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