLOCK_GROUP_RAID10))) {
2972                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2973         }
2974
2975         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2976             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2977                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2978         }
2979
2980         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2981             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2982              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2983              (flags & BTRFS_BLOCK_GROUP_DUP)))
2984                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2985         return flags;
2986 }
2987
2988 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
2989 {
2990         if (flags & BTRFS_BLOCK_GROUP_DATA)
2991                 flags |= root->fs_info->avail_data_alloc_bits &
2992                          root->fs_info->data_alloc_profile;
2993         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2994                 flags |= root->fs_info->avail_system_alloc_bits &
2995                          root->fs_info->system_alloc_profile;
2996         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
2997                 flags |= root->fs_info->avail_metadata_alloc_bits &
2998                          root->fs_info->metadata_alloc_profile;
2999         return btrfs_reduce_alloc_profile(root, flags);
3000 }
3001
3002 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3003 {
3004         u64 flags;
3005
3006         if (data)
3007                 flags = BTRFS_BLOCK_GROUP_DATA;
3008         else if (root == root->fs_info->chunk_root)
3009                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3010         else
3011                 flags = BTRFS_BLOCK_GROUP_METADATA;
3012
3013         return get_alloc_profile(root, flags);
3014 }
3015
3016 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3017 {
3018         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3019                                                        BTRFS_BLOCK_GROUP_DATA);
3020 }
3021
3022 /*
3023  * This will check the space that the inode allocates from to make sure we have
3024  * enough space for bytes.
3025  */
3026 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3027 {
3028         struct btrfs_space_info *data_sinfo;
3029         struct btrfs_root *root = BTRFS_I(inode)->root;
3030         u64 used;
3031         int ret = 0, committed = 0, alloc_chunk = 1;
3032
3033         /* make sure bytes are sectorsize aligned */
3034         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3035
3036         if (root == root->fs_info->tree_root ||
3037             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3038                 alloc_chunk = 0;
3039                 committed = 1;
3040         }
3041
3042         data_sinfo = BTRFS_I(inode)->space_info;
3043         if (!data_sinfo)
3044                 goto alloc;
3045
3046 again:
3047         /* make sure we have enough space to handle the data first */
3048         spin_lock(&data_sinfo->lock);
3049         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3050                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3051                 data_sinfo->bytes_may_use;
3052
3053         if (used + bytes > data_sinfo->total_bytes) {
3054                 struct btrfs_trans_handle *trans;
3055
3056                 /*
3057                  * if we don't have enough free bytes in this space then we need
3058                  * to alloc a new chunk.
3059                  */
3060                 if (!data_sinfo->full && alloc_chunk) {
3061                         u64 alloc_target;
3062
3063                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3064                         spin_unlock(&data_sinfo->lock);
3065 alloc:
3066                         alloc_target = btrfs_get_alloc_profile(root, 1);
3067                         trans = btrfs_join_transaction(root);
3068                         if (IS_ERR(trans))
3069                                 return PTR_ERR(trans);
3070
3071                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3072                                              bytes + 2 * 1024 * 1024,
3073                                              alloc_target,
3074                                              CHUNK_ALLOC_NO_FORCE);
3075                         btrfs_end_transaction(trans, root);
3076                         if (ret < 0) {
3077                                 if (ret != -ENOSPC)
3078                                         return ret;
3079                                 else
3080                                         goto commit_trans;
3081                         }
3082
3083                         if (!data_sinfo) {
3084                                 btrfs_set_inode_space_info(root, inode);
3085                                 data_sinfo = BTRFS_I(inode)->space_info;
3086                         }
3087                         goto again;
3088                 }
3089
3090                 /*
3091                  * If we have less pinned bytes than we want to allocate then
3092                  * don't bother committing the transaction, it won't help us.
3093                  */
3094                 if (data_sinfo->bytes_pinned < bytes)
3095                         committed = 1;
3096                 spin_unlock(&data_sinfo->lock);
3097
3098                 /* commit the current transaction and try again */
3099 commit_trans:
3100                 if (!committed &&
3101                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3102                         committed = 1;
3103                         trans = btrfs_join_transaction(root);
3104                         if (IS_ERR(trans))
3105                                 return PTR_ERR(trans);
3106                         ret = btrfs_commit_transaction(trans, root);
3107                         if (ret)
3108                                 return ret;
3109                         goto again;
3110                 }
3111
3112                 return -ENOSPC;
3113         }
3114         data_sinfo->bytes_may_use += bytes;
3115         BTRFS_I(inode)->reserved_bytes += bytes;
3116         spin_unlock(&data_sinfo->lock);
3117
3118         return 0;
3119 }
3120
3121 /*
3122  * called when we are clearing an delalloc extent from the
3123  * inode's io_tree or there was an error for whatever reason
3124  * after calling btrfs_check_data_free_space
3125  */
3126 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3127 {
3128         struct btrfs_root *root = BTRFS_I(inode)->root;
3129         struct btrfs_space_info *data_sinfo;
3130
3131         /* make sure bytes are sectorsize aligned */
3132         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3133
3134         data_sinfo = BTRFS_I(inode)->space_info;
3135         spin_lock(&data_sinfo->lock);
3136         data_sinfo->bytes_may_use -= bytes;
3137         BTRFS_I(inode)->reserved_bytes -= bytes;
3138         spin_unlock(&data_sinfo->lock);
3139 }
3140
3141 static void force_metadata_allocation(struct btrfs_fs_info *info)
3142 {
3143         struct list_head *head = &info->space_info;
3144         struct btrfs_space_info *found;
3145
3146         rcu_read_lock();
3147         list_for_each_entry_rcu(found, head, list) {
3148                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3149                         found->force_alloc = CHUNK_ALLOC_FORCE;
3150         }
3151         rcu_read_unlock();
3152 }
3153
3154 static int should_alloc_chunk(struct btrfs_root *root,
3155                               struct btrfs_space_info *sinfo, u64 alloc_bytes,
3156                               int force)
3157 {
3158         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3159         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3160         u64 thresh;
3161
3162         if (force == CHUNK_ALLOC_FORCE)
3163                 return 1;
3164
3165         /*
3166          * in limited mode, we want to have some free space up to
3167          * about 1% of the FS size.
3168          */
3169         if (force == CHUNK_ALLOC_LIMITED) {
3170                 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3171                 thresh = max_t(u64, 64 * 1024 * 1024,
3172                                div_factor_fine(thresh, 1));
3173
3174                 if (num_bytes - num_allocated < thresh)
3175                         return 1;
3176         }
3177
3178         /*
3179          * we have two similar checks here, one based on percentage
3180          * and once based on a hard number of 256MB.  The idea
3181          * is that if we have a good amount of free
3182          * room, don't allocate a chunk.  A good mount is
3183          * less than 80% utilized of the chunks we have allocated,
3184          * or more than 256MB free
3185          */
3186         if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3187                 return 0;
3188
3189         if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3190                 return 0;
3191
3192         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3193
3194         /* 256MB or 5% of the FS */
3195         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3196
3197         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3198                 return 0;
3199         return 1;
3200 }
3201
3202 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3203                           struct btrfs_root *extent_root, u64 alloc_bytes,
3204                           u64 flags, int force)
3205 {
3206         struct btrfs_space_info *space_info;
3207         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3208         int wait_for_alloc = 0;
3209         int ret = 0;
3210
3211         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3212
3213         space_info = __find_space_info(extent_root->fs_info, flags);
3214         if (!space_info) {
3215                 ret = update_space_info(extent_root->fs_info, flags,
3216                                         0, 0, &space_info);
3217                 BUG_ON(ret);
3218         }
3219         BUG_ON(!space_info);
3220
3221 again:
3222         spin_lock(&space_info->lock);
3223         if (space_info->force_alloc)
3224                 force = space_info->force_alloc;
3225         if (space_info->full) {
3226                 spin_unlock(&space_info->lock);
3227                 return 0;
3228         }
3229
3230         if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3231                 spin_unlock(&space_info->lock);
3232                 return 0;
3233         } else if (space_info->chunk_alloc) {
3234                 wait_for_alloc = 1;
3235         } else {
3236                 space_info->chunk_alloc = 1;
3237         }
3238
3239         spin_unlock(&space_info->lock);
3240
3241         mutex_lock(&fs_info->chunk_mutex);
3242
3243         /*
3244          * The chunk_mutex is held throughout the entirety of a chunk
3245          * allocation, so once we've acquired the chunk_mutex we know that the
3246          * other guy is done and we need to recheck and see if we should
3247          * allocate.
3248          */
3249         if (wait_for_alloc) {
3250                 mutex_unlock(&fs_info->chunk_mutex);
3251                 wait_for_alloc = 0;
3252                 goto again;
3253         }
3254
3255         /*
3256          * If we have mixed data/metadata chunks we want to make sure we keep
3257          * allocating mixed chunks instead of individual chunks.
3258          */
3259         if (btrfs_mixed_space_info(space_info))
3260                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3261
3262         /*
3263          * if we're doing a data chunk, go ahead and make sure that
3264          * we keep a reasonable number of metadata chunks allocated in the
3265          * FS as well.
3266          */
3267         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3268                 fs_info->data_chunk_allocations++;
3269                 if (!(fs_info->data_chunk_allocations %
3270                       fs_info->metadata_ratio))
3271                         force_metadata_allocation(fs_info);
3272         }
3273
3274         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3275         spin_lock(&space_info->lock);
3276         if (ret)
3277                 space_info->full = 1;
3278         else
3279                 ret = 1;
3280
3281         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3282         space_info->chunk_alloc = 0;
3283         spin_unlock(&space_info->lock);
3284         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3285         return ret;
3286 }
3287
3288 /*
3289  * shrink metadata reservation for delalloc
3290  */
3291 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3292                            struct btrfs_root *root, u64 to_reclaim, int sync)
3293 {
3294         struct btrfs_block_rsv *block_rsv;
3295         struct btrfs_space_info *space_info;
3296         u64 reserved;
3297         u64 max_reclaim;
3298         u64 reclaimed = 0;
3299         long time_left;
3300         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3301         int loops = 0;
3302         unsigned long progress;
3303
3304         block_rsv = &root->fs_info->delalloc_block_rsv;
3305         space_info = block_rsv->space_info;
3306
3307         smp_mb();
3308         reserved = space_info->bytes_reserved;
3309         progress = space_info->reservation_progress;
3310
3311         if (reserved == 0)
3312                 return 0;
3313
3314         smp_mb();
3315         if (root->fs_info->delalloc_bytes == 0) {
3316                 if (trans)
3317                         return 0;
3318                 btrfs_wait_ordered_extents(root, 0, 0);
3319                 return 0;
3320         }
3321
3322         max_reclaim = min(reserved, to_reclaim);
3323
3324         while (loops < 1024) {
3325                 /* have the flusher threads jump in and do some IO */
3326                 smp_mb();
3327                 nr_pages = min_t(unsigned long, nr_pages,
3328                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3329                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3330
3331                 spin_lock(&space_info->lock);
3332                 if (reserved > space_info->bytes_reserved)
3333                         reclaimed += reserved - space_info->bytes_reserved;
3334                 reserved = space_info->bytes_reserved;
3335                 spin_unlock(&space_info->lock);
3336
3337                 loops++;
3338
3339                 if (reserved == 0 || reclaimed >= max_reclaim)
3340                         break;
3341
3342                 if (trans && trans->transaction->blocked)
3343                         return -EAGAIN;
3344
3345                 time_left = schedule_timeout_interruptible(1);
3346
3347                 /* We were interrupted, exit */
3348                 if (time_left)
3349                         break;
3350
3351                 /* we've kicked the IO a few times, if anything has been freed,
3352                  * exit.  There is no sense in looping here for a long time
3353                  * when we really need to commit the transaction, or there are
3354                  * just too many writers without enough free space
3355                  */
3356
3357                 if (loops > 3) {
3358                         smp_mb();
3359                         if (progress != space_info->reservation_progress)
3360                                 break;
3361                 }
3362
3363         }
3364         if (reclaimed >= to_reclaim && !trans)
3365                 btrfs_wait_ordered_extents(root, 0, 0);
3366         return reclaimed >= to_reclaim;
3367 }
3368
3369 /*
3370  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3371  * idea is if this is the first call (retries == 0) then we will add to our
3372  * reserved count if we can't make the allocation in order to hold our place
3373  * while we go and try and free up space.  That way for retries > 1 we don't try
3374  * and add space, we just check to see if the amount of unused space is >= the
3375  * total space, meaning that our reservation is valid.
3376  *
3377  * However if we don't intend to retry this reservation, pass -1 as retries so
3378  * that it short circuits this logic.
3379  */
3380 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3381                                   struct btrfs_root *root,
3382                                   struct btrfs_block_rsv *block_rsv,
3383                                   u64 orig_bytes, int flush)
3384 {
3385         struct btrfs_space_info *space_info = block_rsv->space_info;
3386         u64 unused;
3387         u64 num_bytes = orig_bytes;
3388         int retries = 0;
3389         int ret = 0;
3390         bool committed = false;
3391         bool flushing = false;
3392
3393 again:
3394         ret = 0;
3395         spin_lock(&space_info->lock);
3396         /*
3397          * We only want to wait if somebody other than us is flushing and we are
3398          * actually alloed to flush.
3399          */
3400         while (flush && !flushing && space_info->flush) {
3401                 spin_unlock(&space_info->lock);
3402                 /*
3403                  * If we have a trans handle we can't wait because the flusher
3404                  * may have to commit the transaction, which would mean we would
3405                  * deadlock since we are waiting for the flusher to finish, but
3406                  * hold the current transaction open.
3407                  */
3408                 if (trans)
3409                         return -EAGAIN;
3410                 ret = wait_event_interruptible(space_info->wait,
3411                                                !space_info->flush);
3412                 /* Must have been interrupted, return */
3413                 if (ret)
3414                         return -EINTR;
3415
3416                 spin_lock(&space_info->lock);
3417         }
3418
3419         ret = -ENOSPC;
3420         unused = space_info->bytes_used + space_info->bytes_reserved +
3421                  space_info->bytes_pinned + space_info->bytes_readonly +
3422                  space_info->bytes_may_use;
3423
3424         /*
3425          * The idea here is that we've not already over-reserved the block group
3426          * then we can go ahead and save our reservation first and then start
3427          * flushing if we need to.  Otherwise if we've already overcommitted
3428          * lets start flushing stuff first and then come back and try to make
3429          * our reservation.
3430          */
3431         if (unused <= space_info->total_bytes) {
3432                 unused = space_info->total_bytes - unused;
3433                 if (unused >= num_bytes) {
3434                         space_info->bytes_reserved += orig_bytes;
3435                         ret = 0;
3436                 } else {
3437                         /*
3438                          * Ok set num_bytes to orig_bytes since we aren't
3439                          * overocmmitted, this way we only try and reclaim what
3440                          * we need.
3441                          */
3442                         num_bytes = orig_bytes;
3443                 }
3444         } else {
3445                 /*
3446                  * Ok we're over committed, set num_bytes to the overcommitted
3447                  * amount plus the amount of bytes that we need for this
3448                  * reservation.
3449                  */
3450                 num_bytes = unused - space_info->total_bytes +
3451                         (orig_bytes * (retries + 1));
3452         }
3453
3454         /*
3455          * Couldn't make our reservation, save our place so while we're trying
3456          * to reclaim space we can actually use it instead of somebody else
3457          * stealing it from us.
3458          */
3459         if (ret && flush) {
3460                 flushing = true;
3461                 space_info->flush = 1;
3462         }
3463
3464         spin_unlock(&space_info->lock);
3465
3466         if (!ret || !flush)
3467                 goto out;
3468
3469         /*
3470          * We do synchronous shrinking since we don't actually unreserve
3471          * metadata until after the IO is completed.
3472          */
3473         ret = shrink_delalloc(trans, root, num_bytes, 1);
3474         if (ret < 0)
3475                 goto out;
3476
3477         /*
3478          * So if we were overcommitted it's possible that somebody else flushed
3479          * out enough space and we simply didn't have enough space to reclaim,
3480          * so go back around and try again.
3481          */
3482         if (retries < 2) {
3483                 retries++;
3484                 goto again;
3485         }
3486
3487         /*
3488          * Not enough space to be reclaimed, don't bother committing the
3489          * transaction.
