fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/kthread.h>
  25 #include <linux/slab.h>
  26 #include "compat.h"
  27 #include "hash.h"
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "print-tree.h"
  31 #include "transaction.h"
  32 #include "volumes.h"
  33 #include "locking.h"
  34 #include "free-space-cache.h"
  35
  36 /* control flags for do_chunk_alloc's force field
  37  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
  38  * if we really need one.
  39  *
  40  * CHUNK_ALLOC_FORCE means it must try to allocate one
  41  *
  42  * CHUNK_ALLOC_LIMITED means to only try and allocate one
  43  * if we have very few chunks already allocated.  This is
  44  * used as part of the clustering code to help make sure
  45  * we have a good pool of storage to cluster in, without
  46  * filling the FS with empty chunks
  47  *
  48  */
  49 enum {
  50         CHUNK_ALLOC_NO_FORCE = 0,
  51         CHUNK_ALLOC_FORCE = 1,
  52         CHUNK_ALLOC_LIMITED = 2,
  53 };
  54
  55 static int update_block_group(struct btrfs_trans_handle *trans,
  56                               struct btrfs_root *root,
  57                               u64 bytenr, u64 num_bytes, int alloc);
  58 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  59                                 struct btrfs_root *root,
  60                                 u64 bytenr, u64 num_bytes, u64 parent,
  61                                 u64 root_objectid, u64 owner_objectid,
  62                                 u64 owner_offset, int refs_to_drop,
  63                                 struct btrfs_delayed_extent_op *extra_op);
  64 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  65                                     struct extent_buffer *leaf,
  66                                     struct btrfs_extent_item *ei);
  67 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  68                                       struct btrfs_root *root,
  69                                       u64 parent, u64 root_objectid,
  70                                       u64 flags, u64 owner, u64 offset,
  71                                       struct btrfs_key *ins, int ref_mod);
  72 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  73                                      struct btrfs_root *root,
  74                                      u64 parent, u64 root_objectid,
  75                                      u64 flags, struct btrfs_disk_key *key,
  76                                      int level, struct btrfs_key *ins);
  77 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  78                           struct btrfs_root *extent_root, u64 alloc_bytes,
  79                           u64 flags, int force);
  80 static int find_next_key(struct btrfs_path *path, int level,
  81                          struct btrfs_key *key);
  82 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  83                             int dump_block_groups);
  84
  85 static noinline int
  86 block_group_cache_done(struct btrfs_block_group_cache *cache)
  87 {
  88         smp_mb();
  89         return cache->cached == BTRFS_CACHE_FINISHED;
  90 }
  91
  92 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  93 {
  94         return (cache->flags & bits) == bits;
  95 }
  96
  97 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
  98 {
  99         atomic_inc(&cache->count);
 100 }
 101
 102 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 103 {
 104         if (atomic_dec_and_test(&cache->count)) {
 105                 WARN_ON(cache->pinned > 0);
 106                 WARN_ON(cache->reserved > 0);
 107                 WARN_ON(cache->reserved_pinned > 0);
 108                 kfree(cache->free_space_ctl);
 109                 kfree(cache);
 110         }
 111 }
 112
 113 /*
 114  * this adds the block group to the fs_info rb tree for the block group
 115  * cache
 116  */
 117 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 118                                 struct btrfs_block_group_cache *block_group)
 119 {
 120         struct rb_node **p;
 121         struct rb_node *parent = NULL;
 122         struct btrfs_block_group_cache *cache;
 123
 124         spin_lock(&info->block_group_cache_lock);
 125         p = &info->block_group_cache_tree.rb_node;
 126
 127         while (*p) {
 128                 parent = *p;
 129                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 130                                  cache_node);
 131                 if (block_group->key.objectid < cache->key.objectid) {
 132                         p = &(*p)->rb_left;
 133                 } else if (block_group->key.objectid > cache->key.objectid) {
 134                         p = &(*p)->rb_right;
 135                 } else {
 136                         spin_unlock(&info->block_group_cache_lock);
 137                         return -EEXIST;
 138                 }
 139         }
 140
 141         rb_link_node(&block_group->cache_node, parent, p);
 142         rb_insert_color(&block_group->cache_node,
 143                         &info->block_group_cache_tree);
 144         spin_unlock(&info->block_group_cache_lock);
 145
 146         return 0;
 147 }
 148
 149 /*
 150  * This will return the block group at or after bytenr if contains is 0, else
 151  * it will return the block group that contains the bytenr
 152  */
 153 static struct btrfs_block_group_cache *
 154 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 155                               int contains)
 156 {
 157         struct btrfs_block_group_cache *cache, *ret = NULL;
 158         struct rb_node *n;
 159         u64 end, start;
 160
 161         spin_lock(&info->block_group_cache_lock);
 162         n = info->block_group_cache_tree.rb_node;
 163
 164         while (n) {
 165                 cache = rb_entry(n, struct btrfs_block_group_cache,
 166                                  cache_node);
 167                 end = cache->key.objectid + cache->key.offset - 1;
 168                 start = cache->key.objectid;
 169
 170                 if (bytenr < start) {
 171                         if (!contains && (!ret || start < ret->key.objectid))
 172                                 ret = cache;
 173                         n = n->rb_left;
 174                 } else if (bytenr > start) {
 175                         if (contains && bytenr <= end) {
 176                                 ret = cache;
 177                                 break;
 178                         }
 179                         n = n->rb_right;
 180                 } else {
 181                         ret = cache;
 182                         break;
 183                 }
 184         }
 185         if (ret)
 186                 btrfs_get_block_group(ret);
 187         spin_unlock(&info->block_group_cache_lock);
 188
 189         return ret;
 190 }
 191
 192 static int add_excluded_extent(struct btrfs_root *root,
 193                                u64 start, u64 num_bytes)
 194 {
 195         u64 end = start + num_bytes - 1;
 196         set_extent_bits(&root->fs_info->freed_extents[0],
 197                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 198         set_extent_bits(&root->fs_info->freed_extents[1],
 199                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 200         return 0;
 201 }
 202
 203 static void free_excluded_extents(struct btrfs_root *root,
 204                                   struct btrfs_block_group_cache *cache)
 205 {
 206         u64 start, end;
 207
 208         start = cache->key.objectid;
 209         end = start + cache->key.offset - 1;
 210
 211         clear_extent_bits(&root->fs_info->freed_extents[0],
 212                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 213         clear_extent_bits(&root->fs_info->freed_extents[1],
 214                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 215 }
 216
 217 static int exclude_super_stripes(struct btrfs_root *root,
 218                                  struct btrfs_block_group_cache *cache)
 219 {
 220         u64 bytenr;
 221         u64 *logical;
 222         int stripe_len;
 223         int i, nr, ret;
 224
 225         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 226                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 227                 cache->bytes_super += stripe_len;
 228                 ret = add_excluded_extent(root, cache->key.objectid,
 229                                           stripe_len);
 230                 BUG_ON(ret);
 231         }
 232
 233         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 234                 bytenr = btrfs_sb_offset(i);
 235                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 236                                        cache->key.objectid, bytenr,
 237                                        0, &logical, &nr, &stripe_len);
 238                 BUG_ON(ret);
 239
 240                 while (nr--) {
 241                         cache->bytes_super += stripe_len;
 242                         ret = add_excluded_extent(root, logical[nr],
 243                                                   stripe_len);
 244                         BUG_ON(ret);
 245                 }
 246
 247                 kfree(logical);
 248         }
 249         return 0;
 250 }
 251
 252 static struct btrfs_caching_control *
 253 get_caching_control(struct btrfs_block_group_cache *cache)
 254 {
 255         struct btrfs_caching_control *ctl;
 256
 257         spin_lock(&cache->lock);
 258         if (cache->cached != BTRFS_CACHE_STARTED) {
 259                 spin_unlock(&cache->lock);
 260                 return NULL;
 261         }
 262
 263         /* We're loading it the fast way, so we don't have a caching_ctl. */
 264         if (!cache->caching_ctl) {
 265                 spin_unlock(&cache->lock);
 266                 return NULL;
 267         }
 268
 269         ctl = cache->caching_ctl;
 270         atomic_inc(&ctl->count);
 271         spin_unlock(&cache->lock);
 272         return ctl;
 273 }
 274
 275 static void put_caching_control(struct btrfs_caching_control *ctl)
 276 {
 277         if (atomic_dec_and_test(&ctl->count))
 278                 kfree(ctl);
 279 }
 280
 281 /*
 282  * this is only called by cache_block_group, since we could have freed extents
 283  * we need to check the pinned_extents for any extents that can't be used yet
 284  * since their free space will be released as soon as the transaction commits.
 285  */
 286 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 287                               struct btrfs_fs_info *info, u64 start, u64 end)
 288 {
 289         u64 extent_start, extent_end, size, total_added = 0;
 290         int ret;
 291
 292         while (start < end) {
 293                 ret = find_first_extent_bit(info->pinned_extents, start,
 294                                             &extent_start, &extent_end,
 295                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 296                 if (ret)
 297                         break;
 298
 299                 if (extent_start <= start) {
 300                         start = extent_end + 1;
 301                 } else if (extent_start > start && extent_start < end) {
 302                         size = extent_start - start;
 303                         total_added += size;
 304                         ret = btrfs_add_free_space(block_group, start,
 305                                                    size);
 306                         BUG_ON(ret);
 307                         start = extent_end + 1;
 308                 } else {
 309                         break;
 310                 }
 311         }
 312
 313         if (start < end) {
 314                 size = end - start;
 315                 total_added += size;
 316                 ret = btrfs_add_free_space(block_group, start, size);
 317                 BUG_ON(ret);
 318         }
 319
 320         return total_added;
 321 }
 322
 323 static noinline void caching_thread(struct btrfs_work *work)
 324 {
 325         struct btrfs_block_group_cache *block_group;
 326         struct btrfs_fs_info *fs_info;
 327         struct btrfs_caching_control *caching_ctl;
 328         struct btrfs_root *extent_root;
 329         struct btrfs_path *path;
 330         struct extent_buffer *leaf;
 331         struct btrfs_key key;
 332         u64 total_found = 0;
 333         u64 last = 0;
 334         u32 nritems;
 335         int ret = 0;
 336
 337         caching_ctl = container_of(work, struct btrfs_caching_control, work);
 338         block_group = caching_ctl->block_group;
 339         fs_info = block_group->fs_info;
 340         extent_root = fs_info->extent_root;
 341
 342         path = btrfs_alloc_path();
 343         if (!path)
 344                 goto out;
 345
 346         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 347
 348         /*
 349          * We don't want to deadlock with somebody trying to allocate a new
 350          * extent for the extent root while also trying to search the extent
 351          * root to add free space.  So we skip locking and search the commit
 352          * root, since its read-only
 353          */
 354         path->skip_locking = 1;
 355         path->search_commit_root = 1;
 356         path->reada = 1;
 357
 358         key.objectid = last;
 359         key.offset = 0;
 360         key.type = BTRFS_EXTENT_ITEM_KEY;
 361 again:
 362         mutex_lock(&caching_ctl->mutex);
 363         /* need to make sure the commit_root doesn't disappear */
 364         down_read(&fs_info->extent_commit_sem);
 365
 366         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 367         if (ret < 0)
 368                 goto err;
 369
 370         leaf = path->nodes[0];
 371         nritems = btrfs_header_nritems(leaf);
 372
 373         while (1) {
 374                 if (btrfs_fs_closing(fs_info) > 1) {
 375                         last = (u64)-1;
 376                         break;
 377                 }
 378
 379                 if (path->slots[0] < nritems) {
 380                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 381                 } else {
 382                         ret = find_next_key(path, 0, &key);
 383                         if (ret)
 384                                 break;
 385
 386                         if (need_resched() ||
 387                             btrfs_next_leaf(extent_root, path)) {
 388                                 caching_ctl->progress = last;
 389                                 btrfs_release_path(path);
 390                                 up_read(&fs_info->extent_commit_sem);
 391                                 mutex_unlock(&caching_ctl->mutex);
 392                                 cond_resched();
 393                                 goto again;
 394                         }
 395                         leaf = path->nodes[0];
 396                         nritems = btrfs_header_nritems(leaf);
 397                         continue;
 398                 }
 399
 400                 if (key.objectid < block_group->key.objectid) {
 401                         path->slots[0]++;
 402                         continue;
 403                 }
 404
 405                 if (key.objectid >= block_group->key.objectid +
 406                     block_group->key.offset)
 407                         break;
 408
 409                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 410                         total_found += add_new_free_space(block_group,
 411                                                           fs_info, last,
 412                                                           key.objectid);
 413                         last = key.objectid + key.offset;
 414
 415                         if (total_found > (1024 * 1024 * 2)) {
 416                                 total_found = 0;
 417                                 wake_up(&caching_ctl->wait);
 418                         }
 419                 }
 420                 path->slots[0]++;
 421         }
 422         ret = 0;
 423
 424         total_found += add_new_free_space(block_group, fs_info, last,
 425                                           block_group->key.objectid +
 426                                           block_group->key.offset);
 427         caching_ctl->progress = (u64)-1;
 428
 429         spin_lock(&block_group->lock);
 430         block_group->caching_ctl = NULL;
 431         block_group->cached = BTRFS_CACHE_FINISHED;
 432         spin_unlock(&block_group->lock);
 433
 434 err:
 435         btrfs_free_path(path);
 436         up_read(&fs_info->extent_commit_sem);
 437
 438         free_excluded_extents(extent_root, block_group);
 439
 440         mutex_unlock(&caching_ctl->mutex);
 441 out:
 442         wake_up(&caching_ctl->wait);
 443
 444         put_caching_control(caching_ctl);
 445         btrfs_put_block_group(block_group);
 446 }
 447
 448 static int cache_block_group(struct btrfs_block_group_cache *cache,
 449                              struct btrfs_trans_handle *trans,
 450                              struct btrfs_root *root,
 451                              int load_cache_only)
 452 {
 453         struct btrfs_fs_info *fs_info = cache->fs_info;
 454         struct btrfs_caching_control *caching_ctl;
 455         int ret = 0;
 456
 457         smp_mb();
 458         if (cache->cached != BTRFS_CACHE_NO)
 459                 return 0;
 460
 461         /*
 462          * We can't do the read from on-disk cache during a commit since we need
 463          * to have the normal tree locking.  Also if we are currently trying to
 464          * allocate blocks for the tree root we can't do the fast caching since
 465          * we likely hold important locks.
 466          */
 467         if (trans && (!trans->transaction->in_commit) &&
 468             (root && root != root->fs_info->tree_root)) {
 469                 spin_lock(&cache->lock);
 470                 if (cache->cached != BTRFS_CACHE_NO) {
 471                         spin_unlock(&cache->lock);
 472                         return 0;
 473                 }
 474                 cache->cached = BTRFS_CACHE_STARTED;
 475                 spin_unlock(&cache->lock);
 476
 477                 ret = load_free_space_cache(fs_info, cache);
 478
 479                 spin_lock(&cache->lock);
 480                 if (ret == 1) {
 481                         cache->cached = BTRFS_CACHE_FINISHED;
 482                         cache->last_byte_to_unpin = (u64)-1;
 483                 } else {
 484                         cache->cached = BTRFS_CACHE_NO;
 485                 }
 486                 spin_unlock(&cache->lock);
 487                 if (ret == 1) {
 488                         free_excluded_extents(fs_info->extent_root, cache);
 489                         return 0;
 490                 }
 491         }
 492
 493         if (load_cache_only)
 494                 return 0;
 495
 496         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 497         BUG_ON(!caching_ctl);
 498
 499         INIT_LIST_HEAD(&caching_ctl->list);
 500         mutex_init(&caching_ctl->mutex);
 501         init_waitqueue_head(&caching_ctl->wait);
 502         caching_ctl->block_group = cache;
 503         caching_ctl->progress = cache->key.objectid;
 504         /* one for caching kthread, one for caching block group list */
 505         atomic_set(&caching_ctl->count, 2);
 506         caching_ctl->work.func = caching_thread;
 507
 508         spin_lock(&cache->lock);
 509         if (cache->cached != BTRFS_CACHE_NO) {
 510                 spin_unlock(&cache->lock);
 511                 kfree(caching_ctl);
 512                 return 0;
 513         }
 514         cache->caching_ctl = caching_ctl;
 515         cache->cached = BTRFS_CACHE_STARTED;
 516         spin_unlock(&cache->lock);
 517
 518         down_write(&fs_info->extent_commit_sem);
 519         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 520         up_write(&fs_info->extent_commit_sem);
 521
 522         btrfs_get_block_group(cache);
 523
 524         btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
 525
 526         return ret;
 527 }
 528
 529 /*
 530  * return the block group that starts at or after bytenr
 531  */
 532 static struct btrfs_block_group_cache *
 533 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 534 {
 535         struct btrfs_block_group_cache *cache;
 536
 537         cache = block_group_cache_tree_search(info, bytenr, 0);
 538
 539         return cache;
 540 }
 541
 542 /*
 543  * return the block group that contains the given bytenr
 544  */
 545 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 546                                                  struct btrfs_fs_info *info,
 547                                                  u64 bytenr)
 548 {
 549         struct btrfs_block_group_cache *cache;
 550
 551         cache = block_group_cache_tree_search(info, bytenr, 1);
 552
 553         return cache;
 554 }
 555
 556 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 557                                                   u64 flags)
 558 {
 559         struct list_head *head = &info->space_info;
 560         struct btrfs_space_info *found;
 561
 562         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 563                  BTRFS_BLOCK_GROUP_METADATA;
 564
 565         rcu_read_lock();
 566         list_for_each_entry_rcu(found, head, list) {
 567                 if (found->flags & flags) {
 568                         rcu_read_unlock();
 569                         return found;
 570                 }
 571         }
 572         rcu_read_unlock();
 573         return NULL;
 574 }
 575
 576 /*
 577  * after adding space to the filesystem, we need to clear the full flags
 578  * on all the space infos.
 579  */
 580 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 581 {
 582         struct list_head *head = &info->space_info;
 583         struct btrfs_space_info *found;
 584
 585         rcu_read_lock();
 586         list_for_each_entry_rcu(found, head, list)
 587                 found->full = 0;
 588         rcu_read_unlock();
 589 }
 590
 591 static u64 div_factor(u64 num, int factor)
 592 {
 593         if (factor == 10)
 594                 return num;
 595         num *= factor;
 596         do_div(num, 10);
 597         return num;
 598 }
 599
 600 static u64 div_factor_fine(u64 num, int factor)
 601 {
 602         if (factor == 100)
 603                 return num;
 604         num *= factor;
 605         do_div(num, 100);
 606         return num;
 607 }
 608
 609 u64 btrfs_find_block_group(struct btrfs_root *root,
 610                            u64 search_start, u64 search_hint, int owner)
 611 {
 612         struct btrfs_block_group_cache *cache;
 613         u64 used;
 614         u64 last = max(search_hint, search_start);
 615         u64 group_start = 0;
 616         int full_search = 0;
 617         int factor = 9;
 618         int wrapped = 0;
 619 again:
 620         while (1) {
 621                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 622                 if (!cache)
 623                         break;
 624
 625                 spin_lock(&cache->lock);
 626                 last = cache->key.objectid + cache->key.offset;
 627                 used = btrfs_block_group_used(&cache->item);
 628
 629                 if ((full_search || !cache->ro) &&
 630                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 631                         if (used + cache->pinned + cache->reserved <
 632                             div_factor(cache->key.offset, factor)) {
 633                                 group_start = cache->key.objectid;
 634                                 spin_unlock(&cache->lock);
 635                                 btrfs_put_block_group(cache);
 636                                 goto found;
 637                         }
 638                 }
 639                 spin_unlock(&cache->lock);
 640                 btrfs_put_block_group(cache);
 641                 cond_resched();
 642         }
 643         if (!wrapped) {
 644                 last = search_start;
 645                 wrapped = 1;
 646                 goto again;
 647         }
 648         if (!full_search && factor < 10) {
 649                 last = search_start;
 650                 full_search = 1;
 651                 factor = 10;
 652                 goto again;
 653         }
 654 found:
 655         return group_start;
 656 }
 657
 658 /* simple helper to search for an existing extent at a given offset */
 659 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 660 {
 661         int ret;
 662         struct btrfs_key key;
 663         struct btrfs_path *path;
 664
 665         path = btrfs_alloc_path();
 666         if (!path)
 667                 return -ENOMEM;
 668
 669         key.objectid = start;
 670         key.offset = len;
 671         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 672         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 673                                 0, 0);
 674         btrfs_free_path(path);
 675         return ret;
 676 }
 677
 678 /*
 679  * helper function to lookup reference count and flags of extent.
 680  *
 681  * the head node for delayed ref is used to store the sum of all the
 682  * reference count modifications queued up in the rbtree. the head
 683  * node may also store the extent flags to set. This way you can check
 684  * to see what the reference count and extent flags would be if all of
 685  * the delayed refs are not processed.