3490          */
3491         spin_lock(&space_info->lock);
3492         if (space_info->bytes_pinned < orig_bytes)
3493                 ret = -ENOSPC;
3494         spin_unlock(&space_info->lock);
3495         if (ret)
3496                 goto out;
3497
3498         ret = -EAGAIN;
3499         if (trans || committed)
3500                 goto out;
3501
3502         ret = -ENOSPC;
3503         trans = btrfs_join_transaction(root);
3504         if (IS_ERR(trans))
3505                 goto out;
3506         ret = btrfs_commit_transaction(trans, root);
3507         if (!ret) {
3508                 trans = NULL;
3509                 committed = true;
3510                 goto again;
3511         }
3512
3513 out:
3514         if (flushing) {
3515                 spin_lock(&space_info->lock);
3516                 space_info->flush = 0;
3517                 wake_up_all(&space_info->wait);
3518                 spin_unlock(&space_info->lock);
3519         }
3520         return ret;
3521 }
3522
3523 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3524                                              struct btrfs_root *root)
3525 {
3526         struct btrfs_block_rsv *block_rsv;
3527         if (root->ref_cows)
3528                 block_rsv = trans->block_rsv;
3529         else
3530                 block_rsv = root->block_rsv;
3531
3532         if (!block_rsv)
3533                 block_rsv = &root->fs_info->empty_block_rsv;
3534
3535         return block_rsv;
3536 }
3537
3538 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3539                                u64 num_bytes)
3540 {
3541         int ret = -ENOSPC;
3542         spin_lock(&block_rsv->lock);
3543         if (block_rsv->reserved >= num_bytes) {
3544                 block_rsv->reserved -= num_bytes;
3545                 if (block_rsv->reserved < block_rsv->size)
3546                         block_rsv->full = 0;
3547                 ret = 0;
3548         }
3549         spin_unlock(&block_rsv->lock);
3550         return ret;
3551 }
3552
3553 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3554                                 u64 num_bytes, int update_size)
3555 {
3556         spin_lock(&block_rsv->lock);
3557         block_rsv->reserved += num_bytes;
3558         if (update_size)
3559                 block_rsv->size += num_bytes;
3560         else if (block_rsv->reserved >= block_rsv->size)
3561                 block_rsv->full = 1;
3562         spin_unlock(&block_rsv->lock);
3563 }
3564
3565 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3566                                     struct btrfs_block_rsv *dest, u64 num_bytes)
3567 {
3568         struct btrfs_space_info *space_info = block_rsv->space_info;
3569
3570         spin_lock(&block_rsv->lock);
3571         if (num_bytes == (u64)-1)
3572                 num_bytes = block_rsv->size;
3573         block_rsv->size -= num_bytes;
3574         if (block_rsv->reserved >= block_rsv->size) {
3575                 num_bytes = block_rsv->reserved - block_rsv->size;
3576                 block_rsv->reserved = block_rsv->size;
3577                 block_rsv->full = 1;
3578         } else {
3579                 num_bytes = 0;
3580         }
3581         spin_unlock(&block_rsv->lock);
3582
3583         if (num_bytes > 0) {
3584                 if (dest) {
3585                         spin_lock(&dest->lock);
3586                         if (!dest->full) {
3587                                 u64 bytes_to_add;
3588
3589                                 bytes_to_add = dest->size - dest->reserved;
3590                                 bytes_to_add = min(num_bytes, bytes_to_add);
3591                                 dest->reserved += bytes_to_add;
3592                                 if (dest->reserved >= dest->size)
3593                                         dest->full = 1;
3594                                 num_bytes -= bytes_to_add;
3595                         }
3596                         spin_unlock(&dest->lock);
3597                 }
3598                 if (num_bytes) {
3599                         spin_lock(&space_info->lock);
3600                         space_info->bytes_reserved -= num_bytes;
3601                         space_info->reservation_progress++;
3602                         spin_unlock(&space_info->lock);
3603                 }
3604         }
3605 }
3606
3607 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3608                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3609 {
3610         int ret;
3611
3612         ret = block_rsv_use_bytes(src, num_bytes);
3613         if (ret)
3614                 return ret;
3615
3616         block_rsv_add_bytes(dst, num_bytes, 1);
3617         return 0;
3618 }
3619
3620 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3621 {
3622         memset(rsv, 0, sizeof(*rsv));
3623         spin_lock_init(&rsv->lock);
3624         atomic_set(&rsv->usage, 1);
3625         rsv->priority = 6;
3626         INIT_LIST_HEAD(&rsv->list);
3627 }
3628
3629 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3630 {
3631         struct btrfs_block_rsv *block_rsv;
3632         struct btrfs_fs_info *fs_info = root->fs_info;
3633
3634         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3635         if (!block_rsv)
3636                 return NULL;
3637
3638         btrfs_init_block_rsv(block_rsv);
3639         block_rsv->space_info = __find_space_info(fs_info,
3640                                                   BTRFS_BLOCK_GROUP_METADATA);
3641         return block_rsv;
3642 }
3643
3644 void btrfs_free_block_rsv(struct btrfs_root *root,
3645                           struct btrfs_block_rsv *rsv)
3646 {
3647         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3648                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3649                 if (!rsv->durable)
3650                         kfree(rsv);
3651         }
3652 }
3653
3654 /*
3655  * make the block_rsv struct be able to capture freed space.
3656  * the captured space will re-add to the the block_rsv struct
3657  * after transaction commit
3658  */
3659 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3660                                  struct btrfs_block_rsv *block_rsv)
3661 {
3662         block_rsv->durable = 1;
3663         mutex_lock(&fs_info->durable_block_rsv_mutex);
3664         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3665         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3666 }
3667
3668 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3669                         struct btrfs_root *root,
3670                         struct btrfs_block_rsv *block_rsv,
3671                         u64 num_bytes)
3672 {
3673         int ret;
3674
3675         if (num_bytes == 0)
3676                 return 0;
3677
3678         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3679         if (!ret) {
3680                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3681                 return 0;
3682         }
3683
3684         return ret;
3685 }
3686
3687 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3688                           struct btrfs_root *root,
3689                           struct btrfs_block_rsv *block_rsv,
3690                           u64 min_reserved, int min_factor)
3691 {
3692         u64 num_bytes = 0;
3693         int commit_trans = 0;
3694         int ret = -ENOSPC;
3695
3696         if (!block_rsv)
3697                 return 0;
3698
3699         spin_lock(&block_rsv->lock);
3700         if (min_factor > 0)
3701                 num_bytes = div_factor(block_rsv->size, min_factor);
3702         if (min_reserved > num_bytes)
3703                 num_bytes = min_reserved;
3704
3705         if (block_rsv->reserved >= num_bytes) {
3706                 ret = 0;
3707         } else {
3708                 num_bytes -= block_rsv->reserved;
3709                 if (block_rsv->durable &&
3710                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3711                         commit_trans = 1;
3712         }
3713         spin_unlock(&block_rsv->lock);
3714         if (!ret)
3715                 return 0;
3716
3717         if (block_rsv->refill_used) {
3718                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3719                                              num_bytes, 0);
3720                 if (!ret) {
3721                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3722                         return 0;
3723                 }
3724         }
3725
3726         if (commit_trans) {
3727                 if (trans)
3728                         return -EAGAIN;
3729                 trans = btrfs_join_transaction(root);
3730                 BUG_ON(IS_ERR(trans));
3731                 ret = btrfs_commit_transaction(trans, root);
3732                 return 0;
3733         }
3734
3735         return -ENOSPC;
3736 }
3737
3738 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3739                             struct btrfs_block_rsv *dst_rsv,
3740                             u64 num_bytes)
3741 {
3742         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3743 }
3744
3745 void btrfs_block_rsv_release(struct btrfs_root *root,
3746                              struct btrfs_block_rsv *block_rsv,
3747                              u64 num_bytes)
3748 {
3749         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3750         if (global_rsv->full || global_rsv == block_rsv ||
3751             block_rsv->space_info != global_rsv->space_info)
3752                 global_rsv = NULL;
3753         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3754 }
3755
3756 /*
3757  * helper to calculate size of global block reservation.
3758  * the desired value is sum of space used by extent tree,
3759  * checksum tree and root tree
3760  */
3761 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3762 {
3763         struct btrfs_space_info *sinfo;
3764         u64 num_bytes;
3765         u64 meta_used;
3766         u64 data_used;
3767         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3768
3769         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3770         spin_lock(&sinfo->lock);
3771         data_used = sinfo->bytes_used;
3772         spin_unlock(&sinfo->lock);
3773
3774         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3775         spin_lock(&sinfo->lock);
3776         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3777                 data_used = 0;
3778         meta_used = sinfo->bytes_used;
3779         spin_unlock(&sinfo->lock);
3780
3781         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3782                     csum_size * 2;
3783         num_bytes += div64_u64(data_used + meta_used, 50);
3784
3785         if (num_bytes * 3 > meta_used)
3786                 num_bytes = div64_u64(meta_used, 3);
3787
3788         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3789 }
3790
3791 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3792 {
3793         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3794         struct btrfs_space_info *sinfo = block_rsv->space_info;
3795         u64 num_bytes;
3796
3797         num_bytes = calc_global_metadata_size(fs_info);
3798
3799         spin_lock(&block_rsv->lock);
3800         spin_lock(&sinfo->lock);
3801
3802         block_rsv->size = num_bytes;
3803
3804         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3805                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3806                     sinfo->bytes_may_use;
3807
3808         if (sinfo->total_bytes > num_bytes) {
3809                 num_bytes = sinfo->total_bytes - num_bytes;
3810                 block_rsv->reserved += num_bytes;
3811                 sinfo->bytes_reserved += num_bytes;
3812         }
3813
3814         if (block_rsv->reserved >= block_rsv->size) {
3815                 num_bytes = block_rsv->reserved - block_rsv->size;
3816                 sinfo->bytes_reserved -= num_bytes;
3817                 sinfo->reservation_progress++;
3818                 block_rsv->reserved = block_rsv->size;
3819                 block_rsv->full = 1;
3820         }
3821
3822         spin_unlock(&sinfo->lock);
3823         spin_unlock(&block_rsv->lock);
3824 }
3825
3826 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3827 {
3828         struct btrfs_space_info *space_info;
3829
3830         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3831         fs_info->chunk_block_rsv.space_info = space_info;
3832         fs_info->chunk_block_rsv.priority = 10;
3833
3834         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3835         fs_info->global_block_rsv.space_info = space_info;
3836         fs_info->global_block_rsv.priority = 10;
3837         fs_info->global_block_rsv.refill_used = 1;
3838         fs_info->delalloc_block_rsv.space_info = space_info;
3839         fs_info->trans_block_rsv.space_info = space_info;
3840         fs_info->empty_block_rsv.space_info = space_info;
3841         fs_info->empty_block_rsv.priority = 10;
3842
3843         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3844         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3845         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3846         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3847         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3848
3849         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3850
3851         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3852
3853         update_global_block_rsv(fs_info);
3854 }
3855
3856 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3857 {
3858         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3859         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3860         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3861         WARN_ON(fs_info->trans_block_rsv.size > 0);
3862         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3863         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3864         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3865 }
3866
3867 int btrfs_truncate_reserve_metadata(struct btrfs_trans_handle *trans,
3868                                     struct btrfs_root *root,
3869                                     struct btrfs_block_rsv *rsv)
3870 {
3871         struct btrfs_block_rsv *trans_rsv = &root->fs_info->trans_block_rsv;
3872         u64 num_bytes;
3873         int ret;
3874
3875         /*
3876          * Truncate should be freeing data, but give us 2 items just in case it
3877          * needs to use some space.  We may want to be smarter about this in the
3878          * future.
3879          */
3880         num_bytes = btrfs_calc_trans_metadata_size(root, 2);
3881
3882         /* We already have enough bytes, just return */
3883         if (rsv->reserved >= num_bytes)
3884                 return 0;
3885
3886         num_bytes -= rsv->reserved;
3887
3888         /*
3889          * You should have reserved enough space before hand to do this, so this
3890          * should not fail.
3891          */
3892         ret = block_rsv_migrate_bytes(trans_rsv, rsv, num_bytes);
3893         BUG_ON(ret);
3894
3895         return 0;
3896 }
3897
3898 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3899                                   struct btrfs_root *root)
3900 {
3901         if (!trans->bytes_reserved)
3902                 return;
3903
3904         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3905         btrfs_block_rsv_release(root, trans->block_rsv,
3906                                 trans->bytes_reserved);
3907         trans->bytes_reserved = 0;
3908 }
3909
3910 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3911                                   struct inode *inode)
3912 {
3913         struct btrfs_root *root = BTRFS_I(inode)->root;
3914         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3915         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3916
3917         /*
3918          * We need to hold space in order to delete our orphan item once we've
3919          * added it, so this takes the reservation so we can release it later
3920          * when we are truly done with the orphan item.
3921          */
3922         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3923         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3924 }
3925
3926 void btrfs_orphan_release_metadata(struct inode *inode)
3927 {
3928         struct btrfs_root *root = BTRFS_I(inode)->root;
3929         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3930         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3931 }
3932
3933 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3934                                 struct btrfs_pending_snapshot *pending)
3935 {
3936         struct btrfs_root *root = pending->root;
3937         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3938         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3939         /*
3940          * two for root back/forward refs, two for directory entries
3941          * and one for root of the snapshot.
3942          */
3943         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3944         dst_rsv->space_info = src_rsv->space_info;
3945         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3946 }
3947
3948 static unsigned drop_outstanding_extent(struct inode *inode)
3949 {
3950         unsigned dropped_extents = 0;
3951
3952         spin_lock(&BTRFS_I(inode)->lock);
3953         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
3954         BTRFS_I(inode)->outstanding_extents--;
3955
3956         /*
3957          * If we have more or the same amount of outsanding extents than we have
3958          * reserved then we need to leave the reserved extents count alone.
3959          */
3960         if (BTRFS_I(inode)->outstanding_extents >=
3961             BTRFS_I(inode)->reserved_extents)
3962                 goto out;
3963
3964         dropped_extents = BTRFS_I(inode)->reserved_extents -
3965                 BTRFS_I(inode)->outstanding_extents;
3966         BTRFS_I(inode)->reserved_extents -= dropped_extents;
3967 out:
3968         spin_unlock(&BTRFS_I(inode)->lock);
3969         return dropped_extents;
3970 }
3971
3972 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3973 {
3974         return num_bytes >>= 3;
3975 }
3976
3977 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3978 {
3979         struct btrfs_root *root = BTRFS_I(inode)->root;
3980         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3981         u64 to_reserve = 0;
3982         unsigned nr_extents = 0;
3983         int ret;
3984
3985         if (btrfs_transaction_in_commit(root->fs_info))
3986                 schedule_timeout(1);
3987
3988         num_bytes = ALIGN(num_bytes, root->sectorsize);
3989
3990         spin_lock(&BTRFS_I(inode)->lock);
3991         BTRFS_I(inode)->outstanding_extents++;
3992
3993         if (BTRFS_I(inode)->outstanding_extents >
3994             BTRFS_I(inode)->reserved_extents) {
3995                 nr_extents = BTRFS_I(inode)->outstanding_extents -
3996                         BTRFS_I(inode)->reserved_extents;
3997                 BTRFS_I(inode)->reserved_extents += nr_extents;
3998
3999                 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4000         }
4001         spin_unlock(&BTRFS_I(inode)->lock);
4002
4003         to_reserve += calc_csum_metadata_size(inode, num_bytes);
4004         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4005         if (ret) {
4006                 unsigned dropped;
4007                 /*
4008                  * We don't need the return value since our reservation failed,
4009                  * we just need to clean up our counter.
4010                  */
4011                 dropped = drop_outstanding_extent(inode);
4012                 WARN_ON(dropped > 1);
4013                 return ret;
4014         }
4015
4016         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4017
4018         return 0;
4019 }
4020
4021 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4022 {
4023         struct btrfs_root *root = BTRFS_I(inode)->root;
4024         u64 to_free = 0;
4025         unsigned dropped;
4026
4027         num_bytes = ALIGN(num_bytes, root->sectorsize);
4028         dropped = drop_outstanding_extent(inode);
4029
4030         to_free = calc_csum_metadata_size(inode, num_bytes);
4031         if (dropped > 0)
4032                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4033
4034         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4035                                 to_free);
4036 }
4037
4038 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4039 {
4040         int ret;
4041
4042         ret = btrfs_check_data_free_space(inode, num_bytes);
4043         if (ret)
4044                 return ret;
4045
4046         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4047         if (ret) {
4048                 btrfs_free_reserved_data_space(inode, num_bytes);
4049                 return ret;
4050         }
4051
4052         return 0;
4053 }
4054
4055 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4056 {
4057         btrfs_delalloc_release_metadata(inode, num_bytes);
4058         btrfs_free_reserved_data_space(inode, num_bytes);
4059 }
4060
4061 static int update_block_group(struct btrfs_trans_handle *trans,
4062                               struct btrfs_root *root,
4063                               u64 bytenr, u64 num_bytes, int alloc)
4064 {
4065         struct btrfs_block_group_cache *cache = NULL;
4066         struct btrfs_fs_info *info = root->fs_info;
4067         u64 total = num_bytes;
4068         u64 old_val;
4069         u64 byte_in_group;
4070         int factor;
4071
4072         /* block accounting for super block */
4073         spin_lock(&info->delalloc_lock);
4074         old_val = btrfs_super_bytes_used(&info->super_copy);
4075         if (alloc)
4076                 old_val += num_bytes;
4077         else
4078                 old_val -= num_bytes;
4079         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4080         spin_unlock(&info->delalloc_lock);
4081
4082         while (total) {
4083                 cache = btrfs_lookup_block_group(info, bytenr);
4084                 if (!cache)
4085                         return -1;
4086                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4087                                     BTRFS_BLOCK_GROUP_RAID1 |
4088                                     BTRFS_BLOCK_GROUP_RAID10))
4089                         factor = 2;
4090                 else
4091                         factor = 1;
4092                 /*
4093                  * If this block group has free space cache written out, we
4094                  * need to make sure to load it if we are removing space.  This
4095                  * is because we need the unpinning stage to actually add the
4096                  * space back to the block group, otherwise we will leak space.
4097                  */
4098                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4099                         cache_block_group(cache, trans, NULL, 1);
4100
4101                 byte_in_group = bytenr - cache->key.objectid;
4102                 WARN_ON(byte_in_group > cache->key.offset);
4103
4104                 spin_lock(&cache->space_info->lock);
4105                 spin_lock(&cache->lock);
4106
4107                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4108                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4109                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4110
4111                 cache->dirty = 1;
4112                 old_val = btrfs_block_group_used(&cache->item);
4113                 num_bytes = min(total, cache->key.offset - byte_in_group);
4114                 if (alloc) {
4115                         old_val += num_bytes;
4116                         btrfs_set_block_group_used(&cache->item, old_val);
4117                         cache->reserved -= num_bytes;
4118                         cache->space_info->bytes_reserved -= num_bytes;
4119                         cache->space_info->reservation_progress++;
4120                         cache->space_info->bytes_used += num_bytes;
4121                         cache->space_info->disk_used += num_bytes * factor;
4122                         spin_unlock(&cache->lock);
4123                         spin_unlock(&cache->space_info->lock);
4124                 } else {
4125                         old_val -= num_bytes;
4126                         btrfs_set_block_group_used(&cache->item, old_val);
4127                         cache->pinned += num_bytes;
4128                         cache->space_info->bytes_pinned += num_bytes;
4129                         cache->space_info->bytes_used -= num_bytes;
4130                         cache->space_info->disk_used -= num_bytes * factor;
4131                         spin_unlock(&cache->lock);
4132                         spin_unlock(&cache->space_info->lock);
4133
4134                         set_extent_dirty(info->pinned_extents,
4135                                          bytenr, bytenr + num_bytes - 1,
4136                                          GFP_NOFS | __GFP_NOFAIL);
4137                 }
4138                 btrfs_put_block_group(cache);
4139                 total -= num_bytes;
4140                 bytenr += num_bytes;
4141         }
4142         return 0;
4143 }
4144
4145 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4146 {
4147         struct btrfs_block_group_cache *cache;
4148         u64 bytenr;
4149
4150         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4151         if (!cache)
4152                 return 0;
4153
4154         bytenr = cache->key.objectid;
4155         btrfs_put_block_group(cache);
4156
4157         return bytenr;
4158 }
4159
4160 static int pin_down_extent(struct btrfs_root *root,
4161                            struct btrfs_block_group_cache *cache,
4162                            u64 bytenr, u64 num_bytes, int reserved)
4163 {
4164         spin_lock(&cache->space_info->lock);
4165         spin_lock(&cache->lock);
4166         cache->pinned += num_bytes;
4167         cache->space_info->bytes_pinned += num_bytes;
4168         if (reserved) {
4169                 cache->reserved -= num_bytes;
4170                 cache->space_info->bytes_reserved -= num_bytes;
4171                 cache->space_info->reservation_progress++;
4172         }
4173         spin_unlock(&cache->lock);
4174         spin_unlock(&cache->space_info->lock);
4175
4176         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4177                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4178         return 0;
4179 }
4180
4181 /*
4182  * this function must be called within transaction
4183  */
4184 int btrfs_pin_extent(struct btrfs_root *root,
4185                      u64 bytenr, u64 num_bytes, int reserved)
4186 {
4187         struct btrfs_block_group_cache *cache;
4188
4189         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4190         BUG_ON(!cache);
4191
4192         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4193
4194         btrfs_put_block_group(cache);
4195         return 0;
4196 }
4197
4198 /*
4199  * update size of reserved extents. this function may return -EAGAIN
4200  * if 'reserve' is true or 'sinfo' is false.
4201  */
4202 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4203                                 u64 num_bytes, int reserve, int sinfo)
4204 {
4205         int ret = 0;
4206         if (sinfo) {
4207                 struct btrfs_space_info *space_info = cache->space_info;
4208                 spin_lock(&space_info->lock);
4209                 spin_lock(&cache->lock);
4210                 if (reserve) {
4211                         if (cache->ro) {
4212                                 ret = -EAGAIN;
4213                         } else {
4214                                 cache->reserved += num_bytes;
4215                                 space_info->bytes_reserved += num_bytes;
4216                         }
4217                 } else {
4218                         if (cache->ro)
4219                                 space_info->bytes_readonly += num_bytes;
4220                         cache->reserved -= num_bytes;
4221                         space_info->bytes_reserved -= num_bytes;
4222                         space_info->reservation_progress++;
4223                 }
4224                 spin_unlock(&cache->lock);
4225                 spin_unlock(&space_info->lock);
4226         } else {
4227                 spin_lock(&cache->lock);
4228                 if (cache->ro) {
4229                         ret = -EAGAIN;
4230                 } else {
4231                         if (reserve)
4232                                 cache->reserved += num_bytes;
4233                         else
4234                                 cache->reserved -= num_bytes;
4235                 }
4236                 spin_unlock(&cache->lock);
4237         }
4238         return ret;
4239 }
4240
4241 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4242                                 struct btrfs_root *root)
4243 {
4244         struct btrfs_fs_info *fs_info = root->fs_info;
4245         struct btrfs_caching_control *next;
4246         struct btrfs_caching_control *caching_ctl;
4247         struct btrfs_block_group_cache *cache;
4248
4249         down_write(&fs_info->extent_commit_sem);
4250
4251         list_for_each_entry_safe(caching_ctl, next,
4252                                  &fs_info->caching_block_groups, list) {
4253                 cache = caching_ctl->block_group;
4254                 if (block_group_cache_done(cache)) {
4255                         cache->last_byte_to_unpin = (u64)-1;
4256                         list_del_init(&caching_ctl->list);
4257                         put_caching_control(caching_ctl);
4258                 } else {
4259                         cache->last_byte_to_unpin = caching_ctl->progress;
4260                 }
4261         }
4262
4263         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4264                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4265         else
4266                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4267
4268         up_write(&fs_info->extent_commit_sem);
4269
4270         update_global_block_rsv(fs_info);
4271         return 0;
4272 }
4273
4274 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4275 {
4276         struct btrfs_fs_info *fs_info = root->fs_info;
4277         struct btrfs_block_group_cache *cache = NULL;
4278         u64 len;
4279
4280         while (start <= end) {
4281                 if (!cache ||
4282                     start >= cache->key.objectid + cache->key.offset) {
4283                         if (cache)
4284                                 btrfs_put_block_group(cache);
4285                         cache = btrfs_lookup_block_group(fs_info, start);
4286                         BUG_ON(!cache);
4287                 }
4288
4289                 len = cache->key.objectid + cache->key.offset - start;
4290                 len = min(len, end + 1 - start);
4291
4292                 if (start < cache->last_byte_to_unpin) {
4293                         len = min(len, cache->last_byte_to_unpin - start);
4294                         btrfs_add_free_space(cache, start, len);
4295                 }
4296
4297                 start += len;
4298
4299                 spin_lock(&cache->space_info->lock);
4300                 spin_lock(&cache->lock);
4301                 cache->pinned -= len;
4302                 cache->space_info->bytes_pinned -= len;
4303                 if (cache->ro) {
4304                         cache->space_info->bytes_readonly += len;
4305                 } else if (cache->reserved_pinned > 0) {
4306                         len = min(len, cache->reserved_pinned);
4307                         cache->reserved_pinned -= len;
4308                         cache->space_info->bytes_reserved += len;
4309                 }
4310                 spin_unlock(&cache->lock);
4311                 spin_unlock(&cache->space_info->lock);
4312         }
4313
4314         if (cache)
4315                 btrfs_put_block_group(cache);
4316         return 0;
4317 }
4318
4319 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4320                                struct btrfs_root *root)
4321 {
4322         struct btrfs_fs_info *fs_info = root->fs_info;
4323         struct extent_io_tree *unpin;
4324         struct btrfs_block_rsv *block_rsv;
4325         struct btrfs_block_rsv *next_rsv;
4326         u64 start;
4327         u64 end;
4328         int idx;
4329         int ret;
4330
4331         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4332                 unpin = &fs_info->freed_extents[1];
4333         else
4334                 unpin = &fs_info->freed_extents[0];
4335
4336         while (1) {
4337                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4338                                             EXTENT_DIRTY);
4339                 if (ret)
4340                         break;
4341
4342                 if (btrfs_test_opt(root, DISCARD))
4343                         ret = btrfs_discard_extent(root, start,
4344                                                    end + 1 - start, NULL);
4345
4346                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4347                 unpin_extent_range(root, start, end);
4348                 cond_resched();
4349         }
4350
4351         mutex_lock(&fs_info->durable_block_rsv_mutex);
4352         list_for_each_entry_safe(block_rsv, next_rsv,
4353                                  &fs_info->durable_block_rsv_list, list) {
4354
4355                 idx = trans->transid & 0x1;
4356                 if (block_rsv->freed[idx] > 0) {
4357                         block_rsv_add_bytes(block_rsv,
4358                                             block_rsv->freed[idx], 0);
4359                         block_rsv->freed[idx] = 0;
4360                 }
4361                 if (atomic_read(&block_rsv->usage) == 0) {
4362                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4363
4364                         if (block_rsv->freed[0] == 0 &&
4365                             block_rsv->freed[1] == 0) {
4366                                 list_del_init(&block_rsv->list);
4367                                 kfree(block_rsv);
4368                         }
4369                 } else {
4370                         btrfs_block_rsv_release(root, block_rsv, 0);
4371                 }
4372         }
4373         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4374
4375         return 0;
4376 }
4377
4378 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4379                                 struct btrfs_root *root,
4380                                 u64 bytenr, u64 num_bytes, u64 parent,
4381                                 u64 root_objectid, u64 owner_objectid,
4382                                 u64 owner_offset, int refs_to_drop,
4383                                 struct btrfs_delayed_extent_op *extent_op)
4384 {
4385         struct btrfs_key key;
4386         struct btrfs_path *path;
4387         struct btrfs_fs_info *info = root->fs_info;
4388         struct btrfs_root *extent_root = info->extent_root;
4389         struct extent_buffer *leaf;
4390         struct btrfs_extent_item *ei;
4391         struct btrfs_extent_inline_ref *iref;
4392         int ret;
4393         int is_data;
4394         int extent_slot = 0;
4395         int found_extent = 0;
4396         int num_to_del = 1;
4397         u32 item_size;
4398         u64 refs;
4399
4400         path = btrfs_alloc_path();
4401         if (!path)
4402                 return -ENOMEM;
4403
4404         path->reada = 1;
4405         path->leave_spinning = 1;
4406
4407         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4408         BUG_ON(!is_data && refs_to_drop != 1);
4409
4410         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4411                                     bytenr, num_bytes, parent,
4412                                     root_objectid, owner_objectid,
4413                                     owner_offset);
4414         if (ret == 0) {
4415                 extent_slot = path->slots[0];
4416                 while (extent_slot >= 0) {
4417                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4418                                               extent_slot);
4419                         if (key.objectid != bytenr)
4420                                 break;
4421                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4422                             key.offset == num_bytes) {
4423                                 found_extent = 1;
4424                                 break;
4425                         }
4426                         if (path->slots[0] - extent_slot > 5)
4427                                 break;
4428                         extent_slot--;
4429                 }
4430 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4431                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4432                 if (found_extent && item_size < sizeof(*ei))
4433                         found_extent = 0;
4434 #endif
4435                 if (!found_extent) {
4436                         BUG_ON(iref);
4437                         ret = remove_extent_backref(trans, extent_root, path,
4438                                                     NULL, refs_to_drop,
4439                                                     is_data);
4440                         BUG_ON(ret);
4441                         btrfs_release_path(path);
4442                         path->leave_spinning = 1;
4443
4444                         key.objectid = bytenr;
4445                         key.type = BTRFS_EXTENT_ITEM_KEY;
4446                         key.offset = num_bytes;
4447
4448                         ret = btrfs_search_slot(trans, extent_root,
4449                                                 &key, path, -1, 1);
4450                         if (ret) {
4451                                 printk(KERN_ERR "umm, got %d back from search"
4452                                        ", was looking for %llu\n", ret,
4453                                        (unsigned long long)bytenr);
4454                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4455                         }
4456                         BUG_ON(ret);
4457                         extent_slot = path->slots[0];
4458                 }
4459         } else {
4460                 btrfs_print_leaf(extent_root, path->nodes[0]);
4461                 WARN_ON(1);
4462                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4463                        "parent %llu root %llu  owner %llu offset %llu\n",
4464                        (unsigned long long)bytenr,
4465                        (unsigned long long)parent,
4466                        (unsigned long long)root_objectid,
4467                        (unsigned long long)owner_objectid,
4468                        (unsigned long long)owner_offset);
4469         }
4470
4471         leaf = path->nodes[0];
4472         item_size = btrfs_item_size_nr(leaf, extent_slot);
4473 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4474         if (item_size < sizeof(*ei)) {
4475                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4476                 ret = convert_extent_item_v0(trans, extent_root, path,
4477                                              owner_objectid, 0);
4478                 BUG_ON(ret < 0);
4479
4480                 btrfs_release_path(path);
4481                 path->leave_spinning = 1;
4482
4483                 key.objectid = bytenr;
4484                 key.type = BTRFS_EXTENT_ITEM_KEY;
4485                 key.offset = num_bytes;
4486
4487                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4488                                         -1, 1);
4489                 if (ret) {
4490                         printk(KERN_ERR "umm, got %d back from search"
4491                                ", was looking for %llu\n", ret,
4492                                (unsigned long long)bytenr);
4493                         btrfs_print_leaf(extent_root, path->nodes[0]);
4494                 }
4495                 BUG_ON(ret);
4496                 extent_slot = path->slots[0];
4497                 leaf = path->nodes[0];
4498                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4499         }
4500 #endif
4501         BUG_ON(item_size < sizeof(*ei));
4502         ei = btrfs_item_ptr(leaf, extent_slot,
4503                             struct btrfs_extent_item);
4504         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4505                 struct btrfs_tree_block_info *bi;
4506                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4507                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4508                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4509         }
4510
4511         refs = btrfs_extent_refs(leaf, ei);
4512         BUG_ON(refs < refs_to_drop);
4513         refs -= refs_to_drop;
4514
4515         if (refs > 0) {
4516                 if (extent_op)
4517                         __run_delayed_extent_op(extent_op, leaf, ei);
4518                 /*
4519                  * In the case of inline back ref, reference count will
4520                  * be updated by remove_extent_backref
4521                  */
4522                 if (iref) {
4523                         BUG_ON(!found_extent);
4524                 } else {
4525                         btrfs_set_extent_refs(leaf, ei, refs);
4526                         btrfs_mark_buffer_dirty(leaf);
4527                 }
4528                 if (found_extent) {
4529                         ret = remove_extent_backref(trans, extent_root, path,
4530                                                     iref, refs_to_drop,
4531                                                     is_data);
4532                         BUG_ON(ret);
4533                 }
4534         } else {
4535                 if (found_extent) {
4536                         BUG_ON(is_data && refs_to_drop !=
4537                                extent_data_ref_count(root, path, iref));
4538                         if (iref) {
4539                                 BUG_ON(path->slots[0] != extent_slot);
4540                         } else {
4541                                 BUG_ON(path->slots[0] != extent_slot + 1);
4542                                 path->slots[0] = extent_slot;
4543                                 num_to_del = 2;
4544                         }
4545                 }
4546
4547                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4548                                       num_to_del);
4549                 BUG_ON(ret);
4550                 btrfs_release_path(path);
4551
4552                 if (is_data) {
4553                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4554                         BUG_ON(ret);
4555                 } else {
4556                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4557                              bytenr >> PAGE_CACHE_SHIFT,
4558                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4559                 }
4560
4561                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4562                 BUG_ON(ret);
4563         }
4564         btrfs_free_path(path);
4565         return ret;
4566 }
4567
4568 /*
4569  * when we free an block, it is possible (and likely) that we free the last
4570  * delayed ref for that extent as well.  This searches the delayed ref tree for
4571  * a given extent, and if there are no other delayed refs to be processed, it
4572  * removes it from the tree.
4573  */
4574 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4575                                       struct btrfs_root *root, u64 bytenr)
4576 {
4577         struct btrfs_delayed_ref_head *head;
4578         struct btrfs_delayed_ref_root *delayed_refs;
4579         struct btrfs_delayed_ref_node *ref;
4580         struct rb_node *node;
4581         int ret = 0;
4582
4583         delayed_refs = &trans->transaction->delayed_refs;
4584         spin_lock(&delayed_refs->lock);
4585         head = btrfs_find_delayed_ref_head(trans, bytenr);
4586         if (!head)
4587                 goto out;
4588
4589         node = rb_prev(&head->node.rb_node);
4590         if (!node)
4591                 goto out;
4592
4593         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4594
4595         /* there are still entries for this ref, we can't drop it */
4596         if (ref->bytenr == bytenr)
4597                 goto out;
4598
4599         if (head->extent_op) {
4600                 if (!head->must_insert_reserved)
4601                         goto out;
4602                 kfree(head->extent_op);
4603                 head->extent_op = NULL;
4604         }
4605
4606         /*
4607          * waiting for the lock here would deadlock.  If someone else has it
4608          * locked they are already in the process of dropping it anyway
4609          */
4610         if (!mutex_trylock(&head->mutex))
4611                 goto out;
4612
4613         /*
4614          * at this point we have a head with no other entries.  Go
4615          * ahead and process it.
4616          */
4617         head->node.in_tree = 0;
4618         rb_erase(&head->node.rb_node, &delayed_refs->root);
4619
4620         delayed_refs->num_entries--;
4621
4622         /*
4623          * we don't take a ref on the node because we're removing it from the
4624          * tree, so we just steal the ref the tree was holding.