 686  */
 687 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 688                              struct btrfs_root *root, u64 bytenr,
 689                              u64 num_bytes, u64 *refs, u64 *flags)
 690 {
 691         struct btrfs_delayed_ref_head *head;
 692         struct btrfs_delayed_ref_root *delayed_refs;
 693         struct btrfs_path *path;
 694         struct btrfs_extent_item *ei;
 695         struct extent_buffer *leaf;
 696         struct btrfs_key key;
 697         u32 item_size;
 698         u64 num_refs;
 699         u64 extent_flags;
 700         int ret;
 701
 702         path = btrfs_alloc_path();
 703         if (!path)
 704                 return -ENOMEM;
 705
 706         key.objectid = bytenr;
 707         key.type = BTRFS_EXTENT_ITEM_KEY;
 708         key.offset = num_bytes;
 709         if (!trans) {
 710                 path->skip_locking = 1;
 711                 path->search_commit_root = 1;
 712         }
 713 again:
 714         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 715                                 &key, path, 0, 0);
 716         if (ret < 0)
 717                 goto out_free;
 718
 719         if (ret == 0) {
 720                 leaf = path->nodes[0];
 721                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 722                 if (item_size >= sizeof(*ei)) {
 723                         ei = btrfs_item_ptr(leaf, path->slots[0],
 724                                             struct btrfs_extent_item);
 725                         num_refs = btrfs_extent_refs(leaf, ei);
 726                         extent_flags = btrfs_extent_flags(leaf, ei);
 727                 } else {
 728 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 729                         struct btrfs_extent_item_v0 *ei0;
 730                         BUG_ON(item_size != sizeof(*ei0));
 731                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 732                                              struct btrfs_extent_item_v0);
 733                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 734                         /* FIXME: this isn't correct for data */
 735                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 736 #else
 737                         BUG();
 738 #endif
 739                 }
 740                 BUG_ON(num_refs == 0);
 741         } else {
 742                 num_refs = 0;
 743                 extent_flags = 0;
 744                 ret = 0;
 745         }
 746
 747         if (!trans)
 748                 goto out;
 749
 750         delayed_refs = &trans->transaction->delayed_refs;
 751         spin_lock(&delayed_refs->lock);
 752         head = btrfs_find_delayed_ref_head(trans, bytenr);
 753         if (head) {
 754                 if (!mutex_trylock(&head->mutex)) {
 755                         atomic_inc(&head->node.refs);
 756                         spin_unlock(&delayed_refs->lock);
 757
 758                         btrfs_release_path(path);
 759
 760                         /*
 761                          * Mutex was contended, block until it's released and try
 762                          * again
 763                          */
 764                         mutex_lock(&head->mutex);
 765                         mutex_unlock(&head->mutex);
 766                         btrfs_put_delayed_ref(&head->node);
 767                         goto again;
 768                 }
 769                 if (head->extent_op && head->extent_op->update_flags)
 770                         extent_flags |= head->extent_op->flags_to_set;
 771                 else
 772                         BUG_ON(num_refs == 0);
 773
 774                 num_refs += head->node.ref_mod;
 775                 mutex_unlock(&head->mutex);
 776         }
 777         spin_unlock(&delayed_refs->lock);
 778 out:
 779         WARN_ON(num_refs == 0);
 780         if (refs)
 781                 *refs = num_refs;
 782         if (flags)
 783                 *flags = extent_flags;
 784 out_free:
 785         btrfs_free_path(path);
 786         return ret;
 787 }
 788
 789 /*
 790  * Back reference rules.  Back refs have three main goals:
 791  *
 792  * 1) differentiate between all holders of references to an extent so that
 793  *    when a reference is dropped we can make sure it was a valid reference
 794  *    before freeing the extent.
 795  *
 796  * 2) Provide enough information to quickly find the holders of an extent
 797  *    if we notice a given block is corrupted or bad.
 798  *
 799  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 800  *    maintenance.  This is actually the same as #2, but with a slightly
 801  *    different use case.
 802  *
 803  * There are two kinds of back refs. The implicit back refs is optimized
 804  * for pointers in non-shared tree blocks. For a given pointer in a block,
 805  * back refs of this kind provide information about the block's owner tree
 806  * and the pointer's key. These information allow us to find the block by
 807  * b-tree searching. The full back refs is for pointers in tree blocks not
 808  * referenced by their owner trees. The location of tree block is recorded
 809  * in the back refs. Actually the full back refs is generic, and can be
 810  * used in all cases the implicit back refs is used. The major shortcoming
 811  * of the full back refs is its overhead. Every time a tree block gets
 812  * COWed, we have to update back refs entry for all pointers in it.
 813  *
 814  * For a newly allocated tree block, we use implicit back refs for
 815  * pointers in it. This means most tree related operations only involve
 816  * implicit back refs. For a tree block created in old transaction, the
 817  * only way to drop a reference to it is COW it. So we can detect the
 818  * event that tree block loses its owner tree's reference and do the
 819  * back refs conversion.
 820  *
 821  * When a tree block is COW'd through a tree, there are four cases:
 822  *
 823  * The reference count of the block is one and the tree is the block's
 824  * owner tree. Nothing to do in this case.
 825  *
 826  * The reference count of the block is one and the tree is not the
 827  * block's owner tree. In this case, full back refs is used for pointers
 828  * in the block. Remove these full back refs, add implicit back refs for
 829  * every pointers in the new block.
 830  *
 831  * The reference count of the block is greater than one and the tree is
 832  * the block's owner tree. In this case, implicit back refs is used for
 833  * pointers in the block. Add full back refs for every pointers in the
 834  * block, increase lower level extents' reference counts. The original
 835  * implicit back refs are entailed to the new block.
 836  *
 837  * The reference count of the block is greater than one and the tree is
 838  * not the block's owner tree. Add implicit back refs for every pointer in
 839  * the new block, increase lower level extents' reference count.
 840  *
 841  * Back Reference Key composing:
 842  *
 843  * The key objectid corresponds to the first byte in the extent,
 844  * The key type is used to differentiate between types of back refs.
 845  * There are different meanings of the key offset for different types
 846  * of back refs.
 847  *
 848  * File extents can be referenced by:
 849  *
 850  * - multiple snapshots, subvolumes, or different generations in one subvol
 851  * - different files inside a single subvolume
 852  * - different offsets inside a file (bookend extents in file.c)
 853  *
 854  * The extent ref structure for the implicit back refs has fields for:
 855  *
 856  * - Objectid of the subvolume root
 857  * - objectid of the file holding the reference
 858  * - original offset in the file
 859  * - how many bookend extents
 860  *
 861  * The key offset for the implicit back refs is hash of the first
 862  * three fields.
 863  *
 864  * The extent ref structure for the full back refs has field for:
 865  *
 866  * - number of pointers in the tree leaf
 867  *
 868  * The key offset for the implicit back refs is the first byte of
 869  * the tree leaf
 870  *
 871  * When a file extent is allocated, The implicit back refs is used.
 872  * the fields are filled in:
 873  *
 874  *     (root_key.objectid, inode objectid, offset in file, 1)
 875  *
 876  * When a file extent is removed file truncation, we find the
 877  * corresponding implicit back refs and check the following fields:
 878  *
 879  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 880  *
 881  * Btree extents can be referenced by:
 882  *
 883  * - Different subvolumes
 884  *
 885  * Both the implicit back refs and the full back refs for tree blocks
 886  * only consist of key. The key offset for the implicit back refs is
 887  * objectid of block's owner tree. The key offset for the full back refs
 888  * is the first byte of parent block.
 889  *
 890  * When implicit back refs is used, information about the lowest key and
 891  * level of the tree block are required. These information are stored in
 892  * tree block info structure.
 893  */
 894
 895 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 896 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 897                                   struct btrfs_root *root,
 898                                   struct btrfs_path *path,
 899                                   u64 owner, u32 extra_size)
 900 {
 901         struct btrfs_extent_item *item;
 902         struct btrfs_extent_item_v0 *ei0;
 903         struct btrfs_extent_ref_v0 *ref0;
 904         struct btrfs_tree_block_info *bi;
 905         struct extent_buffer *leaf;
 906         struct btrfs_key key;
 907         struct btrfs_key found_key;
 908         u32 new_size = sizeof(*item);
 909         u64 refs;
 910         int ret;
 911
 912         leaf = path->nodes[0];
 913         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 914
 915         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 916         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 917                              struct btrfs_extent_item_v0);
 918         refs = btrfs_extent_refs_v0(leaf, ei0);
 919
 920         if (owner == (u64)-1) {
 921                 while (1) {
 922                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 923                                 ret = btrfs_next_leaf(root, path);
 924                                 if (ret < 0)
 925                                         return ret;
 926                                 BUG_ON(ret > 0);
 927                                 leaf = path->nodes[0];
 928                         }
 929                         btrfs_item_key_to_cpu(leaf, &found_key,
 930                                               path->slots[0]);
 931                         BUG_ON(key.objectid != found_key.objectid);
 932                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 933                                 path->slots[0]++;
 934                                 continue;
 935                         }
 936                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 937                                               struct btrfs_extent_ref_v0);
 938                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 939                         break;
 940                 }
 941         }
 942         btrfs_release_path(path);
 943
 944         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 945                 new_size += sizeof(*bi);
 946
 947         new_size -= sizeof(*ei0);
 948         ret = btrfs_search_slot(trans, root, &key, path,
 949                                 new_size + extra_size, 1);
 950         if (ret < 0)
 951                 return ret;
 952         BUG_ON(ret);
 953
 954         ret = btrfs_extend_item(trans, root, path, new_size);
 955
 956         leaf = path->nodes[0];
 957         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 958         btrfs_set_extent_refs(leaf, item, refs);
 959         /* FIXME: get real generation */
 960         btrfs_set_extent_generation(leaf, item, 0);
 961         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 962                 btrfs_set_extent_flags(leaf, item,
 963                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 964                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 965                 bi = (struct btrfs_tree_block_info *)(item + 1);
 966                 /* FIXME: get first key of the block */
 967                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 968                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 969         } else {
 970                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 971         }
 972         btrfs_mark_buffer_dirty(leaf);
 973         return 0;
 974 }
 975 #endif
 976
 977 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 978 {
 979         u32 high_crc = ~(u32)0;
 980         u32 low_crc = ~(u32)0;
 981         __le64 lenum;
 982
 983         lenum = cpu_to_le64(root_objectid);
 984         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 985         lenum = cpu_to_le64(owner);
 986         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 987         lenum = cpu_to_le64(offset);
 988         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 989
 990         return ((u64)high_crc << 31) ^ (u64)low_crc;
 991 }
 992
 993 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 994                                      struct btrfs_extent_data_ref *ref)
 995 {
 996         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 997                                     btrfs_extent_data_ref_objectid(leaf, ref),
 998                                     btrfs_extent_data_ref_offset(leaf, ref));
 999 }
1000
1001 static int match_extent_data_ref(struct extent_buffer *leaf,
1002                                  struct btrfs_extent_data_ref *ref,
1003                                  u64 root_objectid, u64 owner, u64 offset)
1004 {
1005         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1006             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1007             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1008                 return 0;
1009         return 1;
1010 }
1011
1012 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1013                                            struct btrfs_root *root,
1014                                            struct btrfs_path *path,
1015                                            u64 bytenr, u64 parent,
1016                                            u64 root_objectid,
1017                                            u64 owner, u64 offset)
1018 {
1019         struct btrfs_key key;
1020         struct btrfs_extent_data_ref *ref;
1021         struct extent_buffer *leaf;
1022         u32 nritems;
1023         int ret;
1024         int recow;
1025         int err = -ENOENT;
1026
1027         key.objectid = bytenr;
1028         if (parent) {
1029                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1030                 key.offset = parent;
1031         } else {
1032                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1033                 key.offset = hash_extent_data_ref(root_objectid,
1034                                                   owner, offset);
1035         }
1036 again:
1037         recow = 0;
1038         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1039         if (ret < 0) {
1040                 err = ret;
1041                 goto fail;
1042         }
1043
1044         if (parent) {
1045                 if (!ret)
1046                         return 0;
1047 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1048                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1049                 btrfs_release_path(path);
1050                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1051                 if (ret < 0) {
1052                         err = ret;
1053                         goto fail;
1054                 }
1055                 if (!ret)
1056                         return 0;
1057 #endif
1058                 goto fail;
1059         }
1060
1061         leaf = path->nodes[0];
1062         nritems = btrfs_header_nritems(leaf);
1063         while (1) {
1064                 if (path->slots[0] >= nritems) {
1065                         ret = btrfs_next_leaf(root, path);
1066                         if (ret < 0)
1067                                 err = ret;
1068                         if (ret)
1069                                 goto fail;
1070
1071                         leaf = path->nodes[0];
1072                         nritems = btrfs_header_nritems(leaf);
1073                         recow = 1;
1074                 }
1075
1076                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1077                 if (key.objectid != bytenr ||
1078                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1079                         goto fail;
1080
1081                 ref = btrfs_item_ptr(leaf, path->slots[0],
1082                                      struct btrfs_extent_data_ref);
1083
1084                 if (match_extent_data_ref(leaf, ref, root_objectid,
1085                                           owner, offset)) {
1086                         if (recow) {
1087                                 btrfs_release_path(path);
1088                                 goto again;
1089                         }
1090                         err = 0;
1091                         break;
1092                 }
1093                 path->slots[0]++;
1094         }
1095 fail:
1096         return err;
1097 }
1098
1099 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1100                                            struct btrfs_root *root,
1101                                            struct btrfs_path *path,
1102                                            u64 bytenr, u64 parent,
1103                                            u64 root_objectid, u64 owner,
1104                                            u64 offset, int refs_to_add)
1105 {
1106         struct btrfs_key key;
1107         struct extent_buffer *leaf;
1108         u32 size;
1109         u32 num_refs;
1110         int ret;
1111
1112         key.objectid = bytenr;
1113         if (parent) {
1114                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1115                 key.offset = parent;
1116                 size = sizeof(struct btrfs_shared_data_ref);
1117         } else {
1118                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1119                 key.offset = hash_extent_data_ref(root_objectid,
1120                                                   owner, offset);
1121                 size = sizeof(struct btrfs_extent_data_ref);
1122         }
1123
1124         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1125         if (ret && ret != -EEXIST)
1126                 goto fail;
1127
1128         leaf = path->nodes[0];
1129         if (parent) {
1130                 struct btrfs_shared_data_ref *ref;
1131                 ref = btrfs_item_ptr(leaf, path->slots[0],
1132                                      struct btrfs_shared_data_ref);
1133                 if (ret == 0) {
1134                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1135                 } else {
1136                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1137                         num_refs += refs_to_add;
1138                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1139                 }
1140         } else {
1141                 struct btrfs_extent_data_ref *ref;
1142                 while (ret == -EEXIST) {
1143                         ref = btrfs_item_ptr(leaf, path->slots[0],
1144                                              struct btrfs_extent_data_ref);
1145                         if (match_extent_data_ref(leaf, ref, root_objectid,
1146                                                   owner, offset))
1147                                 break;
1148                         btrfs_release_path(path);
1149                         key.offset++;
1150                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1151                                                       size);
1152                         if (ret && ret != -EEXIST)
1153                                 goto fail;
1154
1155                         leaf = path->nodes[0];
1156                 }
1157                 ref = btrfs_item_ptr(leaf, path->slots[0],
1158                                      struct btrfs_extent_data_ref);
1159                 if (ret == 0) {
1160                         btrfs_set_extent_data_ref_root(leaf, ref,
1161                                                        root_objectid);
1162                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1163                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1164                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1165                 } else {
1166                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1167                         num_refs += refs_to_add;
1168                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1169                 }
1170         }
1171         btrfs_mark_buffer_dirty(leaf);
1172         ret = 0;
1173 fail:
1174         btrfs_release_path(path);
1175         return ret;
1176 }
1177
1178 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1179                                            struct btrfs_root *root,
1180                                            struct btrfs_path *path,
1181                                            int refs_to_drop)
1182 {
1183         struct btrfs_key key;
1184         struct btrfs_extent_data_ref *ref1 = NULL;
1185         struct btrfs_shared_data_ref *ref2 = NULL;
1186         struct extent_buffer *leaf;
1187         u32 num_refs = 0;
1188         int ret = 0;
1189
1190         leaf = path->nodes[0];
1191         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1192
1193         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1194                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1195                                       struct btrfs_extent_data_ref);
1196                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1197         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1198                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1199                                       struct btrfs_shared_data_ref);
1200                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1201 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1202         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1203                 struct btrfs_extent_ref_v0 *ref0;
1204                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1205                                       struct btrfs_extent_ref_v0);
1206                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1207 #endif
1208         } else {
1209                 BUG();
1210         }
1211
1212         BUG_ON(num_refs < refs_to_drop);
1213         num_refs -= refs_to_drop;
1214
1215         if (num_refs == 0) {
1216                 ret = btrfs_del_item(trans, root, path);
1217         } else {
1218                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1219                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1220                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1221                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1222 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1223                 else {
1224                         struct btrfs_extent_ref_v0 *ref0;
1225                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1226                                         struct btrfs_extent_ref_v0);
1227                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1228                 }
1229 #endif
1230                 btrfs_mark_buffer_dirty(leaf);
1231         }
1232         return ret;
1233 }
1234
1235 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1236                                           struct btrfs_path *path,
1237                                           struct btrfs_extent_inline_ref *iref)
1238 {
1239         struct btrfs_key key;
1240         struct extent_buffer *leaf;
1241         struct btrfs_extent_data_ref *ref1;
1242         struct btrfs_shared_data_ref *ref2;
1243         u32 num_refs = 0;
1244
1245         leaf = path->nodes[0];
1246         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1247         if (iref) {
1248                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1249                     BTRFS_EXTENT_DATA_REF_KEY) {
1250                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1251                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1252                 } else {
1253                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1254                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1255                 }
1256         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1257                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1258                                       struct btrfs_extent_data_ref);
1259                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1260         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1261                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1262                                       struct btrfs_shared_data_ref);
1263                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1264 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1265         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1266                 struct btrfs_extent_ref_v0 *ref0;
1267                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1268                                       struct btrfs_extent_ref_v0);
1269                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1270 #endif
1271         } else {
1272                 WARN_ON(1);
1273         }
1274         return num_refs;
1275 }
1276
1277 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1278                                           struct btrfs_root *root,
1279                                           struct btrfs_path *path,
1280                                           u64 bytenr, u64 parent,
1281                                           u64 root_objectid)
1282 {
1283         struct btrfs_key key;
1284         int ret;
1285
1286         key.objectid = bytenr;
1287         if (parent) {
1288                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1289                 key.offset = parent;
1290         } else {
1291                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1292                 key.offset = root_objectid;
1293         }
1294
1295         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1296         if (ret > 0)
1297                 ret = -ENOENT;
1298 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1299         if (ret == -ENOENT && parent) {
1300                 btrfs_release_path(path);
1301                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1302                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1303                 if (ret > 0)
1304                         ret = -ENOENT;
1305         }
1306 #endif
1307         return ret;
1308 }
1309
1310 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1311                                           struct btrfs_root *root,
1312                                           struct btrfs_path *path,
1313                                           u64 bytenr, u64 parent,
1314                                           u64 root_objectid)
1315 {
1316         struct btrfs_key key;
1317         int ret;
1318
1319         key.objectid = bytenr;
1320         if (parent) {
1321                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1322                 key.offset = parent;
1323         } else {
1324                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1325                 key.offset = root_objectid;
1326         }
1327
1328         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1329         btrfs_release_path(path);
1330         return ret;
1331 }
1332
1333 static inline int extent_ref_type(u64 parent, u64 owner)
1334 {
1335         int type;
1336         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1337                 if (parent > 0)
1338                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1339                 else
1340                         type = BTRFS_TREE_BLOCK_REF_KEY;
1341         } else {
1342                 if (parent > 0)
1343                         type = BTRFS_SHARED_DATA_REF_KEY;
1344                 else
1345                         type = BTRFS_EXTENT_DATA_REF_KEY;
1346         }
1347         return type;
1348 }
1349
1350 static int find_next_key(struct btrfs_path *path, int level,
1351                          struct btrfs_key *key)
1352
1353 {
1354         for (; level < BTRFS_MAX_LEVEL; level++) {
1355                 if (!path->nodes[level])
1356                         break;
1357                 if (path->slots[level] + 1 >=
1358                     btrfs_header_nritems(path->nodes[level]))
1359                         continue;
1360                 if (level == 0)
1361                         btrfs_item_key_to_cpu(path->nodes[level], key,
1362                                               path->slots[level] + 1);
1363                 else
1364                         btrfs_node_key_to_cpu(path->nodes[level], key,
1365                                               path->slots[level] + 1);
1366                 return 0;
1367         }
1368         return 1;
1369 }
1370
1371 /*
1372  * look for inline back ref. if back ref is found, *ref_ret is set
1373  * to the address of inline back ref, and 0 is returned.
1374  *
1375  * if back ref isn't found, *ref_ret is set to the address where it
1376  * should be inserted, and -ENOENT is returned.
1377  *
1378  * if insert is true and there are too many inline back refs, the path
1379  * points to the extent item, and -EAGAIN is returned.
1380  *
1381  * NOTE: inline back refs are ordered in the same way that back ref
1382  *       items in the tree are ordered.
1383  */
1384 static noinline_for_stack
1385 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1386                                  struct btrfs_root *root,
1387                                  struct btrfs_path *path,
1388                                  struct btrfs_extent_inline_ref **ref_ret,
1389                                  u64 bytenr, u64 num_bytes,
1390                                  u64 parent, u64 root_objectid,
1391                                  u64 owner, u64 offset, int insert)
1392 {
1393         struct btrfs_key key;
1394         struct extent_buffer *leaf;
1395         struct btrfs_extent_item *ei;
1396         struct btrfs_extent_inline_ref *iref;
1397         u64 flags;
1398         u64 item_size;
1399         unsigned long ptr;
1400         unsigned long end;
1401         int extra_size;
1402         int type;
1403         int want;
1404         int ret;
1405         int err = 0;
1406
1407         key.objectid = bytenr;
1408         key.type = BTRFS_EXTENT_ITEM_KEY;
1409         key.offset = num_bytes;
1410
1411         want = extent_ref_type(parent, owner);
1412         if (insert) {
1413                 extra_size = btrfs_extent_inline_ref_size(want);
1414                 path->keep_locks = 1;
1415         } else
1416                 extra_size = -1;
1417         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1418         if (ret < 0) {
1419                 err = ret;
1420                 goto out;
1421         }
1422         BUG_ON(ret);
1423
1424         leaf = path->nodes[0];
1425         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1426 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1427         if (item_size < sizeof(*ei)) {
1428                 if (!insert) {
1429                         err = -ENOENT;
1430                         goto out;
1431                 }
1432                 ret = convert_extent_item_v0(trans, root, path, owner,
1433                                              extra_size);
1434                 if (ret < 0) {
1435                         err = ret;
1436                         goto out;
1437                 }
1438                 leaf = path->nodes[0];
1439                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1440         }
1441 #endif
1442         BUG_ON(item_size < sizeof(*ei));
1443
1444         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1445         flags = btrfs_extent_flags(leaf, ei);
1446
1447         ptr = (unsigned long)(ei + 1);
1448         end = (unsigned long)ei + item_size;
1449
1450         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1451                 ptr += sizeof(struct btrfs_tree_block_info);
1452                 BUG_ON(ptr > end);
1453         } else {
1454                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1455         }
1456
1457         err = -ENOENT;
1458         while (1) {
1459                 if (ptr >= end) {
1460                         WARN_ON(ptr > end);
1461                         break;
1462                 }
1463                 iref = (struct btrfs_extent_inline_ref *)ptr;
1464                 type = btrfs_extent_inline_ref_type(leaf, iref);
1465                 if (want < type)
1466                         break;
1467                 if (want > type) {
1468                         ptr += btrfs_extent_inline_ref_size(type);
1469                         continue;
1470                 }
1471
1472                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1473                         struct btrfs_extent_data_ref *dref;
1474                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1475                         if (match_extent_data_ref(leaf, dref, root_objectid,
1476                                                   owner, offset)) {
1477                                 err = 0;
1478                                 break;
1479                         }
1480                         if (hash_extent_data_ref_item(leaf, dref) <
1481                             hash_extent_data_ref(root_objectid, owner, offset))
1482                                 break;
1483                 } else {
1484                         u64 ref_offset;
1485                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1486                         if (parent > 0) {
1487                                 if (parent == ref_offset) {
1488                                         err = 0;
1489                                         break;
1490                                 }
1491                                 if (ref_offset < parent)
1492                                         break;
1493                         } else {
1494                                 if (root_objectid == ref_offset) {
1495                                         err = 0;
1496                                         break;
1497                                 }
1498                                 if (ref_offset < root_objectid)
1499                                         break;
1500                         }
1501                 }
1502                 ptr += btrfs_extent_inline_ref_size(type);
1503         }
1504         if (err == -ENOENT && insert) {
1505                 if (item_size + extra_size >=
1506                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1507                         err = -EAGAIN;
1508                         goto out;
1509                 }
1510                 /*
1511                  * To add new inline back ref, we have to make sure
1512                  * there is no corresponding back ref item.