4625          */
4626         delayed_refs->num_heads--;
4627         if (list_empty(&head->cluster))
4628                 delayed_refs->num_heads_ready--;
4629
4630         list_del_init(&head->cluster);
4631         spin_unlock(&delayed_refs->lock);
4632
4633         BUG_ON(head->extent_op);
4634         if (head->must_insert_reserved)
4635                 ret = 1;
4636
4637         mutex_unlock(&head->mutex);
4638         btrfs_put_delayed_ref(&head->node);
4639         return ret;
4640 out:
4641         spin_unlock(&delayed_refs->lock);
4642         return 0;
4643 }
4644
4645 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4646                            struct btrfs_root *root,
4647                            struct extent_buffer *buf,
4648                            u64 parent, int last_ref)
4649 {
4650         struct btrfs_block_rsv *block_rsv;
4651         struct btrfs_block_group_cache *cache = NULL;
4652         int ret;
4653
4654         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4655                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4656                                                 parent, root->root_key.objectid,
4657                                                 btrfs_header_level(buf),
4658                                                 BTRFS_DROP_DELAYED_REF, NULL);
4659                 BUG_ON(ret);
4660         }
4661
4662         if (!last_ref)
4663                 return;
4664
4665         block_rsv = get_block_rsv(trans, root);
4666         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4667         if (block_rsv->space_info != cache->space_info)
4668                 goto out;
4669
4670         if (btrfs_header_generation(buf) == trans->transid) {
4671                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4672                         ret = check_ref_cleanup(trans, root, buf->start);
4673                         if (!ret)
4674                                 goto pin;
4675                 }
4676
4677                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4678                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4679                         goto pin;
4680                 }
4681
4682                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4683
4684                 btrfs_add_free_space(cache, buf->start, buf->len);
4685                 ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0);
4686                 if (ret == -EAGAIN) {
4687                         /* block group became read-only */
4688                         btrfs_update_reserved_bytes(cache, buf->len, 0, 1);
4689                         goto out;
4690                 }
4691
4692                 ret = 1;
4693                 spin_lock(&block_rsv->lock);
4694                 if (block_rsv->reserved < block_rsv->size) {
4695                         block_rsv->reserved += buf->len;
4696                         ret = 0;
4697                 }
4698                 spin_unlock(&block_rsv->lock);
4699
4700                 if (ret) {
4701                         spin_lock(&cache->space_info->lock);
4702                         cache->space_info->bytes_reserved -= buf->len;
4703                         cache->space_info->reservation_progress++;
4704                         spin_unlock(&cache->space_info->lock);
4705                 }
4706                 goto out;
4707         }
4708 pin:
4709         if (block_rsv->durable && !cache->ro) {
4710                 ret = 0;
4711                 spin_lock(&cache->lock);
4712                 if (!cache->ro) {
4713                         cache->reserved_pinned += buf->len;
4714                         ret = 1;
4715                 }
4716                 spin_unlock(&cache->lock);
4717
4718                 if (ret) {
4719                         spin_lock(&block_rsv->lock);
4720                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4721                         spin_unlock(&block_rsv->lock);
4722                 }
4723         }
4724 out:
4725         /*
4726          * Deleting the buffer, clear the corrupt flag since it doesn't matter
4727          * anymore.
4728          */
4729         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4730         btrfs_put_block_group(cache);
4731 }
4732
4733 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4734                       struct btrfs_root *root,
4735                       u64 bytenr, u64 num_bytes, u64 parent,
4736                       u64 root_objectid, u64 owner, u64 offset)
4737 {
4738         int ret;
4739
4740         /*
4741          * tree log blocks never actually go into the extent allocation
4742          * tree, just update pinning info and exit early.
4743          */
4744         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4745                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4746                 /* unlocks the pinned mutex */
4747                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4748                 ret = 0;
4749         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4750                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4751                                         parent, root_objectid, (int)owner,
4752                                         BTRFS_DROP_DELAYED_REF, NULL);
4753                 BUG_ON(ret);
4754         } else {
4755                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4756                                         parent, root_objectid, owner,
4757                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4758                 BUG_ON(ret);
4759         }
4760         return ret;
4761 }
4762
4763 static u64 stripe_align(struct btrfs_root *root, u64 val)
4764 {
4765         u64 mask = ((u64)root->stripesize - 1);
4766         u64 ret = (val + mask) & ~mask;
4767         return ret;
4768 }
4769
4770 /*
4771  * when we wait for progress in the block group caching, its because
4772  * our allocation attempt failed at least once.  So, we must sleep
4773  * and let some progress happen before we try again.
4774  *
4775  * This function will sleep at least once waiting for new free space to
4776  * show up, and then it will check the block group free space numbers
4777  * for our min num_bytes.  Another option is to have it go ahead
4778  * and look in the rbtree for a free extent of a given size, but this
4779  * is a good start.
4780  */
4781 static noinline int
4782 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4783                                 u64 num_bytes)
4784 {
4785         struct btrfs_caching_control *caching_ctl;
4786         DEFINE_WAIT(wait);
4787
4788         caching_ctl = get_caching_control(cache);
4789         if (!caching_ctl)
4790                 return 0;
4791
4792         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4793                    (cache->free_space_ctl->free_space >= num_bytes));
4794
4795         put_caching_control(caching_ctl);
4796         return 0;
4797 }
4798
4799 static noinline int
4800 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4801 {
4802         struct btrfs_caching_control *caching_ctl;
4803         DEFINE_WAIT(wait);
4804
4805         caching_ctl = get_caching_control(cache);
4806         if (!caching_ctl)
4807                 return 0;
4808
4809         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4810
4811         put_caching_control(caching_ctl);
4812         return 0;
4813 }
4814
4815 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4816 {
4817         int index;
4818         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4819                 index = 0;
4820         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4821                 index = 1;
4822         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4823                 index = 2;
4824         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4825                 index = 3;
4826         else
4827                 index = 4;
4828         return index;
4829 }
4830
4831 enum btrfs_loop_type {
4832         LOOP_FIND_IDEAL = 0,
4833         LOOP_CACHING_NOWAIT = 1,
4834         LOOP_CACHING_WAIT = 2,
4835         LOOP_ALLOC_CHUNK = 3,
4836         LOOP_NO_EMPTY_SIZE = 4,
4837 };
4838
4839 /*
4840  * walks the btree of allocated extents and find a hole of a given size.
4841  * The key ins is changed to record the hole:
4842  * ins->objectid == block start
4843  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4844  * ins->offset == number of blocks
4845  * Any available blocks before search_start are skipped.
4846  */
4847 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4848                                      struct btrfs_root *orig_root,
4849                                      u64 num_bytes, u64 empty_size,
4850                                      u64 search_start, u64 search_end,
4851                                      u64 hint_byte, struct btrfs_key *ins,
4852                                      u64 data)
4853 {
4854         int ret = 0;
4855         struct btrfs_root *root = orig_root->fs_info->extent_root;
4856         struct btrfs_free_cluster *last_ptr = NULL;
4857         struct btrfs_block_group_cache *block_group = NULL;
4858         int empty_cluster = 2 * 1024 * 1024;
4859         int allowed_chunk_alloc = 0;
4860         int done_chunk_alloc = 0;
4861         struct btrfs_space_info *space_info;
4862         int last_ptr_loop = 0;
4863         int loop = 0;
4864         int index = 0;
4865         bool found_uncached_bg = false;
4866         bool failed_cluster_refill = false;
4867         bool failed_alloc = false;
4868         bool use_cluster = true;
4869         u64 ideal_cache_percent = 0;
4870         u64 ideal_cache_offset = 0;
4871
4872         WARN_ON(num_bytes < root->sectorsize);
4873         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4874         ins->objectid = 0;
4875         ins->offset = 0;
4876
4877         space_info = __find_space_info(root->fs_info, data);
4878         if (!space_info) {
4879                 printk(KERN_ERR "No space info for %llu\n", data);
4880                 return -ENOSPC;
4881         }
4882
4883         /*
4884          * If the space info is for both data and metadata it means we have a
4885          * small filesystem and we can't use the clustering stuff.
4886          */
4887         if (btrfs_mixed_space_info(space_info))
4888                 use_cluster = false;
4889
4890         if (orig_root->ref_cows || empty_size)
4891                 allowed_chunk_alloc = 1;
4892
4893         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4894                 last_ptr = &root->fs_info->meta_alloc_cluster;
4895                 if (!btrfs_test_opt(root, SSD))
4896                         empty_cluster = 64 * 1024;
4897         }
4898
4899         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4900             btrfs_test_opt(root, SSD)) {
4901                 last_ptr = &root->fs_info->data_alloc_cluster;
4902         }
4903
4904         if (last_ptr) {
4905                 spin_lock(&last_ptr->lock);
4906                 if (last_ptr->block_group)
4907                         hint_byte = last_ptr->window_start;
4908                 spin_unlock(&last_ptr->lock);
4909         }
4910
4911         search_start = max(search_start, first_logical_byte(root, 0));
4912         search_start = max(search_start, hint_byte);
4913
4914         if (!last_ptr)
4915                 empty_cluster = 0;
4916
4917         if (search_start == hint_byte) {
4918 ideal_cache:
4919                 block_group = btrfs_lookup_block_group(root->fs_info,
4920                                                        search_start);
4921                 /*
4922                  * we don't want to use the block group if it doesn't match our
4923                  * allocation bits, or if its not cached.
4924                  *
4925                  * However if we are re-searching with an ideal block group
4926                  * picked out then we don't care that the block group is cached.
4927                  */
4928                 if (block_group && block_group_bits(block_group, data) &&
4929                     (block_group->cached != BTRFS_CACHE_NO ||
4930                      search_start == ideal_cache_offset)) {
4931                         down_read(&space_info->groups_sem);
4932                         if (list_empty(&block_group->list) ||
4933                             block_group->ro) {
4934                                 /*
4935                                  * someone is removing this block group,
4936                                  * we can't jump into the have_block_group
4937                                  * target because our list pointers are not
4938                                  * valid
4939                                  */
4940                                 btrfs_put_block_group(block_group);
4941                                 up_read(&space_info->groups_sem);
4942                         } else {
4943                                 index = get_block_group_index(block_group);
4944                                 goto have_block_group;
4945                         }
4946                 } else if (block_group) {
4947                         btrfs_put_block_group(block_group);
4948                 }
4949         }
4950 search:
4951         down_read(&space_info->groups_sem);
4952         list_for_each_entry(block_group, &space_info->block_groups[index],
4953                             list) {
4954                 u64 offset;
4955                 int cached;
4956
4957                 btrfs_get_block_group(block_group);
4958                 search_start = block_group->key.objectid;
4959
4960                 /*
4961                  * this can happen if we end up cycling through all the
4962                  * raid types, but we want to make sure we only allocate
4963                  * for the proper type.
4964                  */
4965                 if (!block_group_bits(block_group, data)) {
4966                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
4967                                 BTRFS_BLOCK_GROUP_RAID1 |
4968                                 BTRFS_BLOCK_GROUP_RAID10;
4969
4970                         /*
4971                          * if they asked for extra copies and this block group
4972                          * doesn't provide them, bail.  This does allow us to
4973                          * fill raid0 from raid1.
4974                          */
4975                         if ((data & extra) && !(block_group->flags & extra))
4976                                 goto loop;
4977                 }
4978
4979 have_block_group:
4980                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4981                         u64 free_percent;
4982
4983                         ret = cache_block_group(block_group, trans,
4984                                                 orig_root, 1);
4985                         if (block_group->cached == BTRFS_CACHE_FINISHED)
4986                                 goto have_block_group;
4987
4988                         free_percent = btrfs_block_group_used(&block_group->item);
4989                         free_percent *= 100;
4990                         free_percent = div64_u64(free_percent,
4991                                                  block_group->key.offset);
4992                         free_percent = 100 - free_percent;
4993                         if (free_percent > ideal_cache_percent &&
4994                             likely(!block_group->ro)) {
4995                                 ideal_cache_offset = block_group->key.objectid;
4996                                 ideal_cache_percent = free_percent;
4997                         }
4998
4999                         /*
5000                          * The caching workers are limited to 2 threads, so we
5001                          * can queue as much work as we care to.
5002                          */
5003                         if (loop > LOOP_FIND_IDEAL) {
5004                                 ret = cache_block_group(block_group, trans,
5005                                                         orig_root, 0);
5006                                 BUG_ON(ret);
5007                         }
5008                         found_uncached_bg = true;
5009
5010                         /*
5011                          * If loop is set for cached only, try the next block
5012                          * group.
5013                          */
5014                         if (loop == LOOP_FIND_IDEAL)
5015                                 goto loop;
5016                 }
5017
5018                 cached = block_group_cache_done(block_group);
5019                 if (unlikely(!cached))
5020                         found_uncached_bg = true;
5021
5022                 if (unlikely(block_group->ro))
5023                         goto loop;
5024
5025                 spin_lock(&block_group->free_space_ctl->tree_lock);
5026                 if (cached &&
5027                     block_group->free_space_ctl->free_space <
5028                     num_bytes + empty_size) {
5029                         spin_unlock(&block_group->free_space_ctl->tree_lock);
5030                         goto loop;
5031                 }
5032                 spin_unlock(&block_group->free_space_ctl->tree_lock);
5033
5034                 /*
5035                  * Ok we want to try and use the cluster allocator, so lets look
5036                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5037                  * have tried the cluster allocator plenty of times at this
5038                  * point and not have found anything, so we are likely way too
5039                  * fragmented for the clustering stuff to find anything, so lets
5040                  * just skip it and let the allocator find whatever block it can
5041                  * find
5042                  */
5043                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5044                         /*
5045                          * the refill lock keeps out other
5046                          * people trying to start a new cluster
5047                          */
5048                         spin_lock(&last_ptr->refill_lock);
5049                         if (last_ptr->block_group &&
5050                             (last_ptr->block_group->ro ||
5051                             !block_group_bits(last_ptr->block_group, data))) {
5052                                 offset = 0;
5053                                 goto refill_cluster;
5054                         }
5055
5056                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5057                                                  num_bytes, search_start);
5058                         if (offset) {
5059                                 /* we have a block, we're done */
5060                                 spin_unlock(&last_ptr->refill_lock);
5061                                 goto checks;
5062                         }
5063
5064                         spin_lock(&last_ptr->lock);
5065                         /*
5066                          * whoops, this cluster doesn't actually point to
5067                          * this block group.  Get a ref on the block
5068                          * group is does point to and try again
5069                          */
5070                         if (!last_ptr_loop && last_ptr->block_group &&
5071                             last_ptr->block_group != block_group) {
5072
5073                                 btrfs_put_block_group(block_group);
5074                                 block_group = last_ptr->block_group;
5075                                 btrfs_get_block_group(block_group);
5076                                 spin_unlock(&last_ptr->lock);
5077                                 spin_unlock(&last_ptr->refill_lock);
5078
5079                                 last_ptr_loop = 1;
5080                                 search_start = block_group->key.objectid;
5081                                 /*
5082                                  * we know this block group is properly
5083                                  * in the list because
5084                                  * btrfs_remove_block_group, drops the
5085                                  * cluster before it removes the block
5086                                  * group from the list
5087                                  */
5088                                 goto have_block_group;
5089                         }
5090                         spin_unlock(&last_ptr->lock);
5091 refill_cluster:
5092                         /*
5093                          * this cluster didn't work out, free it and
5094                          * start over
5095                          */
5096                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5097
5098                         last_ptr_loop = 0;
5099
5100                         /* allocate a cluster in this block group */
5101                         ret = btrfs_find_space_cluster(trans, root,
5102                                                block_group, last_ptr,
5103                                                offset, num_bytes,
5104                                                empty_cluster + empty_size);
5105                         if (ret == 0) {
5106                                 /*
5107                                  * now pull our allocation out of this
5108                                  * cluster
5109                                  */
5110                                 offset = btrfs_alloc_from_cluster(block_group,
5111                                                   last_ptr, num_bytes,
5112                                                   search_start);
5113                                 if (offset) {
5114                                         /* we found one, proceed */
5115                                         spin_unlock(&last_ptr->refill_lock);
5116                                         goto checks;
5117                                 }
5118                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5119                                    && !failed_cluster_refill) {
5120                                 spin_unlock(&last_ptr->refill_lock);
5121
5122                                 failed_cluster_refill = true;
5123                                 wait_block_group_cache_progress(block_group,
5124                                        num_bytes + empty_cluster + empty_size);
5125                                 goto have_block_group;
5126                         }
5127
5128                         /*
5129                          * at this point we either didn't find a cluster
5130                          * or we weren't able to allocate a block from our
5131                          * cluster.  Free the cluster we've been trying
5132                          * to use, and go to the next block group
5133                          */
5134                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5135                         spin_unlock(&last_ptr->refill_lock);
5136                         goto loop;
5137                 }
5138
5139                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5140                                                     num_bytes, empty_size);
5141                 /*
5142                  * If we didn't find a chunk, and we haven't failed on this
5143                  * block group before, and this block group is in the middle of
5144                  * caching and we are ok with waiting, then go ahead and wait
5145                  * for progress to be made, and set failed_alloc to true.
5146                  *
5147                  * If failed_alloc is true then we've already waited on this
5148                  * block group once and should move on to the next block group.
5149                  */
5150                 if (!offset && !failed_alloc && !cached &&
5151                     loop > LOOP_CACHING_NOWAIT) {
5152                         wait_block_group_cache_progress(block_group,
5153                                                 num_bytes + empty_size);
5154                         failed_alloc = true;
5155                         goto have_block_group;
5156                 } else if (!offset) {
5157                         goto loop;
5158                 }
5159 checks:
5160                 search_start = stripe_align(root, offset);
5161                 /* move on to the next group */
5162                 if (search_start + num_bytes >= search_end) {
5163                         btrfs_add_free_space(block_group, offset, num_bytes);
5164                         goto loop;
5165                 }
5166
5167                 /* move on to the next group */
5168                 if (search_start + num_bytes >
5169                     block_group->key.objectid + block_group->key.offset) {
5170                         btrfs_add_free_space(block_group, offset, num_bytes);
5171                         goto loop;
5172                 }
5173
5174                 ins->objectid = search_start;
5175                 ins->offset = num_bytes;
5176
5177                 if (offset < search_start)
5178                         btrfs_add_free_space(block_group, offset,
5179                                              search_start - offset);
5180                 BUG_ON(offset > search_start);
5181
5182                 ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1,
5183                                             (data & BTRFS_BLOCK_GROUP_DATA));
5184                 if (ret == -EAGAIN) {
5185                         btrfs_add_free_space(block_group, offset, num_bytes);
5186                         goto loop;
5187                 }
5188
5189                 /* we are all good, lets return */
5190                 ins->objectid = search_start;
5191                 ins->offset = num_bytes;
5192
5193                 if (offset < search_start)
5194                         btrfs_add_free_space(block_group, offset,
5195                                              search_start - offset);
5196                 BUG_ON(offset > search_start);
5197                 btrfs_put_block_group(block_group);
5198                 break;
5199 loop:
5200                 failed_cluster_refill = false;
5201                 failed_alloc = false;
5202                 BUG_ON(index != get_block_group_index(block_group));
5203                 btrfs_put_block_group(block_group);
5204         }
5205         up_read(&space_info->groups_sem);
5206
5207         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5208                 goto search;
5209
5210         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5211          *                      for them to make caching progress.  Also
5212          *                      determine the best possible bg to cache
5213          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5214          *                      caching kthreads as we move along
5215          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5216          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5217          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5218          *                      again
5219          */
5220         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5221                 index = 0;
5222                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5223                         found_uncached_bg = false;
5224                         loop++;
5225                         if (!ideal_cache_percent)
5226                                 goto search;
5227
5228                         /*
5229                          * 1 of the following 2 things have happened so far
5230                          *
5231                          * 1) We found an ideal block group for caching that
5232                          * is mostly full and will cache quickly, so we might
5233                          * as well wait for it.
5234                          *
5235                          * 2) We searched for cached only and we didn't find
5236                          * anything, and we didn't start any caching kthreads
5237                          * either, so chances are we will loop through and
5238                          * start a couple caching kthreads, and then come back
5239                          * around and just wait for them.  This will be slower
5240                          * because we will have 2 caching kthreads reading at
5241                          * the same time when we could have just started one
5242                          * and waited for it to get far enough to give us an
5243                          * allocation, so go ahead and go to the wait caching
5244                          * loop.
5245                          */
5246                         loop = LOOP_CACHING_WAIT;
5247                         search_start = ideal_cache_offset;
5248                         ideal_cache_percent = 0;
5249                         goto ideal_cache;
5250                 } else if (loop == LOOP_FIND_IDEAL) {
5251                         /*
5252                          * Didn't find a uncached bg, wait on anything we find
5253                          * next.
5254                          */
5255                         loop = LOOP_CACHING_WAIT;
5256                         goto search;
5257                 }
5258
5259                 loop++;
5260
5261                 if (loop == LOOP_ALLOC_CHUNK) {
5262                        if (allowed_chunk_alloc) {
5263                                 ret = do_chunk_alloc(trans, root, num_bytes +
5264                                                      2 * 1024 * 1024, data,
5265                                                      CHUNK_ALLOC_LIMITED);
5266                                 allowed_chunk_alloc = 0;
5267                                 if (ret == 1)
5268                                         done_chunk_alloc = 1;
5269                         } else if (!done_chunk_alloc &&
5270                                    space_info->force_alloc ==
5271                                    CHUNK_ALLOC_NO_FORCE) {
5272                                 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5273                         }
5274
5275                        /*
5276                         * We didn't allocate a chunk, go ahead and drop the
5277                         * empty size and loop again.
5278                         */
5279                        if (!done_chunk_alloc)
5280                                loop = LOOP_NO_EMPTY_SIZE;
5281                 }
5282
5283                 if (loop == LOOP_NO_EMPTY_SIZE) {
5284                         empty_size = 0;
5285                         empty_cluster = 0;
5286                 }
5287
5288                 goto search;
5289         } else if (!ins->objectid) {
5290                 ret = -ENOSPC;
5291         } else if (ins->objectid) {
5292                 ret = 0;
5293         }
5294
5295         return ret;
5296 }
5297
5298 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5299                             int dump_block_groups)
5300 {
5301         struct btrfs_block_group_cache *cache;
5302         int index = 0;
5303
5304         spin_lock(&info->lock);
5305         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5306                (unsigned long long)(info->total_bytes - info->bytes_used -
5307                                     info->bytes_pinned - info->bytes_reserved -
5308                                     info->bytes_readonly),
5309                (info->full) ? "" : "not ");
5310         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5311                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5312                (unsigned long long)info->total_bytes,
5313                (unsigned long long)info->bytes_used,
5314                (unsigned long long)info->bytes_pinned,
5315                (unsigned long long)info->bytes_reserved,
5316                (unsigned long long)info->bytes_may_use,
5317                (unsigned long long)info->bytes_readonly);
5318         spin_unlock(&info->lock);
5319
5320         if (!dump_block_groups)
5321                 return;
5322
5323         down_read(&info->groups_sem);
5324 again:
5325         list_for_each_entry(cache, &info->block_groups[index], list) {
5326                 spin_lock(&cache->lock);
5327                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5328                        "%llu pinned %llu reserved\n",
5329                        (unsigned long long)cache->key.objectid,
5330                        (unsigned long long)cache->key.offset,
5331                        (unsigned long long)btrfs_block_group_used(&cache->item),
5332                        (unsigned long long)cache->pinned,
5333                        (unsigned long long)cache->reserved);
5334                 btrfs_dump_free_space(cache, bytes);
5335                 spin_unlock(&cache->lock);
5336         }
5337         if (++index < BTRFS_NR_RAID_TYPES)
5338                 goto again;
5339         up_read(&info->groups_sem);
5340 }
5341
5342 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5343                          struct btrfs_root *root,
5344                          u64 num_bytes, u64 min_alloc_size,
5345                          u64 empty_size, u64 hint_byte,
5346                          u64 search_end, struct btrfs_key *ins,
5347                          u64 data)
5348 {
5349         int ret;
5350         u64 search_start = 0;
5351
5352         data = btrfs_get_alloc_profile(root, data);
5353 again:
5354         /*
5355          * the only place that sets empty_size is btrfs_realloc_node, which
5356          * is not called recursively on allocations
5357          */
5358         if (empty_size || root->ref_cows)
5359                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5360                                      num_bytes + 2 * 1024 * 1024, data,
5361                                      CHUNK_ALLOC_NO_FORCE);
5362
5363         WARN_ON(num_bytes < root->sectorsize);
5364         ret = find_free_extent(trans, root, num_bytes, empty_size,
5365                                search_start, search_end, hint_byte,
5366                                ins, data);
5367
5368         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5369                 num_bytes = num_bytes >> 1;
5370                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5371                 num_bytes = max(num_bytes, min_alloc_size);
5372                 do_chunk_alloc(trans, root->fs_info->extent_root,
5373                                num_bytes, data, CHUNK_ALLOC_FORCE);
5374                 goto again;
5375         }
5376         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5377                 struct btrfs_space_info *sinfo;
5378
5379                 sinfo = __find_space_info(root->fs_info, data);
5380                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5381                        "wanted %llu\n", (unsigned long long)data,
5382                        (unsigned long long)num_bytes);
5383                 dump_space_info(sinfo, num_bytes, 1);
5384         }
5385
5386         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5387
5388         return ret;
5389 }
5390
5391 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5392 {
5393         struct btrfs_block_group_cache *cache;
5394         int ret = 0;
5395
5396         cache = btrfs_lookup_block_group(root->fs_info, start);
5397         if (!cache) {
5398                 printk(KERN_ERR "Unable to find block group for %llu\n",
5399                        (unsigned long long)start);
5400                 return -ENOSPC;
5401         }
5402
5403         if (btrfs_test_opt(root, DISCARD))
5404                 ret = btrfs_discard_extent(root, start, len, NULL);
5405
5406         btrfs_add_free_space(cache, start, len);
5407         btrfs_update_reserved_bytes(cache, len, 0, 1);
5408         btrfs_put_block_group(cache);
5409
5410         trace_btrfs_reserved_extent_free(root, start, len);
5411
5412         return ret;
5413 }
5414
5415 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5416                                       struct btrfs_root *root,
5417                                       u64 parent, u64 root_objectid,
5418                                       u64 flags, u64 owner, u64 offset,
5419                                       struct btrfs_key *ins, int ref_mod)
5420 {
5421         int ret;
5422         struct btrfs_fs_info *fs_info = root->fs_info;
5423         struct btrfs_extent_item *extent_item;
5424         struct btrfs_extent_inline_ref *iref;
5425         struct btrfs_path *path;
5426         struct extent_buffer *leaf;
5427         int type;
5428         u32 size;
5429
5430         if (parent > 0)
5431                 type = BTRFS_SHARED_DATA_REF_KEY;
5432         else
5433                 type = BTRFS_EXTENT_DATA_REF_KEY;
5434
5435         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5436
5437         path = btrfs_alloc_path();
5438         if (!path)
5439                 return -ENOMEM;
5440
5441         path->leave_spinning = 1;
5442         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5443                                       ins, size);
5444         BUG_ON(ret);
5445
5446         leaf = path->nodes[0];
5447         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5448                                      struct btrfs_extent_item);
5449         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5450         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5451         btrfs_set_extent_flags(leaf, extent_item,
5452                                flags | BTRFS_EXTENT_FLAG_DATA);
5453
5454         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5455         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5456         if (parent > 0) {
5457                 struct btrfs_shared_data_ref *ref;
5458                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5459                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5460                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5461         } else {
5462                 struct btrfs_extent_data_ref *ref;
5463                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5464                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5465                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5466                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5467                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5468         }
5469
5470         btrfs_mark_buffer_dirty(path->nodes[0]);
5471         btrfs_free_path(path);
5472
5473         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5474         if (ret) {
5475                 printk(KERN_ERR "btrfs update block group failed for %llu "
5476                        "%llu\n", (unsigned long long)ins->objectid,
5477                        (unsigned long long)ins->offset);
5478                 BUG();
5479         }
5480         return ret;
5481 }
5482
5483 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5484                                      struct btrfs_root *root,
5485                                      u64 parent, u64 root_objectid,
5486                                      u64 flags, struct btrfs_disk_key *key,
5487                                      int level, struct btrfs_key *ins)
5488 {
5489         int ret;
5490         struct btrfs_fs_info *fs_info = root->fs_info;
5491         struct btrfs_extent_item *extent_item;
5492         struct btrfs_tree_block_info *block_info;
5493         struct btrfs_extent_inline_ref *iref;
5494         struct btrfs_path *path;
5495         struct extent_buffer *leaf;
5496         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5497
5498         path = btrfs_alloc_path();
5499         BUG_ON(!path);
5500
5501         path->leave_spinning = 1;
5502         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5503                                       ins, size);
5504         BUG_ON(ret);
5505
5506         leaf = path->nodes[0];
5507         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5508                                      struct btrfs_extent_item);
5509         btrfs_set_extent_refs(leaf, extent_item, 1);
5510         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5511         btrfs_set_extent_flags(leaf, extent_item,
5512                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5513         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5514
5515         btrfs_set_tree_block_key(leaf, block_info, key);
5516         btrfs_set_tree_block_level(leaf, block_info, level);
5517
5518         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5519         if (parent > 0) {
5520                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5521                 btrfs_set_extent_inline_ref_type(leaf, iref,
5522                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5523                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5524         } else {
5525                 btrfs_set_extent_inline_ref_type(leaf, iref,
5526                                                  BTRFS_TREE_BLOCK_REF_KEY);
5527                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5528         }
5529
5530         btrfs_mark_buffer_dirty(leaf);
5531         btrfs_free_path(path);
5532
5533         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5534         if (ret) {
5535                 printk(KERN_ERR "btrfs update block group failed for %llu "
5536                        "%llu\n", (unsigned long long)ins->objectid,
5537                        (unsigned long long)ins->offset);
5538                 BUG();
5539         }
5540         return ret;
5541 }
5542
5543 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5544                                      struct btrfs_root *root,
5545                                      u64 root_objectid, u64 owner,
5546                                      u64 offset, struct btrfs_key *ins)
5547 {
5548         int ret;
5549
5550         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5551
5552         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5553                                          0, root_objectid, owner, offset,
5554                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5555         return ret;
5556 }
5557
5558 /*
5559  * this is used by the tree logging recovery code.  It records that
5560  * an extent has been allocated and makes sure to clear the free
5561  * space cache bits as well
5562  */
5563 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5564                                    struct btrfs_root *root,
5565                                    u64 root_objectid, u64 owner, u64 offset,
5566                                    struct btrfs_key *ins)
5567 {
5568         int ret;
5569         struct btrfs_block_group_cache *block_group;
5570         struct btrfs_caching_control *caching_ctl;
5571         u64 start = ins->objectid;
5572         u64 num_bytes = ins->offset;
5573
5574         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5575         cache_block_group(block_group, trans, NULL, 0);
5576         caching_ctl = get_caching_control(block_group);
5577
5578         if (!caching_ctl) {
5579                 BUG_ON(!block_group_cache_done(block_group));
5580                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5581                 BUG_ON(ret);
5582         } else {
5583                 mutex_lock(&caching_ctl->mutex);
5584
5585                 if (start >= caching_ctl->progress) {
5586                         ret = add_excluded_extent(root, start, num_bytes);
5587                         BUG_ON(ret);
5588                 } else if (start + num_bytes <= caching_ctl->progress) {
5589                         ret = btrfs_remove_free_space(block_group,
5590                                                       start, num_bytes);
5591                         BUG_ON(ret);
5592                 } else {
5593                         num_bytes = caching_ctl->progress - start;
5594                         ret = btrfs_remove_free_space(block_group,
5595                                                       start, num_bytes);
5596                         BUG_ON(ret);
5597
5598                         start = caching_ctl->progress;
5599                         num_bytes = ins->objectid + ins->offset -
5600                                     caching_ctl->progress;
5601                         ret = add_excluded_extent(root, start, num_bytes);
5602                         BUG_ON(ret);
5603                 }
5604
5605                 mutex_unlock(&caching_ctl->mutex);
5606                 put_caching_control(caching_ctl);
5607         }
5608
5609         ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1);
5610         BUG_ON(ret);
5611         btrfs_put_block_group(block_group);
5612         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5613                                          0, owner, offset, ins, 1);
5614         return ret;
5615 }
5616
5617 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5618                                             struct btrfs_root *root,
5619                                             u64 bytenr, u32 blocksize,
5620                                             int level)
5621 {
5622         struct extent_buffer *buf;
5623
5624         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5625         if (!buf)
5626                 return ERR_PTR(-ENOMEM);
5627         btrfs_set_header_generation(buf, trans->transid);
5628         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5629         btrfs_tree_lock(buf);
5630         clean_tree_block(trans, root, buf);
5631
5632         btrfs_set_lock_blocking(buf);
5633         btrfs_set_buffer_uptodate(buf);
5634
5635         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5636                 /*
5637                  * we allow two log transactions at a time, use different
5638                  * EXENT bit to differentiate dirty pages.
5639                  */
5640                 if (root->log_transid % 2 == 0)
5641                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5642                                         buf->start + buf->len - 1, GFP_NOFS);
5643                 else
5644                         set_extent_new(&root->dirty_log_pages, buf->start,
5645                                         buf->start + buf->len - 1, GFP_NOFS);
5646         } else {
5647                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5648                          buf->start + buf->len - 1, GFP_NOFS);
5649         }
5650         trans->blocks_used++;
5651         /* this returns a buffer locked for blocking */
5652         return buf;
5653 }
5654
5655 static struct btrfs_block_rsv *
5656 use_block_rsv(struct btrfs_trans_handle *trans,
5657               struct btrfs_root *root, u32 blocksize)
5658 {
5659         struct btrfs_block_rsv *block_rsv;
5660         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5661         int ret;
5662
5663         block_rsv = get_block_rsv(trans, root);
5664
5665         if (block_rsv->size == 0) {
5666                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5667                                              blocksize, 0);
5668                 /*
5669                  * If we couldn't reserve metadata bytes try and use some from
5670                  * the global reserve.