1513                  * For simplicity, we just do not add new inline back
1514                  * ref if there is any kind of item for this block
1515                  */
1516                 if (find_next_key(path, 0, &key) == 0 &&
1517                     key.objectid == bytenr &&
1518                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1519                         err = -EAGAIN;
1520                         goto out;
1521                 }
1522         }
1523         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1524 out:
1525         if (insert) {
1526                 path->keep_locks = 0;
1527                 btrfs_unlock_up_safe(path, 1);
1528         }
1529         return err;
1530 }
1531
1532 /*
1533  * helper to add new inline back ref
1534  */
1535 static noinline_for_stack
1536 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1537                                 struct btrfs_root *root,
1538                                 struct btrfs_path *path,
1539                                 struct btrfs_extent_inline_ref *iref,
1540                                 u64 parent, u64 root_objectid,
1541                                 u64 owner, u64 offset, int refs_to_add,
1542                                 struct btrfs_delayed_extent_op *extent_op)
1543 {
1544         struct extent_buffer *leaf;
1545         struct btrfs_extent_item *ei;
1546         unsigned long ptr;
1547         unsigned long end;
1548         unsigned long item_offset;
1549         u64 refs;
1550         int size;
1551         int type;
1552         int ret;
1553
1554         leaf = path->nodes[0];
1555         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1556         item_offset = (unsigned long)iref - (unsigned long)ei;
1557
1558         type = extent_ref_type(parent, owner);
1559         size = btrfs_extent_inline_ref_size(type);
1560
1561         ret = btrfs_extend_item(trans, root, path, size);
1562
1563         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1564         refs = btrfs_extent_refs(leaf, ei);
1565         refs += refs_to_add;
1566         btrfs_set_extent_refs(leaf, ei, refs);
1567         if (extent_op)
1568                 __run_delayed_extent_op(extent_op, leaf, ei);
1569
1570         ptr = (unsigned long)ei + item_offset;
1571         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1572         if (ptr < end - size)
1573                 memmove_extent_buffer(leaf, ptr + size, ptr,
1574                                       end - size - ptr);
1575
1576         iref = (struct btrfs_extent_inline_ref *)ptr;
1577         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1578         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1579                 struct btrfs_extent_data_ref *dref;
1580                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1581                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1582                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1583                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1584                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1585         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1586                 struct btrfs_shared_data_ref *sref;
1587                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1588                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1589                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1590         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1591                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1592         } else {
1593                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1594         }
1595         btrfs_mark_buffer_dirty(leaf);
1596         return 0;
1597 }
1598
1599 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1600                                  struct btrfs_root *root,
1601                                  struct btrfs_path *path,
1602                                  struct btrfs_extent_inline_ref **ref_ret,
1603                                  u64 bytenr, u64 num_bytes, u64 parent,
1604                                  u64 root_objectid, u64 owner, u64 offset)
1605 {
1606         int ret;
1607
1608         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1609                                            bytenr, num_bytes, parent,
1610                                            root_objectid, owner, offset, 0);
1611         if (ret != -ENOENT)
1612                 return ret;
1613
1614         btrfs_release_path(path);
1615         *ref_ret = NULL;
1616
1617         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1618                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1619                                             root_objectid);
1620         } else {
1621                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1622                                              root_objectid, owner, offset);
1623         }
1624         return ret;
1625 }
1626
1627 /*
1628  * helper to update/remove inline back ref
1629  */
1630 static noinline_for_stack
1631 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1632                                  struct btrfs_root *root,
1633                                  struct btrfs_path *path,
1634                                  struct btrfs_extent_inline_ref *iref,
1635                                  int refs_to_mod,
1636                                  struct btrfs_delayed_extent_op *extent_op)
1637 {
1638         struct extent_buffer *leaf;
1639         struct btrfs_extent_item *ei;
1640         struct btrfs_extent_data_ref *dref = NULL;
1641         struct btrfs_shared_data_ref *sref = NULL;
1642         unsigned long ptr;
1643         unsigned long end;
1644         u32 item_size;
1645         int size;
1646         int type;
1647         int ret;
1648         u64 refs;
1649
1650         leaf = path->nodes[0];
1651         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1652         refs = btrfs_extent_refs(leaf, ei);
1653         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1654         refs += refs_to_mod;
1655         btrfs_set_extent_refs(leaf, ei, refs);
1656         if (extent_op)
1657                 __run_delayed_extent_op(extent_op, leaf, ei);
1658
1659         type = btrfs_extent_inline_ref_type(leaf, iref);
1660
1661         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1662                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1663                 refs = btrfs_extent_data_ref_count(leaf, dref);
1664         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1665                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1666                 refs = btrfs_shared_data_ref_count(leaf, sref);
1667         } else {
1668                 refs = 1;
1669                 BUG_ON(refs_to_mod != -1);
1670         }
1671
1672         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1673         refs += refs_to_mod;
1674
1675         if (refs > 0) {
1676                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1677                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1678                 else
1679                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1680         } else {
1681                 size =  btrfs_extent_inline_ref_size(type);
1682                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1683                 ptr = (unsigned long)iref;
1684                 end = (unsigned long)ei + item_size;
1685                 if (ptr + size < end)
1686                         memmove_extent_buffer(leaf, ptr, ptr + size,
1687                                               end - ptr - size);
1688                 item_size -= size;
1689                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1690         }
1691         btrfs_mark_buffer_dirty(leaf);
1692         return 0;
1693 }
1694
1695 static noinline_for_stack
1696 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1697                                  struct btrfs_root *root,
1698                                  struct btrfs_path *path,
1699                                  u64 bytenr, u64 num_bytes, u64 parent,
1700                                  u64 root_objectid, u64 owner,
1701                                  u64 offset, int refs_to_add,
1702                                  struct btrfs_delayed_extent_op *extent_op)
1703 {
1704         struct btrfs_extent_inline_ref *iref;
1705         int ret;
1706
1707         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1708                                            bytenr, num_bytes, parent,
1709                                            root_objectid, owner, offset, 1);
1710         if (ret == 0) {
1711                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1712                 ret = update_inline_extent_backref(trans, root, path, iref,
1713                                                    refs_to_add, extent_op);
1714         } else if (ret == -ENOENT) {
1715                 ret = setup_inline_extent_backref(trans, root, path, iref,
1716                                                   parent, root_objectid,
1717                                                   owner, offset, refs_to_add,
1718                                                   extent_op);
1719         }
1720         return ret;
1721 }
1722
1723 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1724                                  struct btrfs_root *root,
1725                                  struct btrfs_path *path,
1726                                  u64 bytenr, u64 parent, u64 root_objectid,
1727                                  u64 owner, u64 offset, int refs_to_add)
1728 {
1729         int ret;
1730         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1731                 BUG_ON(refs_to_add != 1);
1732                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1733                                             parent, root_objectid);
1734         } else {
1735                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1736                                              parent, root_objectid,
1737                                              owner, offset, refs_to_add);
1738         }
1739         return ret;
1740 }
1741
1742 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1743                                  struct btrfs_root *root,
1744                                  struct btrfs_path *path,
1745                                  struct btrfs_extent_inline_ref *iref,
1746                                  int refs_to_drop, int is_data)
1747 {
1748         int ret;
1749
1750         BUG_ON(!is_data && refs_to_drop != 1);
1751         if (iref) {
1752                 ret = update_inline_extent_backref(trans, root, path, iref,
1753                                                    -refs_to_drop, NULL);
1754         } else if (is_data) {
1755                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1756         } else {
1757                 ret = btrfs_del_item(trans, root, path);
1758         }
1759         return ret;
1760 }
1761
1762 static int btrfs_issue_discard(struct block_device *bdev,
1763                                 u64 start, u64 len)
1764 {
1765         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1766 }
1767
1768 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1769                                 u64 num_bytes, u64 *actual_bytes)
1770 {
1771         int ret;
1772         u64 discarded_bytes = 0;
1773         struct btrfs_multi_bio *multi = NULL;
1774
1775
1776         /* Tell the block device(s) that the sectors can be discarded */
1777         ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1778                               bytenr, &num_bytes, &multi, 0);
1779         if (!ret) {
1780                 struct btrfs_bio_stripe *stripe = multi->stripes;
1781                 int i;
1782
1783
1784                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1785                         if (!stripe->dev->can_discard)
1786                                 continue;
1787
1788                         ret = btrfs_issue_discard(stripe->dev->bdev,
1789                                                   stripe->physical,
1790                                                   stripe->length);
1791                         if (!ret)
1792                                 discarded_bytes += stripe->length;
1793                         else if (ret != -EOPNOTSUPP)
1794                                 break;
1795
1796                         /*
1797                          * Just in case we get back EOPNOTSUPP for some reason,
1798                          * just ignore the return value so we don't screw up
1799                          * people calling discard_extent.
1800                          */
1801                         ret = 0;
1802                 }
1803                 kfree(multi);
1804         }
1805
1806         if (actual_bytes)
1807                 *actual_bytes = discarded_bytes;
1808
1809
1810         return ret;
1811 }
1812
1813 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1814                          struct btrfs_root *root,
1815                          u64 bytenr, u64 num_bytes, u64 parent,
1816                          u64 root_objectid, u64 owner, u64 offset)
1817 {
1818         int ret;
1819         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1820                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1821
1822         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1823                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1824                                         parent, root_objectid, (int)owner,
1825                                         BTRFS_ADD_DELAYED_REF, NULL);
1826         } else {
1827                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1828                                         parent, root_objectid, owner, offset,
1829                                         BTRFS_ADD_DELAYED_REF, NULL);
1830         }
1831         return ret;
1832 }
1833
1834 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1835                                   struct btrfs_root *root,
1836                                   u64 bytenr, u64 num_bytes,
1837                                   u64 parent, u64 root_objectid,
1838                                   u64 owner, u64 offset, int refs_to_add,
1839                                   struct btrfs_delayed_extent_op *extent_op)
1840 {
1841         struct btrfs_path *path;
1842         struct extent_buffer *leaf;
1843         struct btrfs_extent_item *item;
1844         u64 refs;
1845         int ret;
1846         int err = 0;
1847
1848         path = btrfs_alloc_path();
1849         if (!path)
1850                 return -ENOMEM;
1851
1852         path->reada = 1;
1853         path->leave_spinning = 1;
1854         /* this will setup the path even if it fails to insert the back ref */
1855         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1856                                            path, bytenr, num_bytes, parent,
1857                                            root_objectid, owner, offset,
1858                                            refs_to_add, extent_op);
1859         if (ret == 0)
1860                 goto out;
1861
1862         if (ret != -EAGAIN) {
1863                 err = ret;
1864                 goto out;
1865         }
1866
1867         leaf = path->nodes[0];
1868         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1869         refs = btrfs_extent_refs(leaf, item);
1870         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1871         if (extent_op)
1872                 __run_delayed_extent_op(extent_op, leaf, item);
1873
1874         btrfs_mark_buffer_dirty(leaf);
1875         btrfs_release_path(path);
1876
1877         path->reada = 1;
1878         path->leave_spinning = 1;
1879
1880         /* now insert the actual backref */
1881         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1882                                     path, bytenr, parent, root_objectid,
1883                                     owner, offset, refs_to_add);
1884         BUG_ON(ret);
1885 out:
1886         btrfs_free_path(path);
1887         return err;
1888 }
1889
1890 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1891                                 struct btrfs_root *root,
1892                                 struct btrfs_delayed_ref_node *node,
1893                                 struct btrfs_delayed_extent_op *extent_op,
1894                                 int insert_reserved)
1895 {
1896         int ret = 0;
1897         struct btrfs_delayed_data_ref *ref;
1898         struct btrfs_key ins;
1899         u64 parent = 0;
1900         u64 ref_root = 0;
1901         u64 flags = 0;
1902
1903         ins.objectid = node->bytenr;
1904         ins.offset = node->num_bytes;
1905         ins.type = BTRFS_EXTENT_ITEM_KEY;
1906
1907         ref = btrfs_delayed_node_to_data_ref(node);
1908         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1909                 parent = ref->parent;
1910         else
1911                 ref_root = ref->root;
1912
1913         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1914                 if (extent_op) {
1915                         BUG_ON(extent_op->update_key);
1916                         flags |= extent_op->flags_to_set;
1917                 }
1918                 ret = alloc_reserved_file_extent(trans, root,
1919                                                  parent, ref_root, flags,
1920                                                  ref->objectid, ref->offset,
1921                                                  &ins, node->ref_mod);
1922         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1923                 ret = __btrfs_inc_extent_ref(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 if (node->action == BTRFS_DROP_DELAYED_REF) {
1929                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1930                                           node->num_bytes, parent,
1931                                           ref_root, ref->objectid,
1932                                           ref->offset, node->ref_mod,
1933                                           extent_op);
1934         } else {
1935                 BUG();
1936         }
1937         return ret;
1938 }
1939
1940 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1941                                     struct extent_buffer *leaf,
1942                                     struct btrfs_extent_item *ei)
1943 {
1944         u64 flags = btrfs_extent_flags(leaf, ei);
1945         if (extent_op->update_flags) {
1946                 flags |= extent_op->flags_to_set;
1947                 btrfs_set_extent_flags(leaf, ei, flags);
1948         }
1949
1950         if (extent_op->update_key) {
1951                 struct btrfs_tree_block_info *bi;
1952                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1953                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1954                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1955         }
1956 }
1957
1958 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1959                                  struct btrfs_root *root,
1960                                  struct btrfs_delayed_ref_node *node,
1961                                  struct btrfs_delayed_extent_op *extent_op)
1962 {
1963         struct btrfs_key key;
1964         struct btrfs_path *path;
1965         struct btrfs_extent_item *ei;
1966         struct extent_buffer *leaf;
1967         u32 item_size;
1968         int ret;
1969         int err = 0;
1970
1971         path = btrfs_alloc_path();
1972         if (!path)
1973                 return -ENOMEM;
1974
1975         key.objectid = node->bytenr;
1976         key.type = BTRFS_EXTENT_ITEM_KEY;
1977         key.offset = node->num_bytes;
1978
1979         path->reada = 1;
1980         path->leave_spinning = 1;
1981         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1982                                 path, 0, 1);
1983         if (ret < 0) {
1984                 err = ret;
1985                 goto out;
1986         }
1987         if (ret > 0) {
1988                 err = -EIO;
1989                 goto out;
1990         }
1991
1992         leaf = path->nodes[0];
1993         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1994 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1995         if (item_size < sizeof(*ei)) {
1996                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1997                                              path, (u64)-1, 0);
1998                 if (ret < 0) {
1999                         err = ret;
2000                         goto out;
2001                 }
2002                 leaf = path->nodes[0];
2003                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2004         }
2005 #endif
2006         BUG_ON(item_size < sizeof(*ei));
2007         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2008         __run_delayed_extent_op(extent_op, leaf, ei);
2009
2010         btrfs_mark_buffer_dirty(leaf);
2011 out:
2012         btrfs_free_path(path);
2013         return err;
2014 }
2015
2016 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2017                                 struct btrfs_root *root,
2018                                 struct btrfs_delayed_ref_node *node,
2019                                 struct btrfs_delayed_extent_op *extent_op,
2020                                 int insert_reserved)
2021 {
2022         int ret = 0;
2023         struct btrfs_delayed_tree_ref *ref;
2024         struct btrfs_key ins;
2025         u64 parent = 0;
2026         u64 ref_root = 0;
2027
2028         ins.objectid = node->bytenr;
2029         ins.offset = node->num_bytes;
2030         ins.type = BTRFS_EXTENT_ITEM_KEY;
2031
2032         ref = btrfs_delayed_node_to_tree_ref(node);
2033         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2034                 parent = ref->parent;
2035         else
2036                 ref_root = ref->root;
2037
2038         BUG_ON(node->ref_mod != 1);
2039         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2040                 BUG_ON(!extent_op || !extent_op->update_flags ||
2041                        !extent_op->update_key);
2042                 ret = alloc_reserved_tree_block(trans, root,
2043                                                 parent, ref_root,
2044                                                 extent_op->flags_to_set,
2045                                                 &extent_op->key,
2046                                                 ref->level, &ins);
2047         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2048                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2049                                              node->num_bytes, parent, ref_root,
2050                                              ref->level, 0, 1, extent_op);
2051         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2052                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2053                                           node->num_bytes, parent, ref_root,
2054                                           ref->level, 0, 1, extent_op);
2055         } else {
2056                 BUG();
2057         }
2058         return ret;
2059 }
2060
2061 /* helper function to actually process a single delayed ref entry */
2062 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2063                                struct btrfs_root *root,
2064                                struct btrfs_delayed_ref_node *node,
2065                                struct btrfs_delayed_extent_op *extent_op,
2066                                int insert_reserved)
2067 {
2068         int ret;
2069         if (btrfs_delayed_ref_is_head(node)) {
2070                 struct btrfs_delayed_ref_head *head;
2071                 /*
2072                  * we've hit the end of the chain and we were supposed
2073                  * to insert this extent into the tree.  But, it got
2074                  * deleted before we ever needed to insert it, so all
2075                  * we have to do is clean up the accounting
2076                  */
2077                 BUG_ON(extent_op);
2078                 head = btrfs_delayed_node_to_head(node);
2079                 if (insert_reserved) {
2080                         btrfs_pin_extent(root, node->bytenr,
2081                                          node->num_bytes, 1);
2082                         if (head->is_data) {
2083                                 ret = btrfs_del_csums(trans, root,
2084                                                       node->bytenr,
2085                                                       node->num_bytes);
2086                                 BUG_ON(ret);
2087                         }
2088                 }
2089                 mutex_unlock(&head->mutex);
2090                 return 0;
2091         }
2092
2093         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2094             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2095                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2096                                            insert_reserved);
2097         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2098                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2099                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2100                                            insert_reserved);
2101         else
2102                 BUG();
2103         return ret;
2104 }
2105
2106 static noinline struct btrfs_delayed_ref_node *
2107 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2108 {
2109         struct rb_node *node;
2110         struct btrfs_delayed_ref_node *ref;
2111         int action = BTRFS_ADD_DELAYED_REF;
2112 again:
2113         /*
2114          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2115          * this prevents ref count from going down to zero when
2116          * there still are pending delayed ref.
2117          */
2118         node = rb_prev(&head->node.rb_node);
2119         while (1) {
2120                 if (!node)
2121                         break;
2122                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2123                                 rb_node);
2124                 if (ref->bytenr != head->node.bytenr)
2125                         break;
2126                 if (ref->action == action)
2127                         return ref;
2128                 node = rb_prev(node);
2129         }
2130         if (action == BTRFS_ADD_DELAYED_REF) {
2131                 action = BTRFS_DROP_DELAYED_REF;
2132                 goto again;
2133         }
2134         return NULL;
2135 }
2136
2137 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2138                                        struct btrfs_root *root,
2139                                        struct list_head *cluster)
2140 {
2141         struct btrfs_delayed_ref_root *delayed_refs;
2142         struct btrfs_delayed_ref_node *ref;
2143         struct btrfs_delayed_ref_head *locked_ref = NULL;
2144         struct btrfs_delayed_extent_op *extent_op;
2145         int ret;
2146         int count = 0;
2147         int must_insert_reserved = 0;
2148
2149         delayed_refs = &trans->transaction->delayed_refs;
2150         while (1) {
2151                 if (!locked_ref) {
2152                         /* pick a new head ref from the cluster list */
2153                         if (list_empty(cluster))
2154                                 break;
2155
2156                         locked_ref = list_entry(cluster->next,
2157                                      struct btrfs_delayed_ref_head, cluster);
2158
2159                         /* grab the lock that says we are going to process
2160                          * all the refs for this head */
2161                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2162
2163                         /*
2164                          * we may have dropped the spin lock to get the head
2165                          * mutex lock, and that might have given someone else
2166                          * time to free the head.  If that's true, it has been
2167                          * removed from our list and we can move on.
2168                          */
2169                         if (ret == -EAGAIN) {
2170                                 locked_ref = NULL;
2171                                 count++;
2172                                 continue;
2173                         }
2174                 }
2175
2176                 /*
2177                  * record the must insert reserved flag before we
2178                  * drop the spin lock.
2179                  */
2180                 must_insert_reserved = locked_ref->must_insert_reserved;
2181                 locked_ref->must_insert_reserved = 0;
2182
2183                 extent_op = locked_ref->extent_op;
2184                 locked_ref->extent_op = NULL;
2185
2186                 /*
2187                  * locked_ref is the head node, so we have to go one
2188                  * node back for any delayed ref updates
2189                  */
2190                 ref = select_delayed_ref(locked_ref);
2191                 if (!ref) {
2192                         /* All delayed refs have been processed, Go ahead
2193                          * and send the head node to run_one_delayed_ref,
2194                          * so that any accounting fixes can happen
2195                          */
2196                         ref = &locked_ref->node;
2197
2198                         if (extent_op && must_insert_reserved) {
2199                                 kfree(extent_op);
2200                                 extent_op = NULL;
2201                         }
2202
2203                         if (extent_op) {
2204                                 spin_unlock(&delayed_refs->lock);
2205
2206                                 ret = run_delayed_extent_op(trans, root,
2207                                                             ref, extent_op);
2208                                 BUG_ON(ret);
2209                                 kfree(extent_op);
2210
2211                                 cond_resched();
2212                                 spin_lock(&delayed_refs->lock);
2213                                 continue;
2214                         }
2215
2216                         list_del_init(&locked_ref->cluster);
2217                         locked_ref = NULL;
2218                 }
2219
2220                 ref->in_tree = 0;
2221                 rb_erase(&ref->rb_node, &delayed_refs->root);
2222                 delayed_refs->num_entries--;
2223
2224                 spin_unlock(&delayed_refs->lock);
2225
2226                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2227                                           must_insert_reserved);
2228                 BUG_ON(ret);
2229
2230                 btrfs_put_delayed_ref(ref);
2231                 kfree(extent_op);
2232                 count++;
2233
2234                 cond_resched();
2235                 spin_lock(&delayed_refs->lock);
2236         }
2237         return count;
2238 }
2239
2240 /*
2241  * this starts processing the delayed reference count updates and
2242  * extent insertions we have queued up so far.  count can be
2243  * 0, which means to process everything in the tree at the start
2244  * of the run (but not newly added entries), or it can be some target
2245  * number you'd like to process.
2246  */
2247 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2248                            struct btrfs_root *root, unsigned long count)
2249 {
2250         struct rb_node *node;
2251         struct btrfs_delayed_ref_root *delayed_refs;
2252         struct btrfs_delayed_ref_node *ref;
2253         struct list_head cluster;
2254         int ret;
2255         int run_all = count == (unsigned long)-1;
2256         int run_most = 0;
2257
2258         if (root == root->fs_info->extent_root)
2259                 root = root->fs_info->tree_root;
2260
2261         delayed_refs = &trans->transaction->delayed_refs;
2262         INIT_LIST_HEAD(&cluster);
2263 again:
2264         spin_lock(&delayed_refs->lock);
2265         if (count == 0) {
2266                 count = delayed_refs->num_entries * 2;
2267                 run_most = 1;
2268         }
2269         while (1) {
2270                 if (!(run_all || run_most) &&
2271                     delayed_refs->num_heads_ready < 64)
2272                         break;
2273
2274                 /*
2275                  * go find something we can process in the rbtree.  We start at
2276                  * the beginning of the tree, and then build a cluster
2277                  * of refs to process starting at the first one we are able to
2278                  * lock
2279                  */
2280                 ret = btrfs_find_ref_cluster(trans, &cluster,
2281                                              delayed_refs->run_delayed_start);
2282                 if (ret)
2283                         break;
2284
2285                 ret = run_clustered_refs(trans, root, &cluster);
2286                 BUG_ON(ret < 0);
2287
2288                 count -= min_t(unsigned long, ret, count);
2289
2290                 if (count == 0)
2291                         break;
2292         }
2293
2294         if (run_all) {
2295                 node = rb_first(&delayed_refs->root);
2296                 if (!node)
2297                         goto out;
2298                 count = (unsigned long)-1;
2299
2300                 while (node) {
2301                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2302                                        rb_node);
2303                         if (btrfs_delayed_ref_is_head(ref)) {
2304                                 struct btrfs_delayed_ref_head *head;
2305
2306                                 head = btrfs_delayed_node_to_head(ref);
2307                                 atomic_inc(&ref->refs);
2308
2309                                 spin_unlock(&delayed_refs->lock);
2310                                 /*
2311                                  * Mutex was contended, block until it's
2312                                  * released and try again
2313                                  */
2314                                 mutex_lock(&head->mutex);
2315                                 mutex_unlock(&head->mutex);
2316
2317                                 btrfs_put_delayed_ref(ref);
2318                                 cond_resched();
2319                                 goto again;
2320                         }
2321                         node = rb_next(node);
2322                 }
2323                 spin_unlock(&delayed_refs->lock);
2324                 schedule_timeout(1);
2325                 goto again;
2326         }
2327 out:
2328         spin_unlock(&delayed_refs->lock);
2329         return 0;
2330 }
2331
2332 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2333                                 struct btrfs_root *root,
2334                                 u64 bytenr, u64 num_bytes, u64 flags,
2335                                 int is_data)
2336 {
2337         struct btrfs_delayed_extent_op *extent_op;
2338         int ret;
2339
2340         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2341         if (!extent_op)
2342                 return -ENOMEM;
2343
2344         extent_op->flags_to_set = flags;
2345         extent_op->update_flags = 1;
2346         extent_op->update_key = 0;
2347         extent_op->is_data = is_data ? 1 : 0;
2348
2349         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2350         if (ret)
2351                 kfree(extent_op);
2352         return ret;
2353 }
2354
2355 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2356                                       struct btrfs_root *root,
2357                                       struct btrfs_path *path,
2358                                       u64 objectid, u64 offset, u64 bytenr)
2359 {
2360         struct btrfs_delayed_ref_head *head;
2361         struct btrfs_delayed_ref_node *ref;
2362         struct btrfs_delayed_data_ref *data_ref;
2363         struct btrfs_delayed_ref_root *delayed_refs;
2364         struct rb_node *node;
2365         int ret = 0;
2366
2367         ret = -ENOENT;
2368         delayed_refs = &trans->transaction->delayed_refs;
2369         spin_lock(&delayed_refs->lock);
2370         head = btrfs_find_delayed_ref_head(trans, bytenr);
2371         if (!head)
2372                 goto out;
2373
2374         if (!mutex_trylock(&head->mutex)) {
2375                 atomic_inc(&head->node.refs);
2376                 spin_unlock(&delayed_refs->lock);
2377
2378                 btrfs_release_path(path);
2379
2380                 /*
2381                  * Mutex was contended, block until it's released and let
2382                  * caller try again
2383                  */
2384                 mutex_lock(&head->mutex);
2385                 mutex_unlock(&head->mutex);
2386                 btrfs_put_delayed_ref(&head->node);
2387                 return -EAGAIN;
2388         }
2389
2390         node = rb_prev(&head->node.rb_node);
2391         if (!node)
2392                 goto out_unlock;
2393
2394         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2395
2396         if (ref->bytenr != bytenr)
2397                 goto out_unlock;
2398
2399         ret = 1;
2400         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2401                 goto out_unlock;
2402
2403         data_ref = btrfs_delayed_node_to_data_ref(ref);
2404
2405         node = rb_prev(node);
2406         if (node) {
2407                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2408                 if (ref->bytenr == bytenr)
2409                         goto out_unlock;
2410         }
2411
2412         if (data_ref->root != root->root_key.objectid ||
2413             data_ref->objectid != objectid || data_ref->offset != offset)
2414                 goto out_unlock;
2415
2416         ret = 0;
2417 out_unlock:
2418         mutex_unlock(&head->mutex);
2419 out:
2420         spin_unlock(&delayed_refs->lock);
2421         return ret;
2422 }
2423
2424 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2425                                         struct btrfs_root *root,
2426                                         struct btrfs_path *path,
2427                                         u64 objectid, u64 offset, u64 bytenr)
2428 {
2429         struct btrfs_root *extent_root = root->fs_info->extent_root;
2430         struct extent_buffer *leaf;
2431         struct btrfs_extent_data_ref *ref;
2432         struct btrfs_extent_inline_ref *iref;
2433         struct btrfs_extent_item *ei;
2434         struct btrfs_key key;
2435         u32 item_size;
2436         int ret;
2437
2438         key.objectid = bytenr;
2439         key.offset = (u64)-1;
2440         key.type = BTRFS_EXTENT_ITEM_KEY;
2441
2442         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2443         if (ret < 0)
2444                 goto out;
2445         BUG_ON(ret == 0);
2446
2447         ret = -ENOENT;
2448         if (path->slots[0] == 0)
2449                 goto out;
2450
2451         path->slots[0]--;
2452         leaf = path->nodes[0];
2453         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2454
2455         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2456                 goto out;
2457
2458         ret = 1;
2459         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2460 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2461         if (item_size < sizeof(*ei)) {
2462                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2463                 goto out;
2464         }
2465 #endif
2466         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2467
2468         if (item_size != sizeof(*ei) +
2469             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2470                 goto out;
2471
2472         if (btrfs_extent_generation(leaf, ei) <=
2473             btrfs_root_last_snapshot(&root->root_item))
2474                 goto out;
2475
2476         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2477         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2478             BTRFS_EXTENT_DATA_REF_KEY)
2479                 goto out;
2480
2481         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2482         if (btrfs_extent_refs(leaf, ei) !=
2483             btrfs_extent_data_ref_count(leaf, ref) ||
2484             btrfs_extent_data_ref_root(leaf, ref) !=
2485             root->root_key.objectid ||
2486             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2487             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2488                 goto out;
2489
2490         ret = 0;
2491 out:
2492         return ret;
2493 }
2494
2495 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2496                           struct btrfs_root *root,
2497                           u64 objectid, u64 offset, u64 bytenr)
2498 {
2499         struct btrfs_path *path;
2500         int ret;
2501         int ret2;
2502
2503         path = btrfs_alloc_path();
2504         if (!path)
2505                 return -ENOENT;
2506
2507         do {
2508                 ret = check_committed_ref(trans, root, path, objectid,
2509                                           offset, bytenr);
2510                 if (ret && ret != -ENOENT)
2511                         goto out;
2512
2513                 ret2 = check_delayed_ref(trans, root, path, objectid,
2514                                          offset, bytenr);
2515         } while (ret2 == -EAGAIN);
2516
2517         if (ret2 && ret2 != -ENOENT) {
2518                 ret = ret2;
2519                 goto out;
2520         }
2521
2522         if (ret != -ENOENT || ret2 != -ENOENT)
2523                 ret = 0;
2524 out:
2525         btrfs_free_path(path);
2526         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2527                 WARN_ON(ret > 0);
2528         return ret;
2529 }
2530
2531 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2532                            struct btrfs_root *root,
2533                            struct extent_buffer *buf,
2534                            int full_backref, int inc)
2535 {
2536         u64 bytenr;
2537         u64 num_bytes;
2538         u64 parent;
2539         u64 ref_root;
2540         u32 nritems;
2541         struct btrfs_key key;
2542         struct btrfs_file_extent_item *fi;
2543         int i;
2544         int level;
2545         int ret = 0;
2546         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2547                             u64, u64, u64, u64, u64, u64);
2548
2549         ref_root = btrfs_header_owner(buf);
2550         nritems = btrfs_header_nritems(buf);
2551         level = btrfs_header_level(buf);
2552
2553         if (!root->ref_cows && level == 0)
2554                 return 0;
2555
2556         if (inc)
2557                 process_func = btrfs_inc_extent_ref;
2558         else
2559                 process_func = btrfs_free_extent;
2560
2561         if (full_backref)
2562                 parent = buf->start;
2563         else
2564                 parent = 0;
2565
2566         for (i = 0; i < nritems; i++) {
2567                 if (level == 0) {
2568                         btrfs_item_key_to_cpu(buf, &key, i);
2569                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2570                                 continue;
2571                         fi = btrfs_item_ptr(buf, i,
2572                                             struct btrfs_file_extent_item);
2573                         if (btrfs_file_extent_type(buf, fi) ==
2574                             BTRFS_FILE_EXTENT_INLINE)
2575                                 continue;
2576                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2577                         if (bytenr == 0)
2578                                 continue;
2579
2580                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2581                         key.offset -= btrfs_file_extent_offset(buf, fi);
2582                         ret = process_func(trans, root, bytenr, num_bytes,
2583                                            parent, ref_root, key.objectid,
2584                                            key.offset);
2585                         if (ret)
2586                                 goto fail;
2587                 } else {
2588                         bytenr = btrfs_node_blockptr(buf, i);
2589                         num_bytes = btrfs_level_size(root, level - 1);
2590                         ret = process_func(trans, root, bytenr, num_bytes,
2591                                            parent, ref_root, level - 1, 0);
2592                         if (ret)
2593                                 goto fail;
2594                 }
2595         }
2596         return 0;
2597 fail:
2598         BUG();
2599         return ret;
2600 }
2601
2602 int btrfs_inc_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, 1);
2606 }
2607
2608 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2609                   struct extent_buffer *buf, int full_backref)
2610 {
2611         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2612 }
2613
2614 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2615                                  struct btrfs_root *root,
2616                                  struct btrfs_path *path,
2617                                  struct btrfs_block_group_cache *cache)
2618 {
2619         int ret;
2620         struct btrfs_root *extent_root = root->fs_info->extent_root;
2621         unsigned long bi;
2622         struct extent_buffer *leaf;
2623
2624         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2625         if (ret < 0)
2626                 goto fail;
2627         BUG_ON(ret);
2628
2629         leaf = path->nodes[0];
2630         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2631         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2632         btrfs_mark_buffer_dirty(leaf);
2633         btrfs_release_path(path);
2634 fail:
2635         if (ret)
2636                 return ret;
2637         return 0;
2638
2639 }
2640
2641 static struct btrfs_block_group_cache *
2642 next_block_group(struct btrfs_root *root,
2643                  struct btrfs_block_group_cache *cache)
2644 {
2645         struct rb_node *node;
2646         spin_lock(&root->fs_info->block_group_cache_lock);
2647         node = rb_next(&cache->cache_node);
2648         btrfs_put_block_group(cache);
2649         if (node) {
2650                 cache = rb_entry(node, struct btrfs_block_group_cache,
2651                                  cache_node);
2652                 btrfs_get_block_group(cache);
2653         } else
2654                 cache = NULL;
2655         spin_unlock(&root->fs_info->block_group_cache_lock);
2656         return cache;
2657 }
2658
2659 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2660                             struct btrfs_trans_handle *trans,
2661                             struct btrfs_path *path)
2662 {
2663         struct btrfs_root *root = block_group->fs_info->tree_root;
2664         struct inode *inode = NULL;
2665         u64 alloc_hint = 0;
2666         int dcs = BTRFS_DC_ERROR;
2667         int num_pages = 0;
2668         int retries = 0;
2669         int ret = 0;
2670
2671         /*
2672          * If this block group is smaller than 100 megs don't bother caching the
2673          * block group.
2674          */
2675         if (block_group->key.offset < (100 * 1024 * 1024)) {
2676                 spin_lock(&block_group->lock);
2677                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2678                 spin_unlock(&block_group->lock);
2679                 return 0;
2680         }
2681
2682 again:
2683         inode = lookup_free_space_inode(root, block_group, path);
2684         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2685                 ret = PTR_ERR(inode);
2686                 btrfs_release_path(path);
2687                 goto out;
2688         }
2689
2690         if (IS_ERR(inode)) {
2691                 BUG_ON(retries);
2692                 retries++;
2693
2694                 if (block_group->ro)
2695                         goto out_free;
2696
2697                 ret = create_free_space_inode(root, trans, block_group, path);
2698                 if (ret)
2699                         goto out_free;
2700                 goto again;
2701         }
2702
2703         /*
2704          * We want to set the generation to 0, that way if anything goes wrong
2705          * from here on out we know not to trust this cache when we load up next
2706          * time.
2707          */
2708         BTRFS_I(inode)->generation = 0;
2709         ret = btrfs_update_inode(trans, root, inode);
2710         WARN_ON(ret);
2711
2712         if (i_size_read(inode) > 0) {
2713                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2714                                                       inode);
2715                 if (ret)
2716                         goto out_put;
2717         }
2718
2719         spin_lock(&block_group->lock);
2720         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2721                 /* We're not cached, don't bother trying to write stuff out */
2722                 dcs = BTRFS_DC_WRITTEN;
2723                 spin_unlock(&block_group->lock);
2724                 goto out_put;
2725         }
2726         spin_unlock(&block_group->lock);
2727
2728         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2729         if (!num_pages)
2730                 num_pages = 1;
2731
2732         /*
2733          * Just to make absolutely sure we have enough space, we're going to
2734          * preallocate 12 pages worth of space for each block group.  In
2735          * practice we ought to use at most 8, but we need extra space so we can
2736          * add our header and have a terminator between the extents and the
2737          * bitmaps.
2738          */
2739         num_pages *= 16;
2740         num_pages *= PAGE_CACHE_SIZE;
2741
2742         ret = btrfs_check_data_free_space(inode, num_pages);
2743         if (ret)
2744                 goto out_put;
2745
2746         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2747                                               num_pages, num_pages,
2748                                               &alloc_hint);
2749         if (!ret)
2750                 dcs = BTRFS_DC_SETUP;
2751         btrfs_free_reserved_data_space(inode, num_pages);
2752 out_put:
2753         iput(inode);
2754 out_free:
2755         btrfs_release_path(path);
2756 out:
2757         spin_lock(&block_group->lock);
2758         block_group->disk_cache_state = dcs;
2759         spin_unlock(&block_group->lock);
2760
2761         return ret;
2762 }
2763
2764 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2765                                    struct btrfs_root *root)
2766 {
2767         struct btrfs_block_group_cache *cache;
2768         int err = 0;
2769         struct btrfs_path *path;
2770         u64 last = 0;
2771
2772         path = btrfs_alloc_path();
2773         if (!path)
2774                 return -ENOMEM;
2775
2776 again:
2777         while (1) {
2778                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2779                 while (cache) {
2780                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2781                                 break;
2782                         cache = next_block_group(root, cache);
2783                 }
2784                 if (!cache) {
2785                         if (last == 0)
2786                                 break;
2787                         last = 0;
2788                         continue;
2789                 }
2790                 err = cache_save_setup(cache, trans, path);
2791                 last = cache->key.objectid + cache->key.offset;
2792                 btrfs_put_block_group(cache);
2793         }
2794
2795         while (1) {
2796                 if (last == 0) {
2797                         err = btrfs_run_delayed_refs(trans, root,
2798                                                      (unsigned long)-1);
2799                         BUG_ON(err);
2800                 }
2801
2802                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2803                 while (cache) {
2804                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2805                                 btrfs_put_block_group(cache);
2806                                 goto again;
2807                         }
2808
2809                         if (cache->dirty)
2810                                 break;
2811                         cache = next_block_group(root, cache);
2812                 }
2813                 if (!cache) {
2814                         if (last == 0)
2815                                 break;
2816                         last = 0;
2817                         continue;
2818                 }
2819
2820                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2821                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2822                 cache->dirty = 0;
2823                 last = cache->key.objectid + cache->key.offset;
2824
2825                 err = write_one_cache_group(trans, root, path, cache);
2826                 BUG_ON(err);
2827                 btrfs_put_block_group(cache);
2828         }
2829
2830         while (1) {
2831                 /*
2832                  * I don't think this is needed since we're just marking our
2833                  * preallocated extent as written, but just in case it can't
2834                  * hurt.
2835                  */
2836                 if (last == 0) {
2837                         err = btrfs_run_delayed_refs(trans, root,
2838                                                      (unsigned long)-1);
2839                         BUG_ON(err);
2840                 }
2841
2842                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2843                 while (cache) {
2844                         /*
2845                          * Really this shouldn't happen, but it could if we
2846                          * couldn't write the entire preallocated extent and
2847                          * splitting the extent resulted in a new block.
2848                          */
2849                         if (cache->dirty) {
2850                                 btrfs_put_block_group(cache);
2851                                 goto again;
2852                         }
2853                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2854                                 break;
2855                         cache = next_block_group(root, cache);
2856                 }
2857                 if (!cache) {
2858                         if (last == 0)
2859                                 break;
2860                         last = 0;
2861                         continue;
2862                 }
2863
2864                 btrfs_write_out_cache(root, trans, cache, path);
2865
2866                 /*
2867                  * If we didn't have an error then the cache state is still
2868                  * NEED_WRITE, so we can set it to WRITTEN.
2869                  */
2870                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2871                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2872                 last = cache->key.objectid + cache->key.offset;
2873                 btrfs_put_block_group(cache);
2874         }
2875
2876         btrfs_free_path(path);
2877         return 0;
2878 }
2879
2880 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2881 {
2882         struct btrfs_block_group_cache *block_group;
2883         int readonly = 0;
2884
2885         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2886         if (!block_group || block_group->ro)
2887                 readonly = 1;
2888         if (block_group)
2889                 btrfs_put_block_group(block_group);
2890         return readonly;
2891 }
2892
2893 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2894                              u64 total_bytes, u64 bytes_used,
2895                              struct btrfs_space_info **space_info)
2896 {
2897         struct btrfs_space_info *found;
2898         int i;
2899         int factor;
2900
2901         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2902                      BTRFS_BLOCK_GROUP_RAID10))
2903                 factor = 2;
2904         else
2905                 factor = 1;
2906
2907         found = __find_space_info(info, flags);
2908         if (found) {
2909                 spin_lock(&found->lock);
2910                 found->total_bytes += total_bytes;
2911                 found->disk_total += total_bytes * factor;
2912                 found->bytes_used += bytes_used;
2913                 found->disk_used += bytes_used * factor;
2914                 found->full = 0;
2915                 spin_unlock(&found->lock);
2916                 *space_info = found;
2917                 return 0;
2918         }
2919         found = kzalloc(sizeof(*found), GFP_NOFS);
2920         if (!found)
2921                 return -ENOMEM;
2922
2923         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2924                 INIT_LIST_HEAD(&found->block_groups[i]);
2925         init_rwsem(&found->groups_sem);
2926         spin_lock_init(&found->lock);
2927         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2928                                 BTRFS_BLOCK_GROUP_SYSTEM |
2929                                 BTRFS_BLOCK_GROUP_METADATA);
2930         found->total_bytes = total_bytes;
2931         found->disk_total = total_bytes * factor;
2932         found->bytes_used = bytes_used;
2933         found->disk_used = bytes_used * factor;
2934         found->bytes_pinned = 0;
2935         found->bytes_reserved = 0;
2936         found->bytes_readonly = 0;
2937         found->bytes_may_use = 0;
2938         found->full = 0;
2939         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
2940         found->chunk_alloc = 0;
2941         found->flush = 0;
2942         init_waitqueue_head(&found->wait);
2943         *space_info = found;
2944         list_add_rcu(&found->list, &info->space_info);
2945         return 0;
2946 }
2947
2948 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2949 {
2950         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2951                                    BTRFS_BLOCK_GROUP_RAID1 |
2952                                    BTRFS_BLOCK_GROUP_RAID10 |
2953                                    BTRFS_BLOCK_GROUP_DUP);
2954         if (extra_flags) {
2955                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2956                         fs_info->avail_data_alloc_bits |= extra_flags;
2957                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2958                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2959                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2960                         fs_info->avail_system_alloc_bits |= extra_flags;
2961         }
2962 }
2963
2964 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2965 {
2966         /*
2967          * we add in the count of missing devices because we want
2968          * to make sure that any RAID levels on a degraded FS
2969          * continue to be honored.