5671                  */
5672                 if (ret && block_rsv != global_rsv) {
5673                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5674                         if (!ret)
5675                                 return global_rsv;
5676                         return ERR_PTR(ret);
5677                 } else if (ret) {
5678                         return ERR_PTR(ret);
5679                 }
5680                 return block_rsv;
5681         }
5682
5683         ret = block_rsv_use_bytes(block_rsv, blocksize);
5684         if (!ret)
5685                 return block_rsv;
5686         if (ret) {
5687                 WARN_ON(1);
5688                 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5689                                              0);
5690                 if (!ret) {
5691                         spin_lock(&block_rsv->lock);
5692                         block_rsv->size += blocksize;
5693                         spin_unlock(&block_rsv->lock);
5694                         return block_rsv;
5695                 } else if (ret && block_rsv != global_rsv) {
5696                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5697                         if (!ret)
5698                                 return global_rsv;
5699                 }
5700         }
5701
5702         return ERR_PTR(-ENOSPC);
5703 }
5704
5705 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5706 {
5707         block_rsv_add_bytes(block_rsv, blocksize, 0);
5708         block_rsv_release_bytes(block_rsv, NULL, 0);
5709 }
5710
5711 /*
5712  * finds a free extent and does all the dirty work required for allocation
5713  * returns the key for the extent through ins, and a tree buffer for
5714  * the first block of the extent through buf.
5715  *
5716  * returns the tree buffer or NULL.
5717  */
5718 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5719                                         struct btrfs_root *root, u32 blocksize,
5720                                         u64 parent, u64 root_objectid,
5721                                         struct btrfs_disk_key *key, int level,
5722                                         u64 hint, u64 empty_size)
5723 {
5724         struct btrfs_key ins;
5725         struct btrfs_block_rsv *block_rsv;
5726         struct extent_buffer *buf;
5727         u64 flags = 0;
5728         int ret;
5729
5730
5731         block_rsv = use_block_rsv(trans, root, blocksize);
5732         if (IS_ERR(block_rsv))
5733                 return ERR_CAST(block_rsv);
5734
5735         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5736                                    empty_size, hint, (u64)-1, &ins, 0);
5737         if (ret) {
5738                 unuse_block_rsv(block_rsv, blocksize);
5739                 return ERR_PTR(ret);
5740         }
5741
5742         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5743                                     blocksize, level);
5744         BUG_ON(IS_ERR(buf));
5745
5746         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5747                 if (parent == 0)
5748                         parent = ins.objectid;
5749                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5750         } else
5751                 BUG_ON(parent > 0);
5752
5753         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5754                 struct btrfs_delayed_extent_op *extent_op;
5755                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5756                 BUG_ON(!extent_op);
5757                 if (key)
5758                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5759                 else
5760                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5761                 extent_op->flags_to_set = flags;
5762                 extent_op->update_key = 1;
5763                 extent_op->update_flags = 1;
5764                 extent_op->is_data = 0;
5765
5766                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5767                                         ins.offset, parent, root_objectid,
5768                                         level, BTRFS_ADD_DELAYED_EXTENT,
5769                                         extent_op);
5770                 BUG_ON(ret);
5771         }
5772         return buf;
5773 }
5774
5775 struct walk_control {
5776         u64 refs[BTRFS_MAX_LEVEL];
5777         u64 flags[BTRFS_MAX_LEVEL];
5778         struct btrfs_key update_progress;
5779         int stage;
5780         int level;
5781         int shared_level;
5782         int update_ref;
5783         int keep_locks;
5784         int reada_slot;
5785         int reada_count;
5786 };
5787
5788 #define DROP_REFERENCE  1
5789 #define UPDATE_BACKREF  2
5790
5791 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5792                                      struct btrfs_root *root,
5793                                      struct walk_control *wc,
5794                                      struct btrfs_path *path)
5795 {
5796         u64 bytenr;
5797         u64 generation;
5798         u64 refs;
5799         u64 flags;
5800         u32 nritems;
5801         u32 blocksize;
5802         struct btrfs_key key;
5803         struct extent_buffer *eb;
5804         int ret;
5805         int slot;
5806         int nread = 0;
5807
5808         if (path->slots[wc->level] < wc->reada_slot) {
5809                 wc->reada_count = wc->reada_count * 2 / 3;
5810                 wc->reada_count = max(wc->reada_count, 2);
5811         } else {
5812                 wc->reada_count = wc->reada_count * 3 / 2;
5813                 wc->reada_count = min_t(int, wc->reada_count,
5814                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5815         }
5816
5817         eb = path->nodes[wc->level];
5818         nritems = btrfs_header_nritems(eb);
5819         blocksize = btrfs_level_size(root, wc->level - 1);
5820
5821         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5822                 if (nread >= wc->reada_count)
5823                         break;
5824
5825                 cond_resched();
5826                 bytenr = btrfs_node_blockptr(eb, slot);
5827                 generation = btrfs_node_ptr_generation(eb, slot);
5828
5829                 if (slot == path->slots[wc->level])
5830                         goto reada;
5831
5832                 if (wc->stage == UPDATE_BACKREF &&
5833                     generation <= root->root_key.offset)
5834                         continue;
5835
5836                 /* We don't lock the tree block, it's OK to be racy here */
5837                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5838                                                &refs, &flags);
5839                 BUG_ON(ret);
5840                 BUG_ON(refs == 0);
5841
5842                 if (wc->stage == DROP_REFERENCE) {
5843                         if (refs == 1)
5844                                 goto reada;
5845
5846                         if (wc->level == 1 &&
5847                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5848                                 continue;
5849                         if (!wc->update_ref ||
5850                             generation <= root->root_key.offset)
5851                                 continue;
5852                         btrfs_node_key_to_cpu(eb, &key, slot);
5853                         ret = btrfs_comp_cpu_keys(&key,
5854                                                   &wc->update_progress);
5855                         if (ret < 0)
5856                                 continue;
5857                 } else {
5858                         if (wc->level == 1 &&
5859                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5860                                 continue;
5861                 }
5862 reada:
5863                 ret = readahead_tree_block(root, bytenr, blocksize,
5864                                            generation);
5865                 if (ret)
5866                         break;
5867                 nread++;
5868         }
5869         wc->reada_slot = slot;
5870 }
5871
5872 /*
5873  * hepler to process tree block while walking down the tree.
5874  *
5875  * when wc->stage == UPDATE_BACKREF, this function updates
5876  * back refs for pointers in the block.
5877  *
5878  * NOTE: return value 1 means we should stop walking down.
5879  */
5880 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5881                                    struct btrfs_root *root,
5882                                    struct btrfs_path *path,
5883                                    struct walk_control *wc, int lookup_info)
5884 {
5885         int level = wc->level;
5886         struct extent_buffer *eb = path->nodes[level];
5887         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5888         int ret;
5889
5890         if (wc->stage == UPDATE_BACKREF &&
5891             btrfs_header_owner(eb) != root->root_key.objectid)
5892                 return 1;
5893
5894         /*
5895          * when reference count of tree block is 1, it won't increase
5896          * again. once full backref flag is set, we never clear it.
5897          */
5898         if (lookup_info &&
5899             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5900              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5901                 BUG_ON(!path->locks[level]);
5902                 ret = btrfs_lookup_extent_info(trans, root,
5903                                                eb->start, eb->len,
5904                                                &wc->refs[level],
5905                                                &wc->flags[level]);
5906                 BUG_ON(ret);
5907                 BUG_ON(wc->refs[level] == 0);
5908         }
5909
5910         if (wc->stage == DROP_REFERENCE) {
5911                 if (wc->refs[level] > 1)
5912                         return 1;
5913
5914                 if (path->locks[level] && !wc->keep_locks) {
5915                         btrfs_tree_unlock_rw(eb, path->locks[level]);
5916                         path->locks[level] = 0;
5917                 }
5918                 return 0;
5919         }
5920
5921         /* wc->stage == UPDATE_BACKREF */
5922         if (!(wc->flags[level] & flag)) {
5923                 BUG_ON(!path->locks[level]);
5924                 ret = btrfs_inc_ref(trans, root, eb, 1);
5925                 BUG_ON(ret);
5926                 ret = btrfs_dec_ref(trans, root, eb, 0);
5927                 BUG_ON(ret);
5928                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5929                                                   eb->len, flag, 0);
5930                 BUG_ON(ret);
5931                 wc->flags[level] |= flag;
5932         }
5933
5934         /*
5935          * the block is shared by multiple trees, so it's not good to
5936          * keep the tree lock
5937          */
5938         if (path->locks[level] && level > 0) {
5939                 btrfs_tree_unlock_rw(eb, path->locks[level]);
5940                 path->locks[level] = 0;
5941         }
5942         return 0;
5943 }
5944
5945 /*
5946  * hepler to process tree block pointer.
5947  *
5948  * when wc->stage == DROP_REFERENCE, this function checks
5949  * reference count of the block pointed to. if the block
5950  * is shared and we need update back refs for the subtree
5951  * rooted at the block, this function changes wc->stage to
5952  * UPDATE_BACKREF. if the block is shared and there is no
5953  * need to update back, this function drops the reference
5954  * to the block.
5955  *
5956  * NOTE: return value 1 means we should stop walking down.
5957  */
5958 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5959                                  struct btrfs_root *root,
5960                                  struct btrfs_path *path,
5961                                  struct walk_control *wc, int *lookup_info)
5962 {
5963         u64 bytenr;
5964         u64 generation;
5965         u64 parent;
5966         u32 blocksize;
5967         struct btrfs_key key;
5968         struct extent_buffer *next;
5969         int level = wc->level;
5970         int reada = 0;
5971         int ret = 0;
5972
5973         generation = btrfs_node_ptr_generation(path->nodes[level],
5974                                                path->slots[level]);
5975         /*
5976          * if the lower level block was created before the snapshot
5977          * was created, we know there is no need to update back refs
5978          * for the subtree
5979          */
5980         if (wc->stage == UPDATE_BACKREF &&
5981             generation <= root->root_key.offset) {
5982                 *lookup_info = 1;
5983                 return 1;
5984         }
5985
5986         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5987         blocksize = btrfs_level_size(root, level - 1);
5988
5989         next = btrfs_find_tree_block(root, bytenr, blocksize);
5990         if (!next) {
5991                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5992                 if (!next)
5993                         return -ENOMEM;
5994                 reada = 1;
5995         }
5996         btrfs_tree_lock(next);
5997         btrfs_set_lock_blocking(next);
5998
5999         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6000                                        &wc->refs[level - 1],
6001                                        &wc->flags[level - 1]);
6002         BUG_ON(ret);
6003         BUG_ON(wc->refs[level - 1] == 0);
6004         *lookup_info = 0;
6005
6006         if (wc->stage == DROP_REFERENCE) {
6007                 if (wc->refs[level - 1] > 1) {
6008                         if (level == 1 &&
6009                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6010                                 goto skip;
6011
6012                         if (!wc->update_ref ||
6013                             generation <= root->root_key.offset)
6014                                 goto skip;
6015
6016                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6017                                               path->slots[level]);
6018                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6019                         if (ret < 0)
6020                                 goto skip;
6021
6022                         wc->stage = UPDATE_BACKREF;
6023                         wc->shared_level = level - 1;
6024                 }
6025         } else {
6026                 if (level == 1 &&
6027                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6028                         goto skip;
6029         }
6030
6031         if (!btrfs_buffer_uptodate(next, generation)) {
6032                 btrfs_tree_unlock(next);
6033                 free_extent_buffer(next);
6034                 next = NULL;
6035                 *lookup_info = 1;
6036         }
6037
6038         if (!next) {
6039                 if (reada && level == 1)
6040                         reada_walk_down(trans, root, wc, path);
6041                 next = read_tree_block(root, bytenr, blocksize, generation);
6042                 if (!next)
6043                         return -EIO;
6044                 btrfs_tree_lock(next);
6045                 btrfs_set_lock_blocking(next);
6046         }
6047
6048         level--;
6049         BUG_ON(level != btrfs_header_level(next));
6050         path->nodes[level] = next;
6051         path->slots[level] = 0;
6052         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6053         wc->level = level;
6054         if (wc->level == 1)
6055                 wc->reada_slot = 0;
6056         return 0;
6057 skip:
6058         wc->refs[level - 1] = 0;
6059         wc->flags[level - 1] = 0;
6060         if (wc->stage == DROP_REFERENCE) {
6061                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6062                         parent = path->nodes[level]->start;
6063                 } else {
6064                         BUG_ON(root->root_key.objectid !=
6065                                btrfs_header_owner(path->nodes[level]));
6066                         parent = 0;
6067                 }
6068
6069                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6070                                         root->root_key.objectid, level - 1, 0);
6071                 BUG_ON(ret);
6072         }
6073         btrfs_tree_unlock(next);
6074         free_extent_buffer(next);
6075         *lookup_info = 1;
6076         return 1;
6077 }
6078
6079 /*
6080  * hepler to process tree block while walking up the tree.
6081  *
6082  * when wc->stage == DROP_REFERENCE, this function drops
6083  * reference count on the block.
6084  *
6085  * when wc->stage == UPDATE_BACKREF, this function changes
6086  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6087  * to UPDATE_BACKREF previously while processing the block.
6088  *
6089  * NOTE: return value 1 means we should stop walking up.
6090  */
6091 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6092                                  struct btrfs_root *root,
6093                                  struct btrfs_path *path,
6094                                  struct walk_control *wc)
6095 {
6096         int ret;
6097         int level = wc->level;
6098         struct extent_buffer *eb = path->nodes[level];
6099         u64 parent = 0;
6100
6101         if (wc->stage == UPDATE_BACKREF) {
6102                 BUG_ON(wc->shared_level < level);
6103                 if (level < wc->shared_level)
6104                         goto out;
6105
6106                 ret = find_next_key(path, level + 1, &wc->update_progress);
6107                 if (ret > 0)
6108                         wc->update_ref = 0;
6109
6110                 wc->stage = DROP_REFERENCE;
6111                 wc->shared_level = -1;
6112                 path->slots[level] = 0;
6113
6114                 /*
6115                  * check reference count again if the block isn't locked.
6116                  * we should start walking down the tree again if reference
6117                  * count is one.
6118                  */
6119                 if (!path->locks[level]) {
6120                         BUG_ON(level == 0);
6121                         btrfs_tree_lock(eb);
6122                         btrfs_set_lock_blocking(eb);
6123                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6124
6125                         ret = btrfs_lookup_extent_info(trans, root,
6126                                                        eb->start, eb->len,
6127                                                        &wc->refs[level],
6128                                                        &wc->flags[level]);
6129                         BUG_ON(ret);
6130                         BUG_ON(wc->refs[level] == 0);
6131                         if (wc->refs[level] == 1) {
6132                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6133                                 return 1;
6134                         }
6135                 }
6136         }
6137
6138         /* wc->stage == DROP_REFERENCE */
6139         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6140
6141         if (wc->refs[level] == 1) {
6142                 if (level == 0) {
6143                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6144                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6145                         else
6146                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6147                         BUG_ON(ret);
6148                 }
6149                 /* make block locked assertion in clean_tree_block happy */
6150                 if (!path->locks[level] &&
6151                     btrfs_header_generation(eb) == trans->transid) {
6152                         btrfs_tree_lock(eb);
6153                         btrfs_set_lock_blocking(eb);
6154                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6155                 }
6156                 clean_tree_block(trans, root, eb);
6157         }
6158
6159         if (eb == root->node) {
6160                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6161                         parent = eb->start;
6162                 else
6163                         BUG_ON(root->root_key.objectid !=
6164                                btrfs_header_owner(eb));
6165         } else {
6166                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6167                         parent = path->nodes[level + 1]->start;
6168                 else
6169                         BUG_ON(root->root_key.objectid !=
6170                                btrfs_header_owner(path->nodes[level + 1]));
6171         }
6172
6173         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6174 out:
6175         wc->refs[level] = 0;
6176         wc->flags[level] = 0;
6177         return 0;
6178 }
6179
6180 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6181                                    struct btrfs_root *root,
6182                                    struct btrfs_path *path,
6183                                    struct walk_control *wc)
6184 {
6185         int level = wc->level;
6186         int lookup_info = 1;
6187         int ret;
6188
6189         while (level >= 0) {
6190                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6191                 if (ret > 0)
6192                         break;
6193
6194                 if (level == 0)
6195                         break;
6196
6197                 if (path->slots[level] >=
6198                     btrfs_header_nritems(path->nodes[level]))
6199                         break;
6200
6201                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6202                 if (ret > 0) {
6203                         path->slots[level]++;
6204                         continue;
6205                 } else if (ret < 0)
6206                         return ret;
6207                 level = wc->level;
6208         }
6209         return 0;
6210 }
6211
6212 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6213                                  struct btrfs_root *root,
6214                                  struct btrfs_path *path,
6215                                  struct walk_control *wc, int max_level)
6216 {
6217         int level = wc->level;
6218         int ret;
6219
6220         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6221         while (level < max_level && path->nodes[level]) {
6222                 wc->level = level;
6223                 if (path->slots[level] + 1 <
6224                     btrfs_header_nritems(path->nodes[level])) {
6225                         path->slots[level]++;
6226                         return 0;
6227                 } else {
6228                         ret = walk_up_proc(trans, root, path, wc);
6229                         if (ret > 0)
6230                                 return 0;
6231
6232                         if (path->locks[level]) {
6233                                 btrfs_tree_unlock_rw(path->nodes[level],
6234                                                      path->locks[level]);
6235                                 path->locks[level] = 0;
6236                         }
6237                         free_extent_buffer(path->nodes[level]);
6238                         path->nodes[level] = NULL;
6239                         level++;
6240                 }
6241         }
6242         return 1;
6243 }
6244
6245 /*
6246  * drop a subvolume tree.
6247  *
6248  * this function traverses the tree freeing any blocks that only
6249  * referenced by the tree.
6250  *
6251  * when a shared tree block is found. this function decreases its
6252  * reference count by one. if update_ref is true, this function
6253  * also make sure backrefs for the shared block and all lower level
6254  * blocks are properly updated.