2970          */
2971         u64 num_devices = root->fs_info->fs_devices->rw_devices +
2972                 root->fs_info->fs_devices->missing_devices;
2973
2974         if (num_devices == 1)
2975                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2976         if (num_devices < 4)
2977                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2978
2979         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2980             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2981                       BTRFS_BLOCK_GROUP_RAID10))) {
2982                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2983         }
2984
2985         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2986             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2987                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2988         }
2989
2990         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2991             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2992              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2993              (flags & BTRFS_BLOCK_GROUP_DUP)))
2994                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2995         return flags;
2996 }
2997
2998 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
2999 {
3000         if (flags & BTRFS_BLOCK_GROUP_DATA)
3001                 flags |= root->fs_info->avail_data_alloc_bits &
3002                          root->fs_info->data_alloc_profile;
3003         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3004                 flags |= root->fs_info->avail_system_alloc_bits &
3005                          root->fs_info->system_alloc_profile;
3006         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3007                 flags |= root->fs_info->avail_metadata_alloc_bits &
3008                          root->fs_info->metadata_alloc_profile;
3009         return btrfs_reduce_alloc_profile(root, flags);
3010 }
3011
3012 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3013 {
3014         u64 flags;
3015
3016         if (data)
3017                 flags = BTRFS_BLOCK_GROUP_DATA;
3018         else if (root == root->fs_info->chunk_root)
3019                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3020         else
3021                 flags = BTRFS_BLOCK_GROUP_METADATA;
3022
3023         return get_alloc_profile(root, flags);
3024 }
3025
3026 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3027 {
3028         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3029                                                        BTRFS_BLOCK_GROUP_DATA);
3030 }
3031
3032 /*
3033  * This will check the space that the inode allocates from to make sure we have
3034  * enough space for bytes.
3035  */
3036 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3037 {
3038         struct btrfs_space_info *data_sinfo;
3039         struct btrfs_root *root = BTRFS_I(inode)->root;
3040         u64 used;
3041         int ret = 0, committed = 0, alloc_chunk = 1;
3042
3043         /* make sure bytes are sectorsize aligned */
3044         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3045
3046         if (root == root->fs_info->tree_root ||
3047             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3048                 alloc_chunk = 0;
3049                 committed = 1;
3050         }
3051
3052         data_sinfo = BTRFS_I(inode)->space_info;
3053         if (!data_sinfo)
3054                 goto alloc;
3055
3056 again:
3057         /* make sure we have enough space to handle the data first */
3058         spin_lock(&data_sinfo->lock);
3059         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3060                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3061                 data_sinfo->bytes_may_use;
3062
3063         if (used + bytes > data_sinfo->total_bytes) {
3064                 struct btrfs_trans_handle *trans;
3065
3066                 /*
3067                  * if we don't have enough free bytes in this space then we need
3068                  * to alloc a new chunk.
3069                  */
3070                 if (!data_sinfo->full && alloc_chunk) {
3071                         u64 alloc_target;
3072
3073                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3074                         spin_unlock(&data_sinfo->lock);
3075 alloc:
3076                         alloc_target = btrfs_get_alloc_profile(root, 1);
3077                         trans = btrfs_join_transaction(root);
3078                         if (IS_ERR(trans))
3079                                 return PTR_ERR(trans);
3080
3081                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3082                                              bytes + 2 * 1024 * 1024,
3083                                              alloc_target,
3084                                              CHUNK_ALLOC_NO_FORCE);
3085                         btrfs_end_transaction(trans, root);
3086                         if (ret < 0) {
3087                                 if (ret != -ENOSPC)
3088                                         return ret;
3089                                 else
3090                                         goto commit_trans;
3091                         }
3092
3093                         if (!data_sinfo) {
3094                                 btrfs_set_inode_space_info(root, inode);
3095                                 data_sinfo = BTRFS_I(inode)->space_info;
3096                         }
3097                         goto again;
3098                 }
3099
3100                 /*
3101                  * If we have less pinned bytes than we want to allocate then
3102                  * don't bother committing the transaction, it won't help us.
3103                  */
3104                 if (data_sinfo->bytes_pinned < bytes)
3105                         committed = 1;
3106                 spin_unlock(&data_sinfo->lock);
3107
3108                 /* commit the current transaction and try again */
3109 commit_trans:
3110                 if (!committed &&
3111                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3112                         committed = 1;
3113                         trans = btrfs_join_transaction(root);
3114                         if (IS_ERR(trans))
3115                                 return PTR_ERR(trans);
3116                         ret = btrfs_commit_transaction(trans, root);
3117                         if (ret)
3118                                 return ret;
3119                         goto again;
3120                 }
3121
3122                 return -ENOSPC;
3123         }
3124         data_sinfo->bytes_may_use += bytes;
3125         BTRFS_I(inode)->reserved_bytes += bytes;
3126         spin_unlock(&data_sinfo->lock);
3127
3128         return 0;
3129 }
3130
3131 /*
3132  * called when we are clearing an delalloc extent from the
3133  * inode's io_tree or there was an error for whatever reason
3134  * after calling btrfs_check_data_free_space
3135  */
3136 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3137 {
3138         struct btrfs_root *root = BTRFS_I(inode)->root;
3139         struct btrfs_space_info *data_sinfo;
3140
3141         /* make sure bytes are sectorsize aligned */
3142         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3143
3144         data_sinfo = BTRFS_I(inode)->space_info;
3145         spin_lock(&data_sinfo->lock);
3146         data_sinfo->bytes_may_use -= bytes;
3147         BTRFS_I(inode)->reserved_bytes -= bytes;
3148         spin_unlock(&data_sinfo->lock);
3149 }
3150
3151 static void force_metadata_allocation(struct btrfs_fs_info *info)
3152 {
3153         struct list_head *head = &info->space_info;
3154         struct btrfs_space_info *found;
3155
3156         rcu_read_lock();
3157         list_for_each_entry_rcu(found, head, list) {
3158                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3159                         found->force_alloc = CHUNK_ALLOC_FORCE;
3160         }
3161         rcu_read_unlock();
3162 }
3163
3164 static int should_alloc_chunk(struct btrfs_root *root,
3165                               struct btrfs_space_info *sinfo, u64 alloc_bytes,
3166                               int force)
3167 {
3168         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3169         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3170         u64 thresh;
3171
3172         if (force == CHUNK_ALLOC_FORCE)
3173                 return 1;
3174
3175         /*
3176          * in limited mode, we want to have some free space up to
3177          * about 1% of the FS size.
3178          */
3179         if (force == CHUNK_ALLOC_LIMITED) {
3180                 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3181                 thresh = max_t(u64, 64 * 1024 * 1024,
3182                                div_factor_fine(thresh, 1));
3183
3184                 if (num_bytes - num_allocated < thresh)
3185                         return 1;
3186         }
3187
3188         /*
3189          * we have two similar checks here, one based on percentage
3190          * and once based on a hard number of 256MB.  The idea
3191          * is that if we have a good amount of free
3192          * room, don't allocate a chunk.  A good mount is
3193          * less than 80% utilized of the chunks we have allocated,
3194          * or more than 256MB free
3195          */
3196         if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3197                 return 0;
3198
3199         if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3200                 return 0;
3201
3202         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3203
3204         /* 256MB or 5% of the FS */
3205         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3206
3207         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3208                 return 0;
3209         return 1;
3210 }
3211
3212 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3213                           struct btrfs_root *extent_root, u64 alloc_bytes,
3214                           u64 flags, int force)
3215 {
3216         struct btrfs_space_info *space_info;
3217         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3218         int wait_for_alloc = 0;
3219         int ret = 0;
3220
3221         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3222
3223         space_info = __find_space_info(extent_root->fs_info, flags);
3224         if (!space_info) {
3225                 ret = update_space_info(extent_root->fs_info, flags,
3226                                         0, 0, &space_info);
3227                 BUG_ON(ret);
3228         }
3229         BUG_ON(!space_info);
3230
3231 again:
3232         spin_lock(&space_info->lock);
3233         if (space_info->force_alloc)
3234                 force = space_info->force_alloc;
3235         if (space_info->full) {
3236                 spin_unlock(&space_info->lock);
3237                 return 0;
3238         }
3239
3240         if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3241                 spin_unlock(&space_info->lock);
3242                 return 0;
3243         } else if (space_info->chunk_alloc) {
3244                 wait_for_alloc = 1;
3245         } else {
3246                 space_info->chunk_alloc = 1;
3247         }
3248
3249         spin_unlock(&space_info->lock);
3250
3251         mutex_lock(&fs_info->chunk_mutex);
3252
3253         /*
3254          * The chunk_mutex is held throughout the entirety of a chunk
3255          * allocation, so once we've acquired the chunk_mutex we know that the
3256          * other guy is done and we need to recheck and see if we should
3257          * allocate.
3258          */
3259         if (wait_for_alloc) {
3260                 mutex_unlock(&fs_info->chunk_mutex);
3261                 wait_for_alloc = 0;
3262                 goto again;
3263         }
3264
3265         /*
3266          * If we have mixed data/metadata chunks we want to make sure we keep
3267          * allocating mixed chunks instead of individual chunks.
3268          */
3269         if (btrfs_mixed_space_info(space_info))
3270                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3271
3272         /*
3273          * if we're doing a data chunk, go ahead and make sure that
3274          * we keep a reasonable number of metadata chunks allocated in the
3275          * FS as well.
3276          */
3277         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3278                 fs_info->data_chunk_allocations++;
3279                 if (!(fs_info->data_chunk_allocations %
3280                       fs_info->metadata_ratio))
3281                         force_metadata_allocation(fs_info);
3282         }
3283
3284         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3285         if (ret < 0 && ret != -ENOSPC)
3286                 goto out;
3287
3288         spin_lock(&space_info->lock);
3289         if (ret)
3290                 space_info->full = 1;
3291         else
3292                 ret = 1;
3293
3294         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3295         space_info->chunk_alloc = 0;
3296         spin_unlock(&space_info->lock);
3297 out:
3298         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3299         return ret;
3300 }
3301
3302 /*
3303  * shrink metadata reservation for delalloc
3304  */
3305 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3306                            struct btrfs_root *root, u64 to_reclaim, int sync)
3307 {
3308         struct btrfs_block_rsv *block_rsv;
3309         struct btrfs_space_info *space_info;
3310         u64 reserved;
3311         u64 max_reclaim;
3312         u64 reclaimed = 0;
3313         long time_left;
3314         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3315         int loops = 0;
3316         unsigned long progress;
3317
3318         block_rsv = &root->fs_info->delalloc_block_rsv;
3319         space_info = block_rsv->space_info;
3320
3321         smp_mb();
3322         reserved = space_info->bytes_reserved;
3323         progress = space_info->reservation_progress;
3324
3325         if (reserved == 0)
3326                 return 0;
3327
3328         smp_mb();
3329         if (root->fs_info->delalloc_bytes == 0) {
3330                 if (trans)
3331                         return 0;
3332                 btrfs_wait_ordered_extents(root, 0, 0);
3333                 return 0;
3334         }
3335
3336         max_reclaim = min(reserved, to_reclaim);
3337
3338         while (loops < 1024) {
3339                 /* have the flusher threads jump in and do some IO */
3340                 smp_mb();
3341                 nr_pages = min_t(unsigned long, nr_pages,
3342                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3343                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3344                                                 WB_REASON_FS_FREE_SPACE);
3345
3346                 spin_lock(&space_info->lock);
3347                 if (reserved > space_info->bytes_reserved)
3348                         reclaimed += reserved - space_info->bytes_reserved;
3349                 reserved = space_info->bytes_reserved;
3350                 spin_unlock(&space_info->lock);
3351
3352                 loops++;
3353
3354                 if (reserved == 0 || reclaimed >= max_reclaim)
3355                         break;
3356
3357                 if (trans && trans->transaction->blocked)
3358                         return -EAGAIN;
3359
3360                 time_left = schedule_timeout_interruptible(1);
3361
3362                 /* We were interrupted, exit */
3363                 if (time_left)
3364                         break;
3365
3366                 /* we've kicked the IO a few times, if anything has been freed,
3367                  * exit.  There is no sense in looping here for a long time
3368                  * when we really need to commit the transaction, or there are
3369                  * just too many writers without enough free space
3370                  */
3371
3372                 if (loops > 3) {
3373                         smp_mb();
3374                         if (progress != space_info->reservation_progress)
3375                                 break;
3376                 }
3377
3378         }
3379         if (reclaimed >= to_reclaim && !trans)
3380                 btrfs_wait_ordered_extents(root, 0, 0);
3381         return reclaimed >= to_reclaim;
3382 }
3383
3384 /*
3385  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3386  * idea is if this is the first call (retries == 0) then we will add to our
3387  * reserved count if we can't make the allocation in order to hold our place
3388  * while we go and try and free up space.  That way for retries > 1 we don't try
3389  * and add space, we just check to see if the amount of unused space is >= the
3390  * total space, meaning that our reservation is valid.
3391  *
3392  * However if we don't intend to retry this reservation, pass -1 as retries so
3393  * that it short circuits this logic.
3394  */
3395 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3396                                   struct btrfs_root *root,
3397                                   struct btrfs_block_rsv *block_rsv,
3398                                   u64 orig_bytes, int flush)
3399 {
3400         struct btrfs_space_info *space_info = block_rsv->space_info;
3401         u64 unused;
3402         u64 num_bytes = orig_bytes;
3403         int retries = 0;
3404         int ret = 0;
3405         bool committed = false;
3406         bool flushing = false;
3407
3408 again:
3409         ret = 0;
3410         spin_lock(&space_info->lock);
3411         /*
3412          * We only want to wait if somebody other than us is flushing and we are
3413          * actually alloed to flush.
3414          */
3415         while (flush && !flushing && space_info->flush) {
3416                 spin_unlock(&space_info->lock);
3417                 /*
3418                  * If we have a trans handle we can't wait because the flusher
3419                  * may have to commit the transaction, which would mean we would
3420                  * deadlock since we are waiting for the flusher to finish, but
3421                  * hold the current transaction open.
3422                  */
3423                 if (trans)
3424                         return -EAGAIN;
3425                 ret = wait_event_interruptible(space_info->wait,
3426                                                !space_info->flush);
3427                 /* Must have been interrupted, return */
3428                 if (ret)
3429                         return -EINTR;
3430
3431                 spin_lock(&space_info->lock);
3432         }
3433
3434         ret = -ENOSPC;
3435         unused = space_info->bytes_used + space_info->bytes_reserved +
3436                  space_info->bytes_pinned + space_info->bytes_readonly +
3437                  space_info->bytes_may_use;
3438
3439         /*
3440          * The idea here is that we've not already over-reserved the block group
3441          * then we can go ahead and save our reservation first and then start
3442          * flushing if we need to.  Otherwise if we've already overcommitted
3443          * lets start flushing stuff first and then come back and try to make
3444          * our reservation.
3445          */
3446         if (unused <= space_info->total_bytes) {
3447                 unused = space_info->total_bytes - unused;
3448                 if (unused >= num_bytes) {
3449                         space_info->bytes_reserved += orig_bytes;
3450                         ret = 0;
3451                 } else {
3452                         /*
3453                          * Ok set num_bytes to orig_bytes since we aren't
3454                          * overocmmitted, this way we only try and reclaim what
3455                          * we need.
3456                          */
3457                         num_bytes = orig_bytes;
3458                 }
3459         } else {
3460                 /*
3461                  * Ok we're over committed, set num_bytes to the overcommitted
3462                  * amount plus the amount of bytes that we need for this
3463                  * reservation.
3464                  */
3465                 num_bytes = unused - space_info->total_bytes +
3466                         (orig_bytes * (retries + 1));
3467         }
3468
3469         /*
3470          * Couldn't make our reservation, save our place so while we're trying
3471          * to reclaim space we can actually use it instead of somebody else
3472          * stealing it from us.
3473          */
3474         if (ret && flush) {
3475                 flushing = true;
3476                 space_info->flush = 1;
3477         }
3478
3479         spin_unlock(&space_info->lock);
3480
3481         if (!ret || !flush)
3482                 goto out;
3483
3484         /*
3485          * We do synchronous shrinking since we don't actually unreserve
3486          * metadata until after the IO is completed.
3487          */
3488         ret = shrink_delalloc(trans, root, num_bytes, 1);
3489         if (ret < 0)
3490                 goto out;
3491
3492         ret = 0;
3493
3494         /*
3495          * So if we were overcommitted it's possible that somebody else flushed
3496          * out enough space and we simply didn't have enough space to reclaim,
3497          * so go back around and try again.
3498          */
3499         if (retries < 2) {
3500                 retries++;
3501                 goto again;
3502         }
3503
3504         /*
3505          * Not enough space to be reclaimed, don't bother committing the
3506          * transaction.
3507          */
3508         spin_lock(&space_info->lock);
3509         if (space_info->bytes_pinned < orig_bytes)
3510                 ret = -ENOSPC;
3511         spin_unlock(&space_info->lock);
3512         if (ret)
3513                 goto out;
3514
3515         ret = -EAGAIN;
3516         if (trans)
3517                 goto out;
3518
3519         ret = -ENOSPC;
3520         if (committed)
3521                 goto out;
3522
3523         trans = btrfs_join_transaction(root);
3524         if (IS_ERR(trans))
3525                 goto out;
3526         ret = btrfs_commit_transaction(trans, root);
3527         if (!ret) {
3528                 trans = NULL;
3529                 committed = true;
3530                 goto again;
3531         }
3532
3533 out:
3534         if (flushing) {
3535                 spin_lock(&space_info->lock);
3536                 space_info->flush = 0;
3537                 wake_up_all(&space_info->wait);
3538                 spin_unlock(&space_info->lock);
3539         }
3540         return ret;
3541 }
3542
3543 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3544                                              struct btrfs_root *root)
3545 {
3546         struct btrfs_block_rsv *block_rsv;
3547         if (root->ref_cows)
3548                 block_rsv = trans->block_rsv;
3549         else
3550                 block_rsv = root->block_rsv;
3551
3552         if (!block_rsv)
3553                 block_rsv = &root->fs_info->empty_block_rsv;
3554
3555         return block_rsv;
3556 }
3557
3558 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3559                                u64 num_bytes)
3560 {
3561         int ret = -ENOSPC;
3562         spin_lock(&block_rsv->lock);
3563         if (block_rsv->reserved >= num_bytes) {
3564                 block_rsv->reserved -= num_bytes;
3565                 if (block_rsv->reserved < block_rsv->size)
3566                         block_rsv->full = 0;
3567                 ret = 0;
3568         }
3569         spin_unlock(&block_rsv->lock);
3570         return ret;
3571 }
3572
3573 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3574                                 u64 num_bytes, int update_size)
3575 {
3576         spin_lock(&block_rsv->lock);
3577         block_rsv->reserved += num_bytes;
3578         if (update_size)
3579                 block_rsv->size += num_bytes;
3580         else if (block_rsv->reserved >= block_rsv->size)
3581                 block_rsv->full = 1;
3582         spin_unlock(&block_rsv->lock);
3583 }
3584
3585 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3586                                     struct btrfs_block_rsv *dest, u64 num_bytes)
3587 {
3588         struct btrfs_space_info *space_info = block_rsv->space_info;
3589
3590         spin_lock(&block_rsv->lock);
3591         if (num_bytes == (u64)-1)
3592                 num_bytes = block_rsv->size;
3593         block_rsv->size -= num_bytes;
3594         if (block_rsv->reserved >= block_rsv->size) {
3595                 num_bytes = block_rsv->reserved - block_rsv->size;
3596                 block_rsv->reserved = block_rsv->size;
3597                 block_rsv->full = 1;
3598         } else {
3599                 num_bytes = 0;
3600         }
3601         spin_unlock(&block_rsv->lock);
3602
3603         if (num_bytes > 0) {
3604                 if (dest) {
3605                         spin_lock(&dest->lock);
3606                         if (!dest->full) {
3607                                 u64 bytes_to_add;
3608
3609                                 bytes_to_add = dest->size - dest->reserved;
3610                                 bytes_to_add = min(num_bytes, bytes_to_add);
3611                                 dest->reserved += bytes_to_add;
3612                                 if (dest->reserved >= dest->size)
3613                                         dest->full = 1;
3614                                 num_bytes -= bytes_to_add;
3615                         }
3616                         spin_unlock(&dest->lock);
3617                 }
3618                 if (num_bytes) {
3619                         spin_lock(&space_info->lock);
3620                         space_info->bytes_reserved -= num_bytes;
3621                         space_info->reservation_progress++;
3622                         spin_unlock(&space_info->lock);
3623                 }
3624         }
3625 }
3626
3627 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3628                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3629 {
3630         int ret;
3631
3632         ret = block_rsv_use_bytes(src, num_bytes);
3633         if (ret)
3634                 return ret;
3635
3636         block_rsv_add_bytes(dst, num_bytes, 1);
3637         return 0;
3638 }
3639
3640 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3641 {
3642         memset(rsv, 0, sizeof(*rsv));
3643         spin_lock_init(&rsv->lock);
3644         atomic_set(&rsv->usage, 1);
3645         rsv->priority = 6;
3646         INIT_LIST_HEAD(&rsv->list);
3647 }
3648
3649 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3650 {
3651         struct btrfs_block_rsv *block_rsv;
3652         struct btrfs_fs_info *fs_info = root->fs_info;
3653
3654         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3655         if (!block_rsv)
3656                 return NULL;
3657
3658         btrfs_init_block_rsv(block_rsv);
3659         block_rsv->space_info = __find_space_info(fs_info,
3660                                                   BTRFS_BLOCK_GROUP_METADATA);
3661         return block_rsv;
3662 }
3663
3664 void btrfs_free_block_rsv(struct btrfs_root *root,
3665                           struct btrfs_block_rsv *rsv)
3666 {
3667         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3668                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3669                 if (!rsv->durable)
3670                         kfree(rsv);
3671         }
3672 }
3673
3674 /*
3675  * make the block_rsv struct be able to capture freed space.
3676  * the captured space will re-add to the the block_rsv struct
3677  * after transaction commit
3678  */
3679 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3680                                  struct btrfs_block_rsv *block_rsv)
3681 {
3682         block_rsv->durable = 1;
3683         mutex_lock(&fs_info->durable_block_rsv_mutex);
3684         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3685         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3686 }
3687
3688 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3689                         struct btrfs_root *root,
3690                         struct btrfs_block_rsv *block_rsv,
3691                         u64 num_bytes)
3692 {
3693         int ret;
3694
3695         if (num_bytes == 0)
3696                 return 0;
3697
3698         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3699         if (!ret) {
3700                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3701                 return 0;
3702         }
3703
3704         return ret;
3705 }
3706
3707 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3708                           struct btrfs_root *root,
3709                           struct btrfs_block_rsv *block_rsv,
3710                           u64 min_reserved, int min_factor)
3711 {
3712         u64 num_bytes = 0;
3713         int commit_trans = 0;
3714         int ret = -ENOSPC;
3715
3716         if (!block_rsv)
3717                 return 0;
3718
3719         spin_lock(&block_rsv->lock);
3720         if (min_factor > 0)
3721                 num_bytes = div_factor(block_rsv->size, min_factor);
3722         if (min_reserved > num_bytes)
3723                 num_bytes = min_reserved;
3724
3725         if (block_rsv->reserved >= num_bytes) {
3726                 ret = 0;
3727         } else {
3728                 num_bytes -= block_rsv->reserved;
3729                 if (block_rsv->durable &&
3730                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3731                         commit_trans = 1;
3732         }
3733         spin_unlock(&block_rsv->lock);
3734         if (!ret)
3735                 return 0;
3736
3737         if (block_rsv->refill_used) {
3738                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3739                                              num_bytes, 0);
3740                 if (!ret) {
3741                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3742                         return 0;
3743                 }
3744         }
3745
3746         if (commit_trans) {
3747                 if (trans)
3748                         return -EAGAIN;
3749                 trans = btrfs_join_transaction(root);
3750                 BUG_ON(IS_ERR(trans));
3751                 ret = btrfs_commit_transaction(trans, root);
3752                 return 0;
3753         }
3754
3755         return -ENOSPC;
3756 }
3757
3758 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3759                             struct btrfs_block_rsv *dst_rsv,
3760                             u64 num_bytes)
3761 {
3762         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3763 }
3764
3765 void btrfs_block_rsv_release(struct btrfs_root *root,
3766                              struct btrfs_block_rsv *block_rsv,
3767                              u64 num_bytes)
3768 {
3769         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3770         if (global_rsv->full || global_rsv == block_rsv ||
3771             block_rsv->space_info != global_rsv->space_info)
3772                 global_rsv = NULL;
3773         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3774 }
3775
3776 /*
3777  * helper to calculate size of global block reservation.
3778  * the desired value is sum of space used by extent tree,
3779  * checksum tree and root tree
3780  */
3781 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3782 {
3783         struct btrfs_space_info *sinfo;
3784         u64 num_bytes;
3785         u64 meta_used;
3786         u64 data_used;
3787         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3788
3789         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3790         spin_lock(&sinfo->lock);
3791         data_used = sinfo->bytes_used;
3792         spin_unlock(&sinfo->lock);
3793
3794         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3795         spin_lock(&sinfo->lock);
3796         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3797                 data_used = 0;
3798         meta_used = sinfo->bytes_used;
3799         spin_unlock(&sinfo->lock);
3800
3801         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3802                     csum_size * 2;
3803         num_bytes += div64_u64(data_used + meta_used, 50);
3804
3805         if (num_bytes * 3 > meta_used)
3806                 num_bytes = div64_u64(meta_used, 3);
3807
3808         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3809 }
3810
3811 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3812 {
3813         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3814         struct btrfs_space_info *sinfo = block_rsv->space_info;
3815         u64 num_bytes;
3816
3817         num_bytes = calc_global_metadata_size(fs_info);
3818
3819         spin_lock(&block_rsv->lock);
3820         spin_lock(&sinfo->lock);
3821
3822         block_rsv->size = num_bytes;
3823
3824         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3825                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3826                     sinfo->bytes_may_use;
3827
3828         if (sinfo->total_bytes > num_bytes) {
3829                 num_bytes = sinfo->total_bytes - num_bytes;
3830                 block_rsv->reserved += num_bytes;
3831                 sinfo->bytes_reserved += num_bytes;
3832         }
3833
3834         if (block_rsv->reserved >= block_rsv->size) {
3835                 num_bytes = block_rsv->reserved - block_rsv->size;
3836                 sinfo->bytes_reserved -= num_bytes;
3837                 sinfo->reservation_progress++;
3838                 block_rsv->reserved = block_rsv->size;
3839                 block_rsv->full = 1;
3840         }
3841
3842         spin_unlock(&sinfo->lock);
3843         spin_unlock(&block_rsv->lock);
3844 }
3845
3846 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3847 {
3848         struct btrfs_space_info *space_info;
3849
3850         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3851         fs_info->chunk_block_rsv.space_info = space_info;
3852         fs_info->chunk_block_rsv.priority = 10;
3853
3854         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3855         fs_info->global_block_rsv.space_info = space_info;
3856         fs_info->global_block_rsv.priority = 10;
3857         fs_info->global_block_rsv.refill_used = 1;
3858         fs_info->delalloc_block_rsv.space_info = space_info;
3859         fs_info->trans_block_rsv.space_info = space_info;
3860         fs_info->empty_block_rsv.space_info = space_info;
3861         fs_info->empty_block_rsv.priority = 10;
3862
3863         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3864         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3865         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3866         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3867         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3868
3869         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3870
3871         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3872
3873         update_global_block_rsv(fs_info);
3874 }
3875
3876 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3877 {
3878         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3879         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3880         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3881         WARN_ON(fs_info->trans_block_rsv.size > 0);
3882         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3883         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3884         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3885 }
3886
3887 int btrfs_truncate_reserve_metadata(struct btrfs_trans_handle *trans,
3888                                     struct btrfs_root *root,
3889                                     struct btrfs_block_rsv *rsv)
3890 {
3891         struct btrfs_block_rsv *trans_rsv = &root->fs_info->trans_block_rsv;
3892         u64 num_bytes;
3893         int ret;
3894
3895         /*
3896          * Truncate should be freeing data, but give us 2 items just in case it
3897          * needs to use some space.  We may want to be smarter about this in the
3898          * future.
3899          */
3900         num_bytes = btrfs_calc_trans_metadata_size(root, 2);
3901
3902         /* We already have enough bytes, just return */
3903         if (rsv->reserved >= num_bytes)
3904                 return 0;
3905
3906         num_bytes -= rsv->reserved;
3907
3908         /*
3909          * You should have reserved enough space before hand to do this, so this
3910          * should not fail.
3911          */
3912         ret = block_rsv_migrate_bytes(trans_rsv, rsv, num_bytes);
3913         BUG_ON(ret);
3914
3915         return 0;
3916 }
3917
3918 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3919                                   struct btrfs_root *root)
3920 {
3921         if (!trans->bytes_reserved)
3922                 return;
3923
3924         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3925         btrfs_block_rsv_release(root, trans->block_rsv,
3926                                 trans->bytes_reserved);
3927         trans->bytes_reserved = 0;
3928 }
3929
3930 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3931                                   struct inode *inode)
3932 {
3933         struct btrfs_root *root = BTRFS_I(inode)->root;
3934         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3935         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3936
3937         /*
3938          * We need to hold space in order to delete our orphan item once we've
3939          * added it, so this takes the reservation so we can release it later
3940          * when we are truly done with the orphan item.
3941          */
3942         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3943         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3944 }
3945
3946 void btrfs_orphan_release_metadata(struct inode *inode)
3947 {
3948         struct btrfs_root *root = BTRFS_I(inode)->root;
3949         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3950         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3951 }
3952
3953 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3954                                 struct btrfs_pending_snapshot *pending)
3955 {
3956         struct btrfs_root *root = pending->root;
3957         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3958         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3959         /*
3960          * two for root back/forward refs, two for directory entries
3961          * and one for root of the snapshot.
3962          */
3963         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3964         dst_rsv->space_info = src_rsv->space_info;
3965         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3966 }
3967
3968 static unsigned drop_outstanding_extent(struct inode *inode)
3969 {
3970         unsigned dropped_extents = 0;
3971
3972         spin_lock(&BTRFS_I(inode)->lock);
3973         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
3974         BTRFS_I(inode)->outstanding_extents--;
3975
3976         /*
3977          * If we have more or the same amount of outsanding extents than we have
3978          * reserved then we need to leave the reserved extents count alone.
3979          */
3980         if (BTRFS_I(inode)->outstanding_extents >=
3981             BTRFS_I(inode)->reserved_extents)
3982                 goto out;
3983
3984         dropped_extents = BTRFS_I(inode)->reserved_extents -
3985                 BTRFS_I(inode)->outstanding_extents;
3986         BTRFS_I(inode)->reserved_extents -= dropped_extents;
3987 out:
3988         spin_unlock(&BTRFS_I(inode)->lock);
3989         return dropped_extents;
3990 }
3991
3992 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3993 {
3994         return num_bytes >>= 3;
3995 }
3996
3997 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3998 {
3999         struct btrfs_root *root = BTRFS_I(inode)->root;
4000         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4001         u64 to_reserve = 0;
4002         unsigned nr_extents = 0;
4003         int ret;
4004
4005         if (btrfs_transaction_in_commit(root->fs_info))
4006                 schedule_timeout(1);
4007
4008         num_bytes = ALIGN(num_bytes, root->sectorsize);
4009
4010         spin_lock(&BTRFS_I(inode)->lock);
4011         BTRFS_I(inode)->outstanding_extents++;
4012
4013         if (BTRFS_I(inode)->outstanding_extents >
4014             BTRFS_I(inode)->reserved_extents) {
4015                 nr_extents = BTRFS_I(inode)->outstanding_extents -
4016                         BTRFS_I(inode)->reserved_extents;
4017                 BTRFS_I(inode)->reserved_extents += nr_extents;
4018
4019                 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4020         }
4021         spin_unlock(&BTRFS_I(inode)->lock);
4022
4023         to_reserve += calc_csum_metadata_size(inode, num_bytes);
4024         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4025         if (ret) {
4026                 unsigned dropped;
4027                 /*
4028                  * We don't need the return value since our reservation failed,
4029                  * we just need to clean up our counter.
4030                  */
4031                 dropped = drop_outstanding_extent(inode);
4032                 WARN_ON(dropped > 1);
4033                 return ret;
4034         }
4035
4036         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4037
4038         return 0;
4039 }
4040
4041 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4042 {
4043         struct btrfs_root *root = BTRFS_I(inode)->root;
4044         u64 to_free = 0;
4045         unsigned dropped;
4046
4047         num_bytes = ALIGN(num_bytes, root->sectorsize);
4048         dropped = drop_outstanding_extent(inode);
4049
4050         to_free = calc_csum_metadata_size(inode, num_bytes);
4051         if (dropped > 0)
4052                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4053
4054         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4055                                 to_free);
4056 }
4057
4058 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4059 {
4060         int ret;
4061
4062         ret = btrfs_check_data_free_space(inode, num_bytes);
4063         if (ret)
4064                 return ret;
4065
4066         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4067         if (ret) {
4068                 btrfs_free_reserved_data_space(inode, num_bytes);
4069                 return ret;
4070         }
4071
4072         return 0;
4073 }
4074
4075 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4076 {
4077         btrfs_delalloc_release_metadata(inode, num_bytes);
4078         btrfs_free_reserved_data_space(inode, num_bytes);
4079 }
4080
4081 static int update_block_group(struct btrfs_trans_handle *trans,
4082                               struct btrfs_root *root,
4083                               u64 bytenr, u64 num_bytes, int alloc)
4084 {
4085         struct btrfs_block_group_cache *cache = NULL;
4086         struct btrfs_fs_info *info = root->fs_info;
4087         u64 total = num_bytes;
4088         u64 old_val;
4089         u64 byte_in_group;
4090         int factor;
4091
4092         /* block accounting for super block */
4093         spin_lock(&info->delalloc_lock);
4094         old_val = btrfs_super_bytes_used(&info->super_copy);
4095         if (alloc)
4096                 old_val += num_bytes;
4097         else
4098                 old_val -= num_bytes;
4099         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4100         spin_unlock(&info->delalloc_lock);
4101
4102         while (total) {
4103                 cache = btrfs_lookup_block_group(info, bytenr);
4104                 if (!cache)
4105                         return -1;
4106                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4107                                     BTRFS_BLOCK_GROUP_RAID1 |
4108                                     BTRFS_BLOCK_GROUP_RAID10))
4109                         factor = 2;
4110                 else
4111                         factor = 1;
4112                 /*
4113                  * If this block group has free space cache written out, we
4114                  * need to make sure to load it if we are removing space.  This
4115                  * is because we need the unpinning stage to actually add the
4116                  * space back to the block group, otherwise we will leak space.
4117                  */
4118                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4119                         cache_block_group(cache, trans, NULL, 1);
4120
4121                 byte_in_group = bytenr - cache->key.objectid;
4122                 WARN_ON(byte_in_group > cache->key.offset);
4123
4124                 spin_lock(&cache->space_info->lock);
4125                 spin_lock(&cache->lock);
4126
4127                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4128                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4129                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4130
4131                 cache->dirty = 1;
4132                 old_val = btrfs_block_group_used(&cache->item);
4133                 num_bytes = min(total, cache->key.offset - byte_in_group);
4134                 if (alloc) {
4135                         old_val += num_bytes;
4136                         btrfs_set_block_group_used(&cache->item, old_val);
4137                         cache->reserved -= num_bytes;
4138                         cache->space_info->bytes_reserved -= num_bytes;
4139                         cache->space_info->reservation_progress++;
4140                         cache->space_info->bytes_used += num_bytes;
4141                         cache->space_info->disk_used += num_bytes * factor;
4142                         spin_unlock(&cache->lock);
4143                         spin_unlock(&cache->space_info->lock);
4144                 } else {
4145                         old_val -= num_bytes;
4146                         btrfs_set_block_group_used(&cache->item, old_val);
4147                         cache->pinned += num_bytes;
4148                         cache->space_info->bytes_pinned += num_bytes;
4149                         cache->space_info->bytes_used -= num_bytes;
4150                         cache->space_info->disk_used -= num_bytes * factor;
4151                         spin_unlock(&cache->lock);
4152                         spin_unlock(&cache->space_info->lock);
4153
4154                         set_extent_dirty(info->pinned_extents,
4155                                          bytenr, bytenr + num_bytes - 1,
4156                                          GFP_NOFS | __GFP_NOFAIL);
4157                 }
4158                 btrfs_put_block_group(cache);
4159                 total -= num_bytes;
4160                 bytenr += num_bytes;
4161         }
4162         return 0;
4163 }
4164
4165 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4166 {
4167         struct btrfs_block_group_cache *cache;
4168         u64 bytenr;
4169
4170         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4171         if (!cache)
4172                 return 0;
4173
4174         bytenr = cache->key.objectid;
4175         btrfs_put_block_group(cache);
4176
4177         return bytenr;
4178 }
4179
4180 static int pin_down_extent(struct btrfs_root *root,
4181                            struct btrfs_block_group_cache *cache,
4182                            u64 bytenr, u64 num_bytes, int reserved)
4183 {
4184         spin_lock(&cache->space_info->lock);
4185         spin_lock(&cache->lock);
4186         cache->pinned += num_bytes;
4187         cache->space_info->bytes_pinned += num_bytes;
4188         if (reserved) {
4189                 cache->reserved -= num_bytes;
4190                 cache->space_info->bytes_reserved -= num_bytes;
4191                 cache->space_info->reservation_progress++;
4192         }
4193         spin_unlock(&cache->lock);
4194         spin_unlock(&cache->space_info->lock);
4195
4196         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4197                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4198         return 0;
4199 }
4200
4201 /*
4202  * this function must be called within transaction
4203  */
4204 int btrfs_pin_extent(struct btrfs_root *root,
4205                      u64 bytenr, u64 num_bytes, int reserved)
4206 {
4207         struct btrfs_block_group_cache *cache;
4208
4209         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4210         BUG_ON(!cache);
4211
4212         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4213
4214         btrfs_put_block_group(cache);
4215         return 0;
4216 }
4217
4218 /*
4219  * update size of reserved extents. this function may return -EAGAIN
4220  * if 'reserve' is true or 'sinfo' is false.
4221  */
4222 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4223                                 u64 num_bytes, int reserve, int sinfo)
4224 {
4225         int ret = 0;
4226         if (sinfo) {
4227                 struct btrfs_space_info *space_info = cache->space_info;
4228                 spin_lock(&space_info->lock);
4229                 spin_lock(&cache->lock);
4230                 if (reserve) {
4231                         if (cache->ro) {
4232                                 ret = -EAGAIN;
4233                         } else {
4234                                 cache->reserved += num_bytes;
4235                                 space_info->bytes_reserved += num_bytes;
4236                         }
4237                 } else {
4238                         if (cache->ro)
4239                                 space_info->bytes_readonly += num_bytes;
4240                         cache->reserved -= num_bytes;
4241                         space_info->bytes_reserved -= num_bytes;
4242                         space_info->reservation_progress++;
4243                 }
4244                 spin_unlock(&cache->lock);
4245                 spin_unlock(&space_info->lock);
4246         } else {
4247                 spin_lock(&cache->lock);
4248                 if (cache->ro) {
4249                         ret = -EAGAIN;
4250                 } else {
4251                         if (reserve)
4252                                 cache->reserved += num_bytes;
4253                         else
4254                                 cache->reserved -= num_bytes;
4255                 }
4256                 spin_unlock(&cache->lock);
4257         }
4258         return ret;
4259 }
4260
4261 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4262                                 struct btrfs_root *root)
4263 {
4264         struct btrfs_fs_info *fs_info = root->fs_info;
4265         struct btrfs_caching_control *next;
4266         struct btrfs_caching_control *caching_ctl;
4267         struct btrfs_block_group_cache *cache;
4268
4269         down_write(&fs_info->extent_commit_sem);
4270
4271         list_for_each_entry_safe(caching_ctl, next,
4272                                  &fs_info->caching_block_groups, list) {
4273                 cache = caching_ctl->block_group;
4274                 if (block_group_cache_done(cache)) {
4275                         cache->last_byte_to_unpin = (u64)-1;
4276                         list_del_init(&caching_ctl->list);
4277                         put_caching_control(caching_ctl);
4278                 } else {
4279                         cache->last_byte_to_unpin = caching_ctl->progress;
4280                 }
4281         }
4282
4283         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4284                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4285         else
4286                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4287
4288         up_write(&fs_info->extent_commit_sem);
4289
4290         update_global_block_rsv(fs_info);
4291         return 0;
4292 }
4293
4294 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4295 {
4296         struct btrfs_fs_info *fs_info = root->fs_info;
4297         struct btrfs_block_group_cache *cache = NULL;
4298         u64 len;
4299
4300         while (start <= end) {
4301                 if (!cache ||
4302                     start >= cache->key.objectid + cache->key.offset) {
4303                         if (cache)
4304                                 btrfs_put_block_group(cache);
4305                         cache = btrfs_lookup_block_group(fs_info, start);
4306                         BUG_ON(!cache);
4307                 }
4308
4309                 len = cache->key.objectid + cache->key.offset - start;
4310                 len = min(len, end + 1 - start);
4311
4312                 if (start < cache->last_byte_to_unpin) {
4313                         len = min(len, cache->last_byte_to_unpin - start);
4314                         btrfs_add_free_space(cache, start, len);
4315                 }
4316
4317                 start += len;
4318
4319                 spin_lock(&cache->space_info->lock);
4320                 spin_lock(&cache->lock);
4321                 cache->pinned -= len;
4322                 cache->space_info->bytes_pinned -= len;
4323                 if (cache->ro) {
4324                         cache->space_info->bytes_readonly += len;
4325                 } else if (cache->reserved_pinned > 0) {
4326                         len = min(len, cache->reserved_pinned);
4327                         cache->reserved_pinned -= len;
4328                         cache->space_info->bytes_reserved += len;
4329                 }
4330                 spin_unlock(&cache->lock);
4331                 spin_unlock(&cache->space_info->lock);
4332         }
4333
4334         if (cache)
4335                 btrfs_put_block_group(cache);
4336         return 0;
4337 }
4338
4339 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4340                                struct btrfs_root *root)
4341 {
4342         struct btrfs_fs_info *fs_info = root->fs_info;
4343         struct extent_io_tree *unpin;
4344         struct btrfs_block_rsv *block_rsv;
4345         struct btrfs_block_rsv *next_rsv;
4346         u64 start;
4347         u64 end;
4348         int idx;
4349         int ret;
4350
4351         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4352                 unpin = &fs_info->freed_extents[1];
4353         else
4354                 unpin = &fs_info->freed_extents[0];
4355
4356         while (1) {
4357                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4358                                             EXTENT_DIRTY);
4359                 if (ret)
4360                         break;
4361
4362                 if (btrfs_test_opt(root, DISCARD))
4363                         ret = btrfs_discard_extent(root, start,
4364                                                    end + 1 - start, NULL);
4365
4366                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4367                 unpin_extent_range(root, start, end);
4368                 cond_resched();
4369         }
4370
4371         mutex_lock(&fs_info->durable_block_rsv_mutex);
4372         list_for_each_entry_safe(block_rsv, next_rsv,
4373                                  &fs_info->durable_block_rsv_list, list) {
4374
4375                 idx = trans->transid & 0x1;
4376                 if (block_rsv->freed[idx] > 0) {
4377                         block_rsv_add_bytes(block_rsv,
4378                                             block_rsv->freed[idx], 0);
4379                         block_rsv->freed[idx] = 0;
4380                 }
4381                 if (atomic_read(&block_rsv->usage) == 0) {
4382                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4383
4384                         if (block_rsv->freed[0] == 0 &&
4385                             block_rsv->freed[1] == 0) {
4386                                 list_del_init(&block_rsv->list);
4387                                 kfree(block_rsv);
4388                         }
4389                 } else {
4390                         btrfs_block_rsv_release(root, block_rsv, 0);
4391                 }
4392         }
4393         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4394
4395         return 0;
4396 }
4397
4398 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4399                                 struct btrfs_root *root,
4400                                 u64 bytenr, u64 num_bytes, u64 parent,
4401                                 u64 root_objectid, u64 owner_objectid,
4402                                 u64 owner_offset, int refs_to_drop,
4403                                 struct btrfs_delayed_extent_op *extent_op)
4404 {
4405         struct btrfs_key key;
4406         struct btrfs_path *path;
4407         struct btrfs_fs_info *info = root->fs_info;
4408         struct btrfs_root *extent_root = info->extent_root;
4409         struct extent_buffer *leaf;
4410         struct btrfs_extent_item *ei;
4411         struct btrfs_extent_inline_ref *iref;
4412         int ret;
4413         int is_data;
4414         int extent_slot = 0;
4415         int found_extent = 0;
4416         int num_to_del = 1;
4417         u32 item_size;
4418         u64 refs;
4419
4420         path = btrfs_alloc_path();
4421         if (!path)
4422                 return -ENOMEM;
4423
4424         path->reada = 1;
4425         path->leave_spinning = 1;
4426
4427         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4428         BUG_ON(!is_data && refs_to_drop != 1);
4429
4430         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4431                                     bytenr, num_bytes, parent,
4432                                     root_objectid, owner_objectid,
4433                                     owner_offset);
4434         if (ret == 0) {
4435                 extent_slot = path->slots[0];
4436                 while (extent_slot >= 0) {
4437                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4438                                               extent_slot);
4439                         if (key.objectid != bytenr)
4440                                 break;
4441                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4442                             key.offset == num_bytes) {
4443                                 found_extent = 1;
4444                                 break;
4445                         }
4446                         if (path->slots[0] - extent_slot > 5)
4447                                 break;
4448                         extent_slot--;
4449                 }
4450 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4451                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4452                 if (found_extent && item_size < sizeof(*ei))
4453                         found_extent = 0;
4454 #endif
4455                 if (!found_extent) {
4456                         BUG_ON(iref);
4457                         ret = remove_extent_backref(trans, extent_root, path,
4458                                                     NULL, refs_to_drop,
4459                                                     is_data);
4460                         BUG_ON(ret);
4461                         btrfs_release_path(path);
4462                         path->leave_spinning = 1;
4463
4464                         key.objectid = bytenr;
4465                         key.type = BTRFS_EXTENT_ITEM_KEY;
4466                         key.offset = num_bytes;
4467
4468                         ret = btrfs_search_slot(trans, extent_root,
4469                                                 &key, path, -1, 1);
4470                         if (ret) {
4471                                 printk(KERN_ERR "umm, got %d back from search"
4472                                        ", was looking for %llu\n", ret,
4473                                        (unsigned long long)bytenr);
4474                                 if (ret > 0)
4475                                         btrfs_print_leaf(extent_root,
4476                                                          path->nodes[0]);
4477                         }
4478                         BUG_ON(ret);
4479                         extent_slot = path->slots[0];
4480                 }
4481         } else {
4482                 btrfs_print_leaf(extent_root, path->nodes[0]);
4483                 WARN_ON(1);
4484                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4485                        "parent %llu root %llu  owner %llu offset %llu\n",
4486                        (unsigned long long)bytenr,
4487                        (unsigned long long)parent,
4488                        (unsigned long long)root_objectid,
4489                        (unsigned long long)owner_objectid,
4490                        (unsigned long long)owner_offset);
4491         }
4492
4493         leaf = path->nodes[0];
4494         item_size = btrfs_item_size_nr(leaf, extent_slot);
4495 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4496         if (item_size < sizeof(*ei)) {
4497                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4498                 ret = convert_extent_item_v0(trans, extent_root, path,
4499                                              owner_objectid, 0);
4500                 BUG_ON(ret < 0);
4501
4502                 btrfs_release_path(path);
4503                 path->leave_spinning = 1;
4504
4505                 key.objectid = bytenr;
4506                 key.type = BTRFS_EXTENT_ITEM_KEY;
4507                 key.offset = num_bytes;
4508
4509                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4510                                         -1, 1);
4511                 if (ret) {
4512                         printk(KERN_ERR "umm, got %d back from search"
4513                                ", was looking for %llu\n", ret,
4514                                (unsigned long long)bytenr);
4515                         btrfs_print_leaf(extent_root, path->nodes[0]);
4516                 }
4517                 BUG_ON(ret);
4518                 extent_slot = path->slots[0];
4519                 leaf = path->nodes[0];
4520                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4521         }
4522 #endif
4523         BUG_ON(item_size < sizeof(*ei));
4524         ei = btrfs_item_ptr(leaf, extent_slot,
4525                             struct btrfs_extent_item);
4526         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4527                 struct btrfs_tree_block_info *bi;
4528                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4529                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4530                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4531         }
4532
4533         refs = btrfs_extent_refs(leaf, ei);
4534         BUG_ON(refs < refs_to_drop);
4535         refs -= refs_to_drop;
4536
4537         if (refs > 0) {
4538                 if (extent_op)
4539                         __run_delayed_extent_op(extent_op, leaf, ei);
4540                 /*
4541                  * In the case of inline back ref, reference count will
4542                  * be updated by remove_extent_backref
4543                  */
4544                 if (iref) {
4545                         BUG_ON(!found_extent);
4546                 } else {
4547                         btrfs_set_extent_refs(leaf, ei, refs);
4548                         btrfs_mark_buffer_dirty(leaf);
4549                 }
4550                 if (found_extent) {
4551                         ret = remove_extent_backref(trans, extent_root, path,
4552                                                     iref, refs_to_drop,
4553                                                     is_data);
4554                         BUG_ON(ret);
4555                 }
4556         } else {
4557                 if (found_extent) {
4558                         BUG_ON(is_data && refs_to_drop !=
4559                                extent_data_ref_count(root, path, iref));
4560                         if (iref) {
4561                                 BUG_ON(path->slots[0] != extent_slot);
4562                         } else {
4563                                 BUG_ON(path->slots[0] != extent_slot + 1);
4564                                 path->slots[0] = extent_slot;
4565                                 num_to_del = 2;
4566                         }
4567                 }
4568
4569                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4570                                       num_to_del);
4571                 BUG_ON(ret);
4572                 btrfs_release_path(path);
4573
4574                 if (is_data) {
4575                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4576                         BUG_ON(ret);
4577                 } else {
4578                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4579                              bytenr >> PAGE_CACHE_SHIFT,
4580                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4581                 }
4582
4583                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4584                 BUG_ON(ret);
4585         }
4586         btrfs_free_path(path);
4587         return ret;
4588 }
4589
4590 /*
4591  * when we free an block, it is possible (and likely) that we free the last
4592  * delayed ref for that extent as well.  This searches the delayed ref tree for
4593  * a given extent, and if there are no other delayed refs to be processed, it
4594  * removes it from the tree.