6255  */
6256 int btrfs_drop_snapshot(struct btrfs_root *root,
6257                         struct btrfs_block_rsv *block_rsv, int update_ref)
6258 {
6259         struct btrfs_path *path;
6260         struct btrfs_trans_handle *trans;
6261         struct btrfs_root *tree_root = root->fs_info->tree_root;
6262         struct btrfs_root_item *root_item = &root->root_item;
6263         struct walk_control *wc;
6264         struct btrfs_key key;
6265         int err = 0;
6266         int ret;
6267         int level;
6268
6269         path = btrfs_alloc_path();
6270         BUG_ON(!path);
6271
6272         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6273         BUG_ON(!wc);
6274
6275         trans = btrfs_start_transaction(tree_root, 0);
6276         BUG_ON(IS_ERR(trans));
6277
6278         if (block_rsv)
6279                 trans->block_rsv = block_rsv;
6280
6281         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6282                 level = btrfs_header_level(root->node);
6283                 path->nodes[level] = btrfs_lock_root_node(root);
6284                 btrfs_set_lock_blocking(path->nodes[level]);
6285                 path->slots[level] = 0;
6286                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6287                 memset(&wc->update_progress, 0,
6288                        sizeof(wc->update_progress));
6289         } else {
6290                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6291                 memcpy(&wc->update_progress, &key,
6292                        sizeof(wc->update_progress));
6293
6294                 level = root_item->drop_level;
6295                 BUG_ON(level == 0);
6296                 path->lowest_level = level;
6297                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6298                 path->lowest_level = 0;
6299                 if (ret < 0) {
6300                         err = ret;
6301                         goto out;
6302                 }
6303                 WARN_ON(ret > 0);
6304
6305                 /*
6306                  * unlock our path, this is safe because only this
6307                  * function is allowed to delete this snapshot
6308                  */
6309                 btrfs_unlock_up_safe(path, 0);
6310
6311                 level = btrfs_header_level(root->node);
6312                 while (1) {
6313                         btrfs_tree_lock(path->nodes[level]);
6314                         btrfs_set_lock_blocking(path->nodes[level]);
6315
6316                         ret = btrfs_lookup_extent_info(trans, root,
6317                                                 path->nodes[level]->start,
6318                                                 path->nodes[level]->len,
6319                                                 &wc->refs[level],
6320                                                 &wc->flags[level]);
6321                         BUG_ON(ret);
6322                         BUG_ON(wc->refs[level] == 0);
6323
6324                         if (level == root_item->drop_level)
6325                                 break;
6326
6327                         btrfs_tree_unlock(path->nodes[level]);
6328                         WARN_ON(wc->refs[level] != 1);
6329                         level--;
6330                 }
6331         }
6332
6333         wc->level = level;
6334         wc->shared_level = -1;
6335         wc->stage = DROP_REFERENCE;
6336         wc->update_ref = update_ref;
6337         wc->keep_locks = 0;
6338         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6339
6340         while (1) {
6341                 ret = walk_down_tree(trans, root, path, wc);
6342                 if (ret < 0) {
6343                         err = ret;
6344                         break;
6345                 }
6346
6347                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6348                 if (ret < 0) {
6349                         err = ret;
6350                         break;
6351                 }
6352
6353                 if (ret > 0) {
6354                         BUG_ON(wc->stage != DROP_REFERENCE);
6355                         break;
6356                 }
6357
6358                 if (wc->stage == DROP_REFERENCE) {
6359                         level = wc->level;
6360                         btrfs_node_key(path->nodes[level],
6361                                        &root_item->drop_progress,
6362                                        path->slots[level]);
6363                         root_item->drop_level = level;
6364                 }
6365
6366                 BUG_ON(wc->level == 0);
6367                 if (btrfs_should_end_transaction(trans, tree_root)) {
6368                         ret = btrfs_update_root(trans, tree_root,
6369                                                 &root->root_key,
6370                                                 root_item);
6371                         BUG_ON(ret);
6372
6373                         btrfs_end_transaction_throttle(trans, tree_root);
6374                         trans = btrfs_start_transaction(tree_root, 0);
6375                         BUG_ON(IS_ERR(trans));
6376                         if (block_rsv)
6377                                 trans->block_rsv = block_rsv;
6378                 }
6379         }
6380         btrfs_release_path(path);
6381         BUG_ON(err);
6382
6383         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6384         BUG_ON(ret);
6385
6386         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6387                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6388                                            NULL, NULL);
6389                 BUG_ON(ret < 0);
6390                 if (ret > 0) {
6391                         /* if we fail to delete the orphan item this time
6392                          * around, it'll get picked up the next time.
6393                          *
6394                          * The most common failure here is just -ENOENT.
6395                          */
6396                         btrfs_del_orphan_item(trans, tree_root,
6397                                               root->root_key.objectid);
6398                 }
6399         }
6400
6401         if (root->in_radix) {
6402                 btrfs_free_fs_root(tree_root->fs_info, root);
6403         } else {
6404                 free_extent_buffer(root->node);
6405                 free_extent_buffer(root->commit_root);
6406                 kfree(root);
6407         }
6408 out:
6409         btrfs_end_transaction_throttle(trans, tree_root);
6410         kfree(wc);
6411         btrfs_free_path(path);
6412         return err;
6413 }
6414
6415 /*
6416  * drop subtree rooted at tree block 'node'.
6417  *
6418  * NOTE: this function will unlock and release tree block 'node'
6419  */
6420 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6421                         struct btrfs_root *root,
6422                         struct extent_buffer *node,
6423                         struct extent_buffer *parent)
6424 {
6425         struct btrfs_path *path;
6426         struct walk_control *wc;
6427         int level;
6428         int parent_level;
6429         int ret = 0;
6430         int wret;
6431
6432         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6433
6434         path = btrfs_alloc_path();
6435         if (!path)
6436                 return -ENOMEM;
6437
6438         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6439         if (!wc) {
6440                 btrfs_free_path(path);
6441                 return -ENOMEM;
6442         }
6443
6444         btrfs_assert_tree_locked(parent);
6445         parent_level = btrfs_header_level(parent);
6446         extent_buffer_get(parent);
6447         path->nodes[parent_level] = parent;
6448         path->slots[parent_level] = btrfs_header_nritems(parent);
6449
6450         btrfs_assert_tree_locked(node);
6451         level = btrfs_header_level(node);
6452         path->nodes[level] = node;
6453         path->slots[level] = 0;
6454         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6455
6456         wc->refs[parent_level] = 1;
6457         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6458         wc->level = level;
6459         wc->shared_level = -1;
6460         wc->stage = DROP_REFERENCE;
6461         wc->update_ref = 0;
6462         wc->keep_locks = 1;
6463         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6464
6465         while (1) {
6466                 wret = walk_down_tree(trans, root, path, wc);
6467                 if (wret < 0) {
6468                         ret = wret;
6469                         break;
6470                 }
6471
6472                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6473                 if (wret < 0)
6474                         ret = wret;
6475                 if (wret != 0)
6476                         break;
6477         }
6478
6479         kfree(wc);
6480         btrfs_free_path(path);
6481         return ret;
6482 }
6483
6484 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6485 {
6486         u64 num_devices;
6487         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6488                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6489
6490         /*
6491          * we add in the count of missing devices because we want
6492          * to make sure that any RAID levels on a degraded FS
6493          * continue to be honored.
6494          */
6495         num_devices = root->fs_info->fs_devices->rw_devices +
6496                 root->fs_info->fs_devices->missing_devices;
6497
6498         if (num_devices == 1) {
6499                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6500                 stripped = flags & ~stripped;
6501
6502                 /* turn raid0 into single device chunks */
6503                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6504                         return stripped;
6505
6506                 /* turn mirroring into duplication */
6507                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6508                              BTRFS_BLOCK_GROUP_RAID10))
6509                         return stripped | BTRFS_BLOCK_GROUP_DUP;
6510                 return flags;
6511         } else {
6512                 /* they already had raid on here, just return */
6513                 if (flags & stripped)
6514                         return flags;
6515
6516                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6517                 stripped = flags & ~stripped;
6518
6519                 /* switch duplicated blocks with raid1 */
6520                 if (flags & BTRFS_BLOCK_GROUP_DUP)
6521                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
6522
6523                 /* turn single device chunks into raid0 */
6524                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6525         }
6526         return flags;
6527 }
6528
6529 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6530 {
6531         struct btrfs_space_info *sinfo = cache->space_info;
6532         u64 num_bytes;
6533         u64 min_allocable_bytes;
6534         int ret = -ENOSPC;
6535
6536         if (cache->ro)
6537                 return 0;
6538
6539         /*
6540          * We need some metadata space and system metadata space for
6541          * allocating chunks in some corner cases until we force to set
6542          * it to be readonly.
6543          */
6544         if ((sinfo->flags &
6545              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6546             !force)
6547                 min_allocable_bytes = 1 * 1024 * 1024;
6548         else
6549                 min_allocable_bytes = 0;
6550
6551         spin_lock(&sinfo->lock);
6552         spin_lock(&cache->lock);
6553         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6554                     cache->bytes_super - btrfs_block_group_used(&cache->item);
6555
6556         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6557             sinfo->bytes_may_use + sinfo->bytes_readonly +
6558             cache->reserved_pinned + num_bytes + min_allocable_bytes <=
6559             sinfo->total_bytes) {
6560                 sinfo->bytes_readonly += num_bytes;
6561                 sinfo->bytes_reserved += cache->reserved_pinned;
6562                 cache->reserved_pinned = 0;
6563                 cache->ro = 1;
6564                 ret = 0;
6565         }
6566
6567         spin_unlock(&cache->lock);
6568         spin_unlock(&sinfo->lock);
6569         return ret;
6570 }
6571
6572 int btrfs_set_block_group_ro(struct btrfs_root *root,
6573                              struct btrfs_block_group_cache *cache)
6574
6575 {
6576         struct btrfs_trans_handle *trans;
6577         u64 alloc_flags;
6578         int ret;
6579
6580         BUG_ON(cache->ro);
6581
6582         trans = btrfs_join_transaction(root);
6583         BUG_ON(IS_ERR(trans));
6584
6585         alloc_flags = update_block_group_flags(root, cache->flags);
6586         if (alloc_flags != cache->flags)
6587                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6588                                CHUNK_ALLOC_FORCE);
6589
6590         ret = set_block_group_ro(cache, 0);
6591         if (!ret)
6592                 goto out;
6593         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6594         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6595                              CHUNK_ALLOC_FORCE);
6596         if (ret < 0)
6597                 goto out;
6598         ret = set_block_group_ro(cache, 0);
6599 out:
6600         btrfs_end_transaction(trans, root);
6601         return ret;
6602 }
6603
6604 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6605                             struct btrfs_root *root, u64 type)
6606 {
6607         u64 alloc_flags = get_alloc_profile(root, type);
6608         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6609                               CHUNK_ALLOC_FORCE);
6610 }
6611
6612 /*
6613  * helper to account the unused space of all the readonly block group in the
6614  * list. takes mirrors into account.
6615  */
6616 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6617 {
6618         struct btrfs_block_group_cache *block_group;
6619         u64 free_bytes = 0;
6620         int factor;
6621
6622         list_for_each_entry(block_group, groups_list, list) {
6623                 spin_lock(&block_group->lock);
6624
6625                 if (!block_group->ro) {
6626                         spin_unlock(&block_group->lock);
6627                         continue;
6628                 }
6629
6630                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6631                                           BTRFS_BLOCK_GROUP_RAID10 |
6632                                           BTRFS_BLOCK_GROUP_DUP))
6633                         factor = 2;
6634                 else
6635                         factor = 1;
6636
6637                 free_bytes += (block_group->key.offset -
6638                                btrfs_block_group_used(&block_group->item)) *
6639                                factor;
6640
6641                 spin_unlock(&block_group->lock);
6642         }
6643
6644         return free_bytes;
6645 }
6646
6647 /*
6648  * helper to account the unused space of all the readonly block group in the
6649  * space_info. takes mirrors into account.
6650  */
6651 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6652 {
6653         int i;
6654         u64 free_bytes = 0;
6655
6656         spin_lock(&sinfo->lock);
6657
6658         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6659                 if (!list_empty(&sinfo->block_groups[i]))
6660                         free_bytes += __btrfs_get_ro_block_group_free_space(
6661                                                 &sinfo->block_groups[i]);
6662
6663         spin_unlock(&sinfo->lock);
6664
6665         return free_bytes;
6666 }
6667
6668 int btrfs_set_block_group_rw(struct btrfs_root *root,
6669                               struct btrfs_block_group_cache *cache)
6670 {
6671         struct btrfs_space_info *sinfo = cache->space_info;
6672         u64 num_bytes;
6673
6674         BUG_ON(!cache->ro);
6675
6676         spin_lock(&sinfo->lock);
6677         spin_lock(&cache->lock);
6678         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6679                     cache->bytes_super - btrfs_block_group_used(&cache->item);
6680         sinfo->bytes_readonly -= num_bytes;
6681         cache->ro = 0;
6682         spin_unlock(&cache->lock);
6683         spin_unlock(&sinfo->lock);
6684         return 0;
6685 }
6686
6687 /*
6688  * checks to see if its even possible to relocate this block group.
6689  *
6690  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6691  * ok to go ahead and try.
6692  */
6693 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6694 {
6695         struct btrfs_block_group_cache *block_group;
6696         struct btrfs_space_info *space_info;
6697         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6698         struct btrfs_device *device;
6699         int full = 0;
6700         int ret = 0;
6701
6702         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6703
6704         /* odd, couldn't find the block group, leave it alone */
6705         if (!block_group)
6706                 return -1;
6707
6708         /* no bytes used, we're good */
6709         if (!btrfs_block_group_used(&block_group->item))
6710                 goto out;
6711
6712         space_info = block_group->space_info;
6713         spin_lock(&space_info->lock);
6714
6715         full = space_info->full;
6716
6717         /*
6718          * if this is the last block group we have in this space, we can't
6719          * relocate it unless we're able to allocate a new chunk below.
6720          *
6721          * Otherwise, we need to make sure we have room in the space to handle
6722          * all of the extents from this block group.  If we can, we're good
6723          */
6724         if ((space_info->total_bytes != block_group->key.offset) &&
6725            (space_info->bytes_used + space_info->bytes_reserved +
6726             space_info->bytes_pinned + space_info->bytes_readonly +
6727             btrfs_block_group_used(&block_group->item) <
6728             space_info->total_bytes)) {
6729                 spin_unlock(&space_info->lock);
6730                 goto out;
6731         }
6732         spin_unlock(&space_info->lock);
6733
6734         /*
6735          * ok we don't have enough space, but maybe we have free space on our
6736          * devices to allocate new chunks for relocation, so loop through our
6737          * alloc devices and guess if we have enough space.  However, if we
6738          * were marked as full, then we know there aren't enough chunks, and we
6739          * can just return.
6740          */
6741         ret = -1;
6742         if (full)
6743                 goto out;
6744
6745         mutex_lock(&root->fs_info->chunk_mutex);
6746         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
6747                 u64 min_free = btrfs_block_group_used(&block_group->item);
6748                 u64 dev_offset;
6749
6750                 /*
6751                  * check to make sure we can actually find a chunk with enough
6752                  * space to fit our block group in.
6753                  */
6754                 if (device->total_bytes > device->bytes_used + min_free) {
6755                         ret = find_free_dev_extent(NULL, device, min_free,
6756                                                    &dev_offset, NULL);
6757                         if (!ret)
6758                                 break;
6759                         ret = -1;
6760                 }
6761         }
6762         mutex_unlock(&root->fs_info->chunk_mutex);
6763 out:
6764         btrfs_put_block_group(block_group);
6765         return ret;
6766 }
6767
6768 static int find_first_block_group(struct btrfs_root *root,
6769                 struct btrfs_path *path, struct btrfs_key *key)
6770 {
6771         int ret = 0;
6772         struct btrfs_key found_key;
6773         struct extent_buffer *leaf;
6774         int slot;
6775
6776         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6777         if (ret < 0)
6778                 goto out;
6779
6780         while (1) {
6781                 slot = path->slots[0];
6782                 leaf = path->nodes[0];
6783                 if (slot >= btrfs_header_nritems(leaf)) {
6784                         ret = btrfs_next_leaf(root, path);
6785                         if (ret == 0)
6786                                 continue;
6787                         if (ret < 0)
6788                                 goto out;
6789                         break;
6790                 }
6791                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6792
6793                 if (found_key.objectid >= key->objectid &&
6794                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6795                         ret = 0;
6796                         goto out;
6797                 }
6798                 path->slots[0]++;
6799         }
6800 out:
6801         return ret;
6802 }
6803
6804 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
6805 {
6806         struct btrfs_block_group_cache *block_group;
6807         u64 last = 0;
6808
6809         while (1) {
6810                 struct inode *inode;
6811
6812                 block_group = btrfs_lookup_first_block_group(info, last);
6813                 while (block_group) {
6814                         spin_lock(&block_group->lock);
6815                         if (block_group->iref)
6816                                 break;
6817                         spin_unlock(&block_group->lock);
6818                         block_group = next_block_group(info->tree_root,
6819                                                        block_group);
6820                 }
6821                 if (!block_group) {
6822                         if (last == 0)
6823                                 break;
6824                         last = 0;
6825                         continue;
6826                 }
6827
6828                 inode = block_group->inode;
6829                 block_group->iref = 0;
6830                 block_group->inode = NULL;
6831                 spin_unlock(&block_group->lock);
6832                 iput(inode);
6833                 last = block_group->key.objectid + block_group->key.offset;
6834                 btrfs_put_block_group(block_group);
6835         }
6836 }
6837
6838 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6839 {
6840         struct btrfs_block_group_cache *block_group;
6841         struct btrfs_space_info *space_info;
6842         struct btrfs_caching_control *caching_ctl;
6843         struct rb_node *n;
6844
6845         down_write(&info->extent_commit_sem);
6846         while (!list_empty(&info->caching_block_groups)) {
6847                 caching_ctl = list_entry(info->caching_block_groups.next,
6848                                          struct btrfs_caching_control, list);
6849                 list_del(&caching_ctl->list);
6850                 put_caching_control(caching_ctl);
6851         }
6852         up_write(&info->extent_commit_sem);
6853
6854         spin_lock(&info->block_group_cache_lock);
6855         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6856                 block_group = rb_entry(n, struct btrfs_block_group_cache,
6857                                        cache_node);
6858                 rb_erase(&block_group->cache_node,
6859                          &info->block_group_cache_tree);
6860                 spin_unlock(&info->block_group_cache_lock);
6861
6862                 down_write(&block_group->space_info->groups_sem);
6863                 list_del(&block_group->list);
6864                 up_write(&block_group->space_info->groups_sem);
6865
6866                 if (block_group->cached == BTRFS_CACHE_STARTED)
6867                         wait_block_group_cache_done(block_group);
6868
6869                 /*
6870                  * We haven't cached this block group, which means we could
6871                  * possibly have excluded extents on this block group.