4595  */
4596 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4597                                       struct btrfs_root *root, u64 bytenr)
4598 {
4599         struct btrfs_delayed_ref_head *head;
4600         struct btrfs_delayed_ref_root *delayed_refs;
4601         struct btrfs_delayed_ref_node *ref;
4602         struct rb_node *node;
4603         int ret = 0;
4604
4605         delayed_refs = &trans->transaction->delayed_refs;
4606         spin_lock(&delayed_refs->lock);
4607         head = btrfs_find_delayed_ref_head(trans, bytenr);
4608         if (!head)
4609                 goto out;
4610
4611         node = rb_prev(&head->node.rb_node);
4612         if (!node)
4613                 goto out;
4614
4615         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4616
4617         /* there are still entries for this ref, we can't drop it */
4618         if (ref->bytenr == bytenr)
4619                 goto out;
4620
4621         if (head->extent_op) {
4622                 if (!head->must_insert_reserved)
4623                         goto out;
4624                 kfree(head->extent_op);
4625                 head->extent_op = NULL;
4626         }
4627
4628         /*
4629          * waiting for the lock here would deadlock.  If someone else has it
4630          * locked they are already in the process of dropping it anyway
4631          */
4632         if (!mutex_trylock(&head->mutex))
4633                 goto out;
4634
4635         /*
4636          * at this point we have a head with no other entries.  Go
4637          * ahead and process it.
4638          */
4639         head->node.in_tree = 0;
4640         rb_erase(&head->node.rb_node, &delayed_refs->root);
4641
4642         delayed_refs->num_entries--;
4643
4644         /*
4645          * we don't take a ref on the node because we're removing it from the
4646          * tree, so we just steal the ref the tree was holding.
4647          */
4648         delayed_refs->num_heads--;
4649         if (list_empty(&head->cluster))
4650                 delayed_refs->num_heads_ready--;
4651
4652         list_del_init(&head->cluster);
4653         spin_unlock(&delayed_refs->lock);
4654
4655         BUG_ON(head->extent_op);
4656         if (head->must_insert_reserved)
4657                 ret = 1;
4658
4659         mutex_unlock(&head->mutex);
4660         btrfs_put_delayed_ref(&head->node);
4661         return ret;
4662 out:
4663         spin_unlock(&delayed_refs->lock);
4664         return 0;
4665 }
4666
4667 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4668                            struct btrfs_root *root,
4669                            struct extent_buffer *buf,
4670                            u64 parent, int last_ref)
4671 {
4672         struct btrfs_block_rsv *block_rsv;
4673         struct btrfs_block_group_cache *cache = NULL;
4674         int ret;
4675
4676         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4677                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4678                                                 parent, root->root_key.objectid,
4679                                                 btrfs_header_level(buf),
4680                                                 BTRFS_DROP_DELAYED_REF, NULL);
4681                 BUG_ON(ret);
4682         }
4683
4684         if (!last_ref)
4685                 return;
4686
4687         block_rsv = get_block_rsv(trans, root);
4688         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4689         if (block_rsv->space_info != cache->space_info)
4690                 goto out;
4691
4692         if (btrfs_header_generation(buf) == trans->transid) {
4693                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4694                         ret = check_ref_cleanup(trans, root, buf->start);
4695                         if (!ret)
4696                                 goto pin;
4697                 }
4698
4699                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4700                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4701                         goto pin;
4702                 }
4703
4704                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4705
4706                 btrfs_add_free_space(cache, buf->start, buf->len);
4707                 ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0);
4708                 if (ret == -EAGAIN) {
4709                         /* block group became read-only */
4710                         btrfs_update_reserved_bytes(cache, buf->len, 0, 1);
4711                         goto out;
4712                 }
4713
4714                 ret = 1;
4715                 spin_lock(&block_rsv->lock);
4716                 if (block_rsv->reserved < block_rsv->size) {
4717                         block_rsv->reserved += buf->len;
4718                         ret = 0;
4719                 }
4720                 spin_unlock(&block_rsv->lock);
4721
4722                 if (ret) {
4723                         spin_lock(&cache->space_info->lock);
4724                         cache->space_info->bytes_reserved -= buf->len;
4725                         cache->space_info->reservation_progress++;
4726                         spin_unlock(&cache->space_info->lock);
4727                 }
4728                 goto out;
4729         }
4730 pin:
4731         if (block_rsv->durable && !cache->ro) {
4732                 ret = 0;
4733                 spin_lock(&cache->lock);
4734                 if (!cache->ro) {
4735                         cache->reserved_pinned += buf->len;
4736                         ret = 1;
4737                 }
4738                 spin_unlock(&cache->lock);
4739
4740                 if (ret) {
4741                         spin_lock(&block_rsv->lock);
4742                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4743                         spin_unlock(&block_rsv->lock);
4744                 }
4745         }
4746 out:
4747         /*
4748          * Deleting the buffer, clear the corrupt flag since it doesn't matter
4749          * anymore.
4750          */
4751         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4752         btrfs_put_block_group(cache);
4753 }
4754
4755 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4756                       struct btrfs_root *root,
4757                       u64 bytenr, u64 num_bytes, u64 parent,
4758                       u64 root_objectid, u64 owner, u64 offset)
4759 {
4760         int ret;
4761
4762         /*
4763          * tree log blocks never actually go into the extent allocation
4764          * tree, just update pinning info and exit early.
4765          */
4766         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4767                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4768                 /* unlocks the pinned mutex */
4769                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4770                 ret = 0;
4771         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4772                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4773                                         parent, root_objectid, (int)owner,
4774                                         BTRFS_DROP_DELAYED_REF, NULL);
4775                 BUG_ON(ret);
4776         } else {
4777                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4778                                         parent, root_objectid, owner,
4779                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4780                 BUG_ON(ret);
4781         }
4782         return ret;
4783 }
4784
4785 static u64 stripe_align(struct btrfs_root *root, u64 val)
4786 {
4787         u64 mask = ((u64)root->stripesize - 1);
4788         u64 ret = (val + mask) & ~mask;
4789         return ret;
4790 }
4791
4792 /*
4793  * when we wait for progress in the block group caching, its because
4794  * our allocation attempt failed at least once.  So, we must sleep
4795  * and let some progress happen before we try again.
4796  *
4797  * This function will sleep at least once waiting for new free space to
4798  * show up, and then it will check the block group free space numbers
4799  * for our min num_bytes.  Another option is to have it go ahead
4800  * and look in the rbtree for a free extent of a given size, but this
4801  * is a good start.
4802  */
4803 static noinline int
4804 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4805                                 u64 num_bytes)
4806 {
4807         struct btrfs_caching_control *caching_ctl;
4808         DEFINE_WAIT(wait);
4809
4810         caching_ctl = get_caching_control(cache);
4811         if (!caching_ctl)
4812                 return 0;
4813
4814         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4815                    (cache->free_space_ctl->free_space >= num_bytes));
4816
4817         put_caching_control(caching_ctl);
4818         return 0;
4819 }
4820
4821 static noinline int
4822 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4823 {
4824         struct btrfs_caching_control *caching_ctl;
4825         DEFINE_WAIT(wait);
4826
4827         caching_ctl = get_caching_control(cache);
4828         if (!caching_ctl)
4829                 return 0;
4830
4831         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4832
4833         put_caching_control(caching_ctl);
4834         return 0;
4835 }
4836
4837 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4838 {
4839         int index;
4840         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4841                 index = 0;
4842         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4843                 index = 1;
4844         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4845                 index = 2;
4846         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4847                 index = 3;
4848         else
4849                 index = 4;
4850         return index;
4851 }
4852
4853 enum btrfs_loop_type {
4854         LOOP_FIND_IDEAL = 0,
4855         LOOP_CACHING_NOWAIT = 1,
4856         LOOP_CACHING_WAIT = 2,
4857         LOOP_ALLOC_CHUNK = 3,
4858         LOOP_NO_EMPTY_SIZE = 4,
4859 };
4860
4861 /*
4862  * walks the btree of allocated extents and find a hole of a given size.
4863  * The key ins is changed to record the hole:
4864  * ins->objectid == block start
4865  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4866  * ins->offset == number of blocks
4867  * Any available blocks before search_start are skipped.
4868  */
4869 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4870                                      struct btrfs_root *orig_root,
4871                                      u64 num_bytes, u64 empty_size,
4872                                      u64 search_start, u64 search_end,
4873                                      u64 hint_byte, struct btrfs_key *ins,
4874                                      u64 data)
4875 {
4876         int ret = 0;
4877         struct btrfs_root *root = orig_root->fs_info->extent_root;
4878         struct btrfs_free_cluster *last_ptr = NULL;
4879         struct btrfs_block_group_cache *block_group = NULL;
4880         int empty_cluster = 2 * 1024 * 1024;
4881         int allowed_chunk_alloc = 0;
4882         int done_chunk_alloc = 0;
4883         struct btrfs_space_info *space_info;
4884         int last_ptr_loop = 0;
4885         int loop = 0;
4886         int index = 0;
4887         bool found_uncached_bg = false;
4888         bool failed_cluster_refill = false;
4889         bool failed_alloc = false;
4890         bool use_cluster = true;
4891         u64 ideal_cache_percent = 0;
4892         u64 ideal_cache_offset = 0;
4893
4894         WARN_ON(num_bytes < root->sectorsize);
4895         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4896         ins->objectid = 0;
4897         ins->offset = 0;
4898
4899         space_info = __find_space_info(root->fs_info, data);
4900         if (!space_info) {
4901                 printk(KERN_ERR "No space info for %llu\n", data);
4902                 return -ENOSPC;
4903         }
4904
4905         /*
4906          * If the space info is for both data and metadata it means we have a
4907          * small filesystem and we can't use the clustering stuff.
4908          */
4909         if (btrfs_mixed_space_info(space_info))
4910                 use_cluster = false;
4911
4912         if (orig_root->ref_cows || empty_size)
4913                 allowed_chunk_alloc = 1;
4914
4915         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4916                 last_ptr = &root->fs_info->meta_alloc_cluster;
4917                 if (!btrfs_test_opt(root, SSD))
4918                         empty_cluster = 64 * 1024;
4919         }
4920
4921         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4922             btrfs_test_opt(root, SSD)) {
4923                 last_ptr = &root->fs_info->data_alloc_cluster;
4924         }
4925
4926         if (last_ptr) {
4927                 spin_lock(&last_ptr->lock);
4928                 if (last_ptr->block_group)
4929                         hint_byte = last_ptr->window_start;
4930                 spin_unlock(&last_ptr->lock);
4931         }
4932
4933         search_start = max(search_start, first_logical_byte(root, 0));
4934         search_start = max(search_start, hint_byte);
4935
4936         if (!last_ptr)
4937                 empty_cluster = 0;
4938
4939         if (search_start == hint_byte) {
4940 ideal_cache:
4941                 block_group = btrfs_lookup_block_group(root->fs_info,
4942                                                        search_start);
4943                 /*
4944                  * we don't want to use the block group if it doesn't match our
4945                  * allocation bits, or if its not cached.
4946                  *
4947                  * However if we are re-searching with an ideal block group
4948                  * picked out then we don't care that the block group is cached.
4949                  */
4950                 if (block_group && block_group_bits(block_group, data) &&
4951                     (block_group->cached != BTRFS_CACHE_NO ||
4952                      search_start == ideal_cache_offset)) {
4953                         down_read(&space_info->groups_sem);
4954                         if (list_empty(&block_group->list) ||
4955                             block_group->ro) {
4956                                 /*
4957                                  * someone is removing this block group,
4958                                  * we can't jump into the have_block_group
4959                                  * target because our list pointers are not
4960                                  * valid
4961                                  */
4962                                 btrfs_put_block_group(block_group);
4963                                 up_read(&space_info->groups_sem);
4964                         } else {
4965                                 index = get_block_group_index(block_group);
4966                                 goto have_block_group;
4967                         }
4968                 } else if (block_group) {
4969                         btrfs_put_block_group(block_group);
4970                 }
4971         }
4972 search:
4973         down_read(&space_info->groups_sem);
4974         list_for_each_entry(block_group, &space_info->block_groups[index],
4975                             list) {
4976                 u64 offset;
4977                 int cached;
4978
4979                 btrfs_get_block_group(block_group);
4980                 search_start = block_group->key.objectid;
4981
4982                 /*
4983                  * this can happen if we end up cycling through all the
4984                  * raid types, but we want to make sure we only allocate
4985                  * for the proper type.
4986                  */
4987                 if (!block_group_bits(block_group, data)) {
4988                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
4989                                 BTRFS_BLOCK_GROUP_RAID1 |
4990                                 BTRFS_BLOCK_GROUP_RAID10;
4991
4992                         /*
4993                          * if they asked for extra copies and this block group
4994                          * doesn't provide them, bail.  This does allow us to
4995                          * fill raid0 from raid1.
4996                          */
4997                         if ((data & extra) && !(block_group->flags & extra))
4998                                 goto loop;
4999                 }
5000
5001 have_block_group:
5002                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5003                         u64 free_percent;
5004
5005                         ret = cache_block_group(block_group, trans,
5006                                                 orig_root, 1);
5007                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5008                                 goto have_block_group;
5009
5010                         free_percent = btrfs_block_group_used(&block_group->item);
5011                         free_percent *= 100;
5012                         free_percent = div64_u64(free_percent,
5013                                                  block_group->key.offset);
5014                         free_percent = 100 - free_percent;
5015                         if (free_percent > ideal_cache_percent &&
5016                             likely(!block_group->ro)) {
5017                                 ideal_cache_offset = block_group->key.objectid;
5018                                 ideal_cache_percent = free_percent;
5019                         }
5020
5021                         /*
5022                          * The caching workers are limited to 2 threads, so we
5023                          * can queue as much work as we care to.
5024                          */
5025                         if (loop > LOOP_FIND_IDEAL) {
5026                                 ret = cache_block_group(block_group, trans,
5027                                                         orig_root, 0);
5028                                 BUG_ON(ret);
5029                         }
5030                         found_uncached_bg = true;
5031
5032                         /*
5033                          * If loop is set for cached only, try the next block
5034                          * group.
5035                          */
5036                         if (loop == LOOP_FIND_IDEAL)
5037                                 goto loop;
5038                 }
5039
5040                 cached = block_group_cache_done(block_group);
5041                 if (unlikely(!cached))
5042                         found_uncached_bg = true;
5043
5044                 if (unlikely(block_group->ro))
5045                         goto loop;
5046
5047                 spin_lock(&block_group->free_space_ctl->tree_lock);
5048                 if (cached &&
5049                     block_group->free_space_ctl->free_space <
5050                     num_bytes + empty_size) {
5051                         spin_unlock(&block_group->free_space_ctl->tree_lock);
5052                         goto loop;
5053                 }
5054                 spin_unlock(&block_group->free_space_ctl->tree_lock);
5055
5056                 /*
5057                  * Ok we want to try and use the cluster allocator, so lets look
5058                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5059                  * have tried the cluster allocator plenty of times at this
5060                  * point and not have found anything, so we are likely way too
5061                  * fragmented for the clustering stuff to find anything, so lets
5062                  * just skip it and let the allocator find whatever block it can
5063                  * find
5064                  */
5065                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5066                         /*
5067                          * the refill lock keeps out other
5068                          * people trying to start a new cluster
5069                          */
5070                         spin_lock(&last_ptr->refill_lock);
5071                         if (last_ptr->block_group &&
5072                             (last_ptr->block_group->ro ||
5073                             !block_group_bits(last_ptr->block_group, data))) {
5074                                 offset = 0;
5075                                 goto refill_cluster;
5076                         }
5077
5078                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5079                                                  num_bytes, search_start);
5080                         if (offset) {
5081                                 /* we have a block, we're done */
5082                                 spin_unlock(&last_ptr->refill_lock);
5083                                 goto checks;
5084                         }
5085
5086                         spin_lock(&last_ptr->lock);
5087                         /*
5088                          * whoops, this cluster doesn't actually point to
5089                          * this block group.  Get a ref on the block
5090                          * group is does point to and try again
5091                          */
5092                         if (!last_ptr_loop && last_ptr->block_group &&
5093                             last_ptr->block_group != block_group &&
5094                             index <=
5095                                  get_block_group_index(last_ptr->block_group)) {
5096
5097                                 btrfs_put_block_group(block_group);
5098                                 block_group = last_ptr->block_group;
5099                                 btrfs_get_block_group(block_group);
5100                                 spin_unlock(&last_ptr->lock);
5101                                 spin_unlock(&last_ptr->refill_lock);
5102
5103                                 last_ptr_loop = 1;
5104                                 search_start = block_group->key.objectid;
5105                                 /*
5106                                  * we know this block group is properly
5107                                  * in the list because
5108                                  * btrfs_remove_block_group, drops the
5109                                  * cluster before it removes the block
5110                                  * group from the list
5111                                  */
5112                                 goto have_block_group;
5113                         }
5114                         spin_unlock(&last_ptr->lock);
5115 refill_cluster:
5116                         /*
5117                          * this cluster didn't work out, free it and
5118                          * start over
5119                          */
5120                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5121
5122                         last_ptr_loop = 0;
5123
5124                         /* allocate a cluster in this block group */
5125                         ret = btrfs_find_space_cluster(trans, root,
5126                                                block_group, last_ptr,
5127                                                offset, num_bytes,
5128                                                empty_cluster + empty_size);
5129                         if (ret == 0) {
5130                                 /*
5131                                  * now pull our allocation out of this
5132                                  * cluster
5133                                  */
5134                                 offset = btrfs_alloc_from_cluster(block_group,
5135                                                   last_ptr, num_bytes,
5136                                                   search_start);
5137                                 if (offset) {
5138                                         /* we found one, proceed */
5139                                         spin_unlock(&last_ptr->refill_lock);
5140                                         goto checks;
5141                                 }
5142                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5143                                    && !failed_cluster_refill) {
5144                                 spin_unlock(&last_ptr->refill_lock);
5145
5146                                 failed_cluster_refill = true;
5147                                 wait_block_group_cache_progress(block_group,
5148                                        num_bytes + empty_cluster + empty_size);
5149                                 goto have_block_group;
5150                         }
5151
5152                         /*
5153                          * at this point we either didn't find a cluster
5154                          * or we weren't able to allocate a block from our
5155                          * cluster.  Free the cluster we've been trying
5156                          * to use, and go to the next block group
5157                          */
5158                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5159                         spin_unlock(&last_ptr->refill_lock);
5160                         goto loop;
5161                 }
5162
5163                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5164                                                     num_bytes, empty_size);
5165                 /*
5166                  * If we didn't find a chunk, and we haven't failed on this
5167                  * block group before, and this block group is in the middle of
5168                  * caching and we are ok with waiting, then go ahead and wait
5169                  * for progress to be made, and set failed_alloc to true.