6872                  */
6873                 if (block_group->cached == BTRFS_CACHE_NO)
6874                         free_excluded_extents(info->extent_root, block_group);
6875
6876                 btrfs_remove_free_space_cache(block_group);
6877                 btrfs_put_block_group(block_group);
6878
6879                 spin_lock(&info->block_group_cache_lock);
6880         }
6881         spin_unlock(&info->block_group_cache_lock);
6882
6883         /* now that all the block groups are freed, go through and
6884          * free all the space_info structs.  This is only called during
6885          * the final stages of unmount, and so we know nobody is
6886          * using them.  We call synchronize_rcu() once before we start,
6887          * just to be on the safe side.
6888          */
6889         synchronize_rcu();
6890
6891         release_global_block_rsv(info);
6892
6893         while(!list_empty(&info->space_info)) {
6894                 space_info = list_entry(info->space_info.next,
6895                                         struct btrfs_space_info,
6896                                         list);
6897                 if (space_info->bytes_pinned > 0 ||
6898                     space_info->bytes_reserved > 0) {
6899                         WARN_ON(1);
6900                         dump_space_info(space_info, 0, 0);
6901                 }
6902                 list_del(&space_info->list);
6903                 kfree(space_info);
6904         }
6905         return 0;
6906 }
6907
6908 static void __link_block_group(struct btrfs_space_info *space_info,
6909                                struct btrfs_block_group_cache *cache)
6910 {
6911         int index = get_block_group_index(cache);
6912
6913         down_write(&space_info->groups_sem);
6914         list_add_tail(&cache->list, &space_info->block_groups[index]);
6915         up_write(&space_info->groups_sem);
6916 }
6917
6918 int btrfs_read_block_groups(struct btrfs_root *root)
6919 {
6920         struct btrfs_path *path;
6921         int ret;
6922         struct btrfs_block_group_cache *cache;
6923         struct btrfs_fs_info *info = root->fs_info;
6924         struct btrfs_space_info *space_info;
6925         struct btrfs_key key;
6926         struct btrfs_key found_key;
6927         struct extent_buffer *leaf;
6928         int need_clear = 0;
6929         u64 cache_gen;
6930
6931         root = info->extent_root;
6932         key.objectid = 0;
6933         key.offset = 0;
6934         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
6935         path = btrfs_alloc_path();
6936         if (!path)
6937                 return -ENOMEM;
6938         path->reada = 1;
6939
6940         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
6941         if (cache_gen != 0 &&
6942             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
6943                 need_clear = 1;
6944         if (btrfs_test_opt(root, CLEAR_CACHE))
6945                 need_clear = 1;
6946         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
6947                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
6948
6949         while (1) {
6950                 ret = find_first_block_group(root, path, &key);
6951                 if (ret > 0)
6952                         break;
6953                 if (ret != 0)
6954                         goto error;
6955                 leaf = path->nodes[0];
6956                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6957                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
6958                 if (!cache) {
6959                         ret = -ENOMEM;
6960                         goto error;
6961                 }
6962                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
6963                                                 GFP_NOFS);
6964                 if (!cache->free_space_ctl) {
6965                         kfree(cache);
6966                         ret = -ENOMEM;
6967                         goto error;
6968                 }
6969
6970                 atomic_set(&cache->count, 1);
6971                 spin_lock_init(&cache->lock);
6972                 cache->fs_info = info;
6973                 INIT_LIST_HEAD(&cache->list);
6974                 INIT_LIST_HEAD(&cache->cluster_list);
6975
6976                 if (need_clear)
6977                         cache->disk_cache_state = BTRFS_DC_CLEAR;
6978
6979                 read_extent_buffer(leaf, &cache->item,
6980                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
6981                                    sizeof(cache->item));
6982                 memcpy(&cache->key, &found_key, sizeof(found_key));
6983
6984                 key.objectid = found_key.objectid + found_key.offset;
6985                 btrfs_release_path(path);
6986                 cache->flags = btrfs_block_group_flags(&cache->item);
6987                 cache->sectorsize = root->sectorsize;
6988
6989                 btrfs_init_free_space_ctl(cache);
6990
6991                 /*
6992                  * We need to exclude the super stripes now so that the space
6993                  * info has super bytes accounted for, otherwise we'll think
6994                  * we have more space than we actually do.
6995                  */
6996                 exclude_super_stripes(root, cache);
6997
6998                 /*
6999                  * check for two cases, either we are full, and therefore
7000                  * don't need to bother with the caching work since we won't
7001                  * find any space, or we are empty, and we can just add all
7002                  * the space in and be done with it.  This saves us _alot_ of
7003                  * time, particularly in the full case.
7004                  */
7005                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7006                         cache->last_byte_to_unpin = (u64)-1;
7007                         cache->cached = BTRFS_CACHE_FINISHED;
7008                         free_excluded_extents(root, cache);
7009                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7010                         cache->last_byte_to_unpin = (u64)-1;
7011                         cache->cached = BTRFS_CACHE_FINISHED;
7012                         add_new_free_space(cache, root->fs_info,
7013                                            found_key.objectid,
7014                                            found_key.objectid +
7015                                            found_key.offset);
7016                         free_excluded_extents(root, cache);
7017                 }
7018
7019                 ret = update_space_info(info, cache->flags, found_key.offset,
7020                                         btrfs_block_group_used(&cache->item),
7021                                         &space_info);
7022                 BUG_ON(ret);
7023                 cache->space_info = space_info;
7024                 spin_lock(&cache->space_info->lock);
7025                 cache->space_info->bytes_readonly += cache->bytes_super;
7026                 spin_unlock(&cache->space_info->lock);
7027
7028                 __link_block_group(space_info, cache);
7029
7030                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7031                 BUG_ON(ret);
7032
7033                 set_avail_alloc_bits(root->fs_info, cache->flags);
7034                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7035                         set_block_group_ro(cache, 1);
7036         }
7037
7038         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7039                 if (!(get_alloc_profile(root, space_info->flags) &
7040                       (BTRFS_BLOCK_GROUP_RAID10 |
7041                        BTRFS_BLOCK_GROUP_RAID1 |
7042                        BTRFS_BLOCK_GROUP_DUP)))
7043                         continue;
7044                 /*
7045                  * avoid allocating from un-mirrored block group if there are
7046                  * mirrored block groups.
7047                  */
7048                 list_for_each_entry(cache, &space_info->block_groups[3], list)
7049                         set_block_group_ro(cache, 1);
7050                 list_for_each_entry(cache, &space_info->block_groups[4], list)
7051                         set_block_group_ro(cache, 1);
7052         }
7053
7054         init_global_block_rsv(info);
7055         ret = 0;
7056 error:
7057         btrfs_free_path(path);
7058         return ret;
7059 }
7060
7061 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7062                            struct btrfs_root *root, u64 bytes_used,
7063                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7064                            u64 size)
7065 {
7066         int ret;
7067         struct btrfs_root *extent_root;
7068         struct btrfs_block_group_cache *cache;
7069
7070         extent_root = root->fs_info->extent_root;
7071
7072         root->fs_info->last_trans_log_full_commit = trans->transid;
7073
7074         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7075         if (!cache)
7076                 return -ENOMEM;
7077         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7078                                         GFP_NOFS);
7079         if (!cache->free_space_ctl) {
7080                 kfree(cache);
7081                 return -ENOMEM;
7082         }
7083
7084         cache->key.objectid = chunk_offset;
7085         cache->key.offset = size;
7086         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7087         cache->sectorsize = root->sectorsize;
7088         cache->fs_info = root->fs_info;
7089
7090         atomic_set(&cache->count, 1);
7091         spin_lock_init(&cache->lock);
7092         INIT_LIST_HEAD(&cache->list);
7093         INIT_LIST_HEAD(&cache->cluster_list);
7094
7095         btrfs_init_free_space_ctl(cache);
7096
7097         btrfs_set_block_group_used(&cache->item, bytes_used);
7098         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7099         cache->flags = type;
7100         btrfs_set_block_group_flags(&cache->item, type);
7101
7102         cache->last_byte_to_unpin = (u64)-1;
7103         cache->cached = BTRFS_CACHE_FINISHED;
7104         exclude_super_stripes(root, cache);
7105
7106         add_new_free_space(cache, root->fs_info, chunk_offset,
7107                            chunk_offset + size);
7108
7109         free_excluded_extents(root, cache);
7110
7111         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7112                                 &cache->space_info);
7113         BUG_ON(ret);
7114
7115         spin_lock(&cache->space_info->lock);
7116         cache->space_info->bytes_readonly += cache->bytes_super;
7117         spin_unlock(&cache->space_info->lock);
7118
7119         __link_block_group(cache->space_info, cache);
7120
7121         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7122         BUG_ON(ret);
7123
7124         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7125                                 sizeof(cache->item));
7126         BUG_ON(ret);
7127
7128         set_avail_alloc_bits(extent_root->fs_info, type);
7129
7130         return 0;
7131 }
7132
7133 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7134                              struct btrfs_root *root, u64 group_start)
7135 {
7136         struct btrfs_path *path;
7137         struct btrfs_block_group_cache *block_group;
7138         struct btrfs_free_cluster *cluster;
7139         struct btrfs_root *tree_root = root->fs_info->tree_root;
7140         struct btrfs_key key;
7141         struct inode *inode;
7142         int ret;
7143         int factor;
7144
7145         root = root->fs_info->extent_root;
7146
7147         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7148         BUG_ON(!block_group);
7149         BUG_ON(!block_group->ro);
7150
7151         /*
7152          * Free the reserved super bytes from this block group before
7153          * remove it.
7154          */
7155         free_excluded_extents(root, block_group);
7156
7157         memcpy(&key, &block_group->key, sizeof(key));
7158         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7159                                   BTRFS_BLOCK_GROUP_RAID1 |
7160                                   BTRFS_BLOCK_GROUP_RAID10))
7161                 factor = 2;
7162         else
7163                 factor = 1;
7164
7165         /* make sure this block group isn't part of an allocation cluster */
7166         cluster = &root->fs_info->data_alloc_cluster;
7167         spin_lock(&cluster->refill_lock);
7168         btrfs_return_cluster_to_free_space(block_group, cluster);
7169         spin_unlock(&cluster->refill_lock);
7170
7171         /*
7172          * make sure this block group isn't part of a metadata
7173          * allocation cluster
7174          */
7175         cluster = &root->fs_info->meta_alloc_cluster;
7176         spin_lock(&cluster->refill_lock);
7177         btrfs_return_cluster_to_free_space(block_group, cluster);
7178         spin_unlock(&cluster->refill_lock);
7179
7180         path = btrfs_alloc_path();
7181         BUG_ON(!path);
7182
7183         inode = lookup_free_space_inode(root, block_group, path);
7184         if (!IS_ERR(inode)) {
7185                 btrfs_orphan_add(trans, inode);
7186                 clear_nlink(inode);
7187                 /* One for the block groups ref */
7188                 spin_lock(&block_group->lock);
7189                 if (block_group->iref) {
7190                         block_group->iref = 0;
7191                         block_group->inode = NULL;
7192                         spin_unlock(&block_group->lock);
7193                         iput(inode);
7194                 } else {
7195                         spin_unlock(&block_group->lock);
7196                 }
7197                 /* One for our lookup ref */
7198                 iput(inode);
7199         }
7200
7201         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7202         key.offset = block_group->key.objectid;
7203         key.type = 0;
7204
7205         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7206         if (ret < 0)
7207                 goto out;
7208         if (ret > 0)
7209                 btrfs_release_path(path);
7210         if (ret == 0) {
7211                 ret = btrfs_del_item(trans, tree_root, path);
7212                 if (ret)
7213                         goto out;
7214                 btrfs_release_path(path);
7215         }
7216
7217         spin_lock(&root->fs_info->block_group_cache_lock);
7218         rb_erase(&block_group->cache_node,
7219                  &root->fs_info->block_group_cache_tree);
7220         spin_unlock(&root->fs_info->block_group_cache_lock);
7221
7222         down_write(&block_group->space_info->groups_sem);
7223         /*
7224          * we must use list_del_init so people can check to see if they
7225          * are still on the list after taking the semaphore
7226          */
7227         list_del_init(&block_group->list);
7228         up_write(&block_group->space_info->groups_sem);
7229
7230         if (block_group->cached == BTRFS_CACHE_STARTED)
7231                 wait_block_group_cache_done(block_group);
7232
7233         btrfs_remove_free_space_cache(block_group);
7234
7235         spin_lock(&block_group->space_info->lock);
7236         block_group->space_info->total_bytes -= block_group->key.offset;
7237         block_group->space_info->bytes_readonly -= block_group->key.offset;
7238         block_group->space_info->disk_total -= block_group->key.offset * factor;
7239         spin_unlock(&block_group->space_info->lock);
7240
7241         memcpy(&key, &block_group->key, sizeof(key));
7242
7243         btrfs_clear_space_info_full(root->fs_info);
7244
7245         btrfs_put_block_group(block_group);
7246         btrfs_put_block_group(block_group);
7247
7248         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7249         if (ret > 0)
7250                 ret = -EIO;
7251         if (ret < 0)
7252                 goto out;
7253
7254         ret = btrfs_del_item(trans, root, path);
7255 out:
7256         btrfs_free_path(path);
7257         return ret;
7258 }
7259
7260 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7261 {
7262         struct btrfs_space_info *space_info;
7263         struct btrfs_super_block *disk_super;
7264         u64 features;
7265         u64 flags;
7266         int mixed = 0;
7267         int ret;
7268
7269         disk_super = &fs_info->super_copy;
7270         if (!btrfs_super_root(disk_super))
7271                 return 1;
7272
7273         features = btrfs_super_incompat_flags(disk_super);
7274         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7275                 mixed = 1;
7276
7277         flags = BTRFS_BLOCK_GROUP_SYSTEM;
7278         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7279         if (ret)
7280                 goto out;
7281
7282         if (mixed) {
7283                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7284                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7285         } else {
7286                 flags = BTRFS_BLOCK_GROUP_METADATA;
7287                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7288                 if (ret)
7289                         goto out;
7290
7291                 flags = BTRFS_BLOCK_GROUP_DATA;
7292                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7293         }
7294 out:
7295         return ret;
7296 }
7297
7298 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7299 {
7300         return unpin_extent_range(root, start, end);
7301 }
7302
7303 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7304                                u64 num_bytes, u64 *actual_bytes)
7305 {
7306         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7307 }
7308
7309 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7310 {
7311         struct btrfs_fs_info *fs_info = root->fs_info;
7312         struct btrfs_block_group_cache *cache = NULL;
7313         u64 group_trimmed;
7314         u64 start;
7315         u64 end;
7316         u64 trimmed = 0;
7317         int ret = 0;
7318
7319         cache = btrfs_lookup_block_group(fs_info, range->start);
7320
7321         while (cache) {
7322                 if (cache->key.objectid >= (range->start + range->len)) {
7323                         btrfs_put_block_group(cache);
7324                         break;
7325                 }
7326
7327                 start = max(range->start, cache->key.objectid);
7328                 end = min(range->start + range->len,
7329                                 cache->key.objectid + cache->key.offset);
7330
7331                 if (end - start >= range->minlen) {
7332                         if (!block_group_cache_done(cache)) {
7333                                 ret = cache_block_group(cache, NULL, root, 0);
7334                                 if (!ret)
7335                                         wait_block_group_cache_done(cache);
7336                         }
7337                         ret = btrfs_trim_block_group(cache,
7338                                                      &group_trimmed,
7339                                                      start,
7340                                                      end,
7341                                                      range->minlen);
7342
7343                         trimmed += group_trimmed;
7344                         if (ret) {
7345                                 btrfs_put_block_group(cache);
7346                                 break;
7347                         }
7348                 }
7349
7350                 cache = next_block_group(fs_info->tree_root, cache);
7351         }
7352
7353         range->len = trimmed;
7354         return ret;
7355 }