5170                  *
5171                  * If failed_alloc is true then we've already waited on this
5172                  * block group once and should move on to the next block group.
5173                  */
5174                 if (!offset && !failed_alloc && !cached &&
5175                     loop > LOOP_CACHING_NOWAIT) {
5176                         wait_block_group_cache_progress(block_group,
5177                                                 num_bytes + empty_size);
5178                         failed_alloc = true;
5179                         goto have_block_group;
5180                 } else if (!offset) {
5181                         goto loop;
5182                 }
5183 checks:
5184                 search_start = stripe_align(root, offset);
5185                 /* move on to the next group */
5186                 if (search_start + num_bytes >= search_end) {
5187                         btrfs_add_free_space(block_group, offset, num_bytes);
5188                         goto loop;
5189                 }
5190
5191                 /* move on to the next group */
5192                 if (search_start + num_bytes >
5193                     block_group->key.objectid + block_group->key.offset) {
5194                         btrfs_add_free_space(block_group, offset, num_bytes);
5195                         goto loop;
5196                 }
5197
5198                 ins->objectid = search_start;
5199                 ins->offset = num_bytes;
5200
5201                 if (offset < search_start)
5202                         btrfs_add_free_space(block_group, offset,
5203                                              search_start - offset);
5204                 BUG_ON(offset > search_start);
5205
5206                 ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1,
5207                                             (data & BTRFS_BLOCK_GROUP_DATA));
5208                 if (ret == -EAGAIN) {
5209                         btrfs_add_free_space(block_group, offset, num_bytes);
5210                         goto loop;
5211                 }
5212
5213                 /* we are all good, lets return */
5214                 ins->objectid = search_start;
5215                 ins->offset = num_bytes;
5216
5217                 if (offset < search_start)
5218                         btrfs_add_free_space(block_group, offset,
5219                                              search_start - offset);
5220                 BUG_ON(offset > search_start);
5221                 btrfs_put_block_group(block_group);
5222                 break;
5223 loop:
5224                 failed_cluster_refill = false;
5225                 failed_alloc = false;
5226                 BUG_ON(index != get_block_group_index(block_group));
5227                 btrfs_put_block_group(block_group);
5228         }
5229         up_read(&space_info->groups_sem);
5230
5231         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5232                 goto search;
5233
5234         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5235          *                      for them to make caching progress.  Also
5236          *                      determine the best possible bg to cache
5237          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5238          *                      caching kthreads as we move along
5239          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5240          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5241          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5242          *                      again
5243          */
5244         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5245                 index = 0;
5246                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5247                         found_uncached_bg = false;
5248                         loop++;
5249                         if (!ideal_cache_percent)
5250                                 goto search;
5251
5252                         /*
5253                          * 1 of the following 2 things have happened so far
5254                          *
5255                          * 1) We found an ideal block group for caching that
5256                          * is mostly full and will cache quickly, so we might
5257                          * as well wait for it.
5258                          *
5259                          * 2) We searched for cached only and we didn't find
5260                          * anything, and we didn't start any caching kthreads
5261                          * either, so chances are we will loop through and
5262                          * start a couple caching kthreads, and then come back
5263                          * around and just wait for them.  This will be slower
5264                          * because we will have 2 caching kthreads reading at
5265                          * the same time when we could have just started one
5266                          * and waited for it to get far enough to give us an
5267                          * allocation, so go ahead and go to the wait caching
5268                          * loop.
5269                          */
5270                         loop = LOOP_CACHING_WAIT;
5271                         search_start = ideal_cache_offset;
5272                         ideal_cache_percent = 0;
5273                         goto ideal_cache;
5274                 } else if (loop == LOOP_FIND_IDEAL) {
5275                         /*
5276                          * Didn't find a uncached bg, wait on anything we find
5277                          * next.
5278                          */
5279                         loop = LOOP_CACHING_WAIT;
5280                         goto search;
5281                 }
5282
5283                 loop++;
5284
5285                 if (loop == LOOP_ALLOC_CHUNK) {
5286                        if (allowed_chunk_alloc) {
5287                                 ret = do_chunk_alloc(trans, root, num_bytes +
5288                                                      2 * 1024 * 1024, data,
5289                                                      CHUNK_ALLOC_LIMITED);
5290                                 allowed_chunk_alloc = 0;
5291                                 if (ret == 1)
5292                                         done_chunk_alloc = 1;
5293                         } else if (!done_chunk_alloc &&
5294                                    space_info->force_alloc ==
5295                                    CHUNK_ALLOC_NO_FORCE) {
5296                                 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5297                         }
5298
5299                        /*
5300                         * We didn't allocate a chunk, go ahead and drop the
5301                         * empty size and loop again.
5302                         */
5303                        if (!done_chunk_alloc)
5304                                loop = LOOP_NO_EMPTY_SIZE;
5305                 }
5306
5307                 if (loop == LOOP_NO_EMPTY_SIZE) {
5308                         empty_size = 0;
5309                         empty_cluster = 0;
5310                 }
5311
5312                 goto search;
5313         } else if (!ins->objectid) {
5314                 ret = -ENOSPC;
5315         } else if (ins->objectid) {
5316                 ret = 0;
5317         }
5318
5319         return ret;
5320 }
5321
5322 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5323                             int dump_block_groups)
5324 {
5325         struct btrfs_block_group_cache *cache;
5326         int index = 0;
5327
5328         spin_lock(&info->lock);
5329         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5330                (unsigned long long)(info->total_bytes - info->bytes_used -
5331                                     info->bytes_pinned - info->bytes_reserved -
5332                                     info->bytes_readonly),
5333                (info->full) ? "" : "not ");
5334         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5335                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5336                (unsigned long long)info->total_bytes,
5337                (unsigned long long)info->bytes_used,
5338                (unsigned long long)info->bytes_pinned,
5339                (unsigned long long)info->bytes_reserved,
5340                (unsigned long long)info->bytes_may_use,
5341                (unsigned long long)info->bytes_readonly);
5342         spin_unlock(&info->lock);
5343
5344         if (!dump_block_groups)
5345                 return;
5346
5347         down_read(&info->groups_sem);
5348 again:
5349         list_for_each_entry(cache, &info->block_groups[index], list) {
5350                 spin_lock(&cache->lock);
5351                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5352                        "%llu pinned %llu reserved\n",
5353                        (unsigned long long)cache->key.objectid,
5354                        (unsigned long long)cache->key.offset,
5355                        (unsigned long long)btrfs_block_group_used(&cache->item),
5356                        (unsigned long long)cache->pinned,
5357                        (unsigned long long)cache->reserved);
5358                 btrfs_dump_free_space(cache, bytes);
5359                 spin_unlock(&cache->lock);
5360         }
5361         if (++index < BTRFS_NR_RAID_TYPES)
5362                 goto again;
5363         up_read(&info->groups_sem);
5364 }
5365
5366 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5367                          struct btrfs_root *root,
5368                          u64 num_bytes, u64 min_alloc_size,
5369                          u64 empty_size, u64 hint_byte,
5370                          u64 search_end, struct btrfs_key *ins,
5371                          u64 data)
5372 {
5373         int ret;
5374         u64 search_start = 0;
5375
5376         data = btrfs_get_alloc_profile(root, data);
5377 again:
5378         /*
5379          * the only place that sets empty_size is btrfs_realloc_node, which
5380          * is not called recursively on allocations
5381          */
5382         if (empty_size || root->ref_cows)
5383                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5384                                      num_bytes + 2 * 1024 * 1024, data,
5385                                      CHUNK_ALLOC_NO_FORCE);
5386
5387         WARN_ON(num_bytes < root->sectorsize);
5388         ret = find_free_extent(trans, root, num_bytes, empty_size,
5389                                search_start, search_end, hint_byte,
5390                                ins, data);
5391
5392         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5393                 num_bytes = num_bytes >> 1;
5394                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5395                 num_bytes = max(num_bytes, min_alloc_size);
5396                 do_chunk_alloc(trans, root->fs_info->extent_root,
5397                                num_bytes, data, CHUNK_ALLOC_FORCE);
5398                 goto again;
5399         }
5400         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5401                 struct btrfs_space_info *sinfo;
5402
5403                 sinfo = __find_space_info(root->fs_info, data);
5404                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5405                        "wanted %llu\n", (unsigned long long)data,
5406                        (unsigned long long)num_bytes);
5407                 dump_space_info(sinfo, num_bytes, 1);
5408         }
5409
5410         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5411
5412         return ret;
5413 }
5414
5415 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5416 {
5417         struct btrfs_block_group_cache *cache;
5418         int ret = 0;
5419
5420         cache = btrfs_lookup_block_group(root->fs_info, start);
5421         if (!cache) {
5422                 printk(KERN_ERR "Unable to find block group for %llu\n",
5423                        (unsigned long long)start);
5424                 return -ENOSPC;
5425         }
5426
5427         if (btrfs_test_opt(root, DISCARD))
5428                 ret = btrfs_discard_extent(root, start, len, NULL);
5429
5430         btrfs_add_free_space(cache, start, len);
5431         btrfs_update_reserved_bytes(cache, len, 0, 1);
5432         btrfs_put_block_group(cache);
5433
5434         trace_btrfs_reserved_extent_free(root, start, len);
5435
5436         return ret;
5437 }
5438
5439 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5440                                       struct btrfs_root *root,
5441                                       u64 parent, u64 root_objectid,
5442                                       u64 flags, u64 owner, u64 offset,
5443                                       struct btrfs_key *ins, int ref_mod)
5444 {
5445         int ret;
5446         struct btrfs_fs_info *fs_info = root->fs_info;
5447         struct btrfs_extent_item *extent_item;
5448         struct btrfs_extent_inline_ref *iref;
5449         struct btrfs_path *path;
5450         struct extent_buffer *leaf;
5451         int type;
5452         u32 size;
5453
5454         if (parent > 0)
5455                 type = BTRFS_SHARED_DATA_REF_KEY;
5456         else
5457                 type = BTRFS_EXTENT_DATA_REF_KEY;
5458
5459         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5460
5461         path = btrfs_alloc_path();
5462         if (!path)
5463                 return -ENOMEM;
5464
5465         path->leave_spinning = 1;
5466         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5467                                       ins, size);
5468         BUG_ON(ret);
5469
5470         leaf = path->nodes[0];
5471         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5472                                      struct btrfs_extent_item);
5473         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5474         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5475         btrfs_set_extent_flags(leaf, extent_item,
5476                                flags | BTRFS_EXTENT_FLAG_DATA);
5477
5478         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5479         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5480         if (parent > 0) {
5481                 struct btrfs_shared_data_ref *ref;
5482                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5483                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5484                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5485         } else {
5486                 struct btrfs_extent_data_ref *ref;
5487                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5488                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5489                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5490                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5491                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5492         }
5493
5494         btrfs_mark_buffer_dirty(path->nodes[0]);
5495         btrfs_free_path(path);
5496
5497         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5498         if (ret) {
5499                 printk(KERN_ERR "btrfs update block group failed for %llu "
5500                        "%llu\n", (unsigned long long)ins->objectid,
5501                        (unsigned long long)ins->offset);
5502                 BUG();
5503         }
5504         return ret;
5505 }
5506
5507 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5508                                      struct btrfs_root *root,
5509                                      u64 parent, u64 root_objectid,
5510                                      u64 flags, struct btrfs_disk_key *key,
5511                                      int level, struct btrfs_key *ins)
5512 {
5513         int ret;
5514         struct btrfs_fs_info *fs_info = root->fs_info;
5515         struct btrfs_extent_item *extent_item;
5516         struct btrfs_tree_block_info *block_info;
5517         struct btrfs_extent_inline_ref *iref;
5518         struct btrfs_path *path;
5519         struct extent_buffer *leaf;
5520         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5521
5522         path = btrfs_alloc_path();
5523         if (!path)
5524                 return -ENOMEM;
5525
5526         path->leave_spinning = 1;
5527         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5528                                       ins, size);
5529         BUG_ON(ret);
5530
5531         leaf = path->nodes[0];
5532         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5533                                      struct btrfs_extent_item);
5534         btrfs_set_extent_refs(leaf, extent_item, 1);
5535         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5536         btrfs_set_extent_flags(leaf, extent_item,
5537                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5538         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5539
5540         btrfs_set_tree_block_key(leaf, block_info, key);
5541         btrfs_set_tree_block_level(leaf, block_info, level);
5542
5543         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5544         if (parent > 0) {
5545                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5546                 btrfs_set_extent_inline_ref_type(leaf, iref,
5547                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5548                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5549         } else {
5550                 btrfs_set_extent_inline_ref_type(leaf, iref,
5551                                                  BTRFS_TREE_BLOCK_REF_KEY);
5552                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5553         }
5554
5555         btrfs_mark_buffer_dirty(leaf);
5556         btrfs_free_path(path);
5557
5558         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5559         if (ret) {
5560                 printk(KERN_ERR "btrfs update block group failed for %llu "
5561                        "%llu\n", (unsigned long long)ins->objectid,
5562                        (unsigned long long)ins->offset);
5563                 BUG();
5564         }
5565         return ret;
5566 }
5567
5568 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5569                                      struct btrfs_root *root,
5570                                      u64 root_objectid, u64 owner,
5571                                      u64 offset, struct btrfs_key *ins)
5572 {
5573         int ret;
5574
5575         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5576
5577         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5578                                          0, root_objectid, owner, offset,
5579                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5580         return ret;
5581 }
5582
5583 /*
5584  * this is used by the tree logging recovery code.  It records that
5585  * an extent has been allocated and makes sure to clear the free
5586  * space cache bits as well
5587  */
5588 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5589                                    struct btrfs_root *root,
5590                                    u64 root_objectid, u64 owner, u64 offset,
5591                                    struct btrfs_key *ins)
5592 {
5593         int ret;
5594         struct btrfs_block_group_cache *block_group;
5595         struct btrfs_caching_control *caching_ctl;
5596         u64 start = ins->objectid;
5597         u64 num_bytes = ins->offset;
5598
5599         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5600         cache_block_group(block_group, trans, NULL, 0);
5601         caching_ctl = get_caching_control(block_group);
5602
5603         if (!caching_ctl) {
5604                 BUG_ON(!block_group_cache_done(block_group));
5605                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5606                 BUG_ON(ret);
5607         } else {
5608                 mutex_lock(&caching_ctl->mutex);
5609
5610                 if (start >= caching_ctl->progress) {
5611                         ret = add_excluded_extent(root, start, num_bytes);
5612                         BUG_ON(ret);
5613                 } else if (start + num_bytes <= caching_ctl->progress) {
5614                         ret = btrfs_remove_free_space(block_group,
5615                                                       start, num_bytes);
5616                         BUG_ON(ret);
5617                 } else {
5618                         num_bytes = caching_ctl->progress - start;
5619                         ret = btrfs_remove_free_space(block_group,
5620                                                       start, num_bytes);
5621                         BUG_ON(ret);
5622
5623                         start = caching_ctl->progress;
5624                         num_bytes = ins->objectid + ins->offset -
5625                                     caching_ctl->progress;
5626                         ret = add_excluded_extent(root, start, num_bytes);
5627                         BUG_ON(ret);
5628                 }
5629
5630                 mutex_unlock(&caching_ctl->mutex);
5631                 put_caching_control(caching_ctl);
5632         }
5633
5634         ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1);
5635         BUG_ON(ret);
5636         btrfs_put_block_group(block_group);
5637         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5638                                          0, owner, offset, ins, 1);
5639         return ret;
5640 }
5641
5642 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5643                                             struct btrfs_root *root,
5644                                             u64 bytenr, u32 blocksize,
5645                                             int level)
5646 {
5647         struct extent_buffer *buf;
5648
5649         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5650         if (!buf)
5651                 return ERR_PTR(-ENOMEM);
5652         btrfs_set_header_generation(buf, trans->transid);
5653         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5654         btrfs_tree_lock(buf);
5655         clean_tree_block(trans, root, buf);
5656
5657         btrfs_set_lock_blocking(buf);
5658         btrfs_set_buffer_uptodate(buf);
5659
5660         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5661                 /*
5662                  * we allow two log transactions at a time, use different
5663                  * EXENT bit to differentiate dirty pages.
5664                  */
5665                 if (root->log_transid % 2 == 0)
5666                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5667                                         buf->start + buf->len - 1, GFP_NOFS);
5668                 else
5669                         set_extent_new(&root->dirty_log_pages, buf->start,
5670                                         buf->start + buf->len - 1, GFP_NOFS);
5671         } else {
5672                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5673                          buf->start + buf->len - 1, GFP_NOFS);
5674         }
5675         trans->blocks_used++;
5676         /* this returns a buffer locked for blocking */
5677         return buf;
5678 }
5679
5680 static struct btrfs_block_rsv *
5681 use_block_rsv(struct btrfs_trans_handle *trans,
5682               struct btrfs_root *root, u32 blocksize)
5683 {
5684         struct btrfs_block_rsv *block_rsv;
5685         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5686         int ret;
5687
5688         block_rsv = get_block_rsv(trans, root);
5689
5690         if (block_rsv->size == 0) {
5691                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5692                                              blocksize, 0);
5693                 /*
5694                  * If we couldn't reserve metadata bytes try and use some from
5695                  * the global reserve.
5696                  */
5697                 if (ret && block_rsv != global_rsv) {
5698                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5699                         if (!ret)
5700                                 return global_rsv;
5701                         return ERR_PTR(ret);
5702                 } else if (ret) {
5703                         return ERR_PTR(ret);
5704                 }
5705                 return block_rsv;
5706         }
5707
5708         ret = block_rsv_use_bytes(block_rsv, blocksize);
5709         if (!ret)
5710                 return block_rsv;
5711         if (ret) {
5712                 WARN_ON(1);
5713                 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5714                                              0);
5715                 if (!ret) {
5716                         spin_lock(&block_rsv->lock);
5717                         block_rsv->size += blocksize;
5718                         spin_unlock(&block_rsv->lock);
5719                         return block_rsv;
5720                 } else if (ret && block_rsv != global_rsv) {
5721                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5722                         if (!ret)
5723                                 return global_rsv;
5724                 }
5725         }
5726
5727         return ERR_PTR(-ENOSPC);
5728 }
5729
5730 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5731 {
5732         block_rsv_add_bytes(block_rsv, blocksize, 0);
5733         block_rsv_release_bytes(block_rsv, NULL, 0);
5734 }
5735
5736 /*
5737  * finds a free extent and does all the dirty work required for allocation
5738  * returns the key for the extent through ins, and a tree buffer for
5739  * the first block of the extent through buf.
5740  *
5741  * returns the tree buffer or NULL.
5742  */
5743 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5744                                         struct btrfs_root *root, u32 blocksize,
5745                                         u64 parent, u64 root_objectid,
5746                                         struct btrfs_disk_key *key, int level,
5747                                         u64 hint, u64 empty_size)
5748 {
5749         struct btrfs_key ins;
5750         struct btrfs_block_rsv *block_rsv;
5751         struct extent_buffer *buf;
5752         u64 flags = 0;
5753         int ret;
5754
5755
5756         block_rsv = use_block_rsv(trans, root, blocksize);
5757         if (IS_ERR(block_rsv))
5758                 return ERR_CAST(block_rsv);
5759
5760         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5761                                    empty_size, hint, (u64)-1, &ins, 0);
5762         if (ret) {
5763                 unuse_block_rsv(block_rsv, blocksize);
5764                 return ERR_PTR(ret);
5765         }
5766
5767         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5768                                     blocksize, level);
5769         BUG_ON(IS_ERR(buf));
5770
5771         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5772                 if (parent == 0)
5773                         parent = ins.objectid;
5774                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5775         } else
5776                 BUG_ON(parent > 0);
5777
5778         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5779                 struct btrfs_delayed_extent_op *extent_op;
5780                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5781                 BUG_ON(!extent_op);
5782                 if (key)
5783                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5784                 else
5785                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5786                 extent_op->flags_to_set = flags;
5787                 extent_op->update_key = 1;
5788                 extent_op->update_flags = 1;
5789                 extent_op->is_data = 0;
5790
5791                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5792                                         ins.offset, parent, root_objectid,
5793                                         level, BTRFS_ADD_DELAYED_EXTENT,
5794                                         extent_op);
5795                 BUG_ON(ret);
5796         }
5797         return buf;
5798 }
5799
5800 struct walk_control {
5801         u64 refs[BTRFS_MAX_LEVEL];
5802         u64 flags[BTRFS_MAX_LEVEL];
5803         struct btrfs_key update_progress;
5804         int stage;
5805         int level;
5806         int shared_level;
5807         int update_ref;
5808         int keep_locks;
5809         int reada_slot;
5810         int reada_count;
5811 };
5812
5813 #define DROP_REFERENCE  1
5814 #define UPDATE_BACKREF  2
5815
5816 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5817                                      struct btrfs_root *root,
5818                                      struct walk_control *wc,
5819                                      struct btrfs_path *path)
5820 {
5821         u64 bytenr;
5822         u64 generation;
5823         u64 refs;
5824         u64 flags;
5825         u32 nritems;
5826         u32 blocksize;
5827         struct btrfs_key key;
5828         struct extent_buffer *eb;
5829         int ret;
5830         int slot;
5831         int nread = 0;
5832
5833         if (path->slots[wc->level] < wc->reada_slot) {
5834                 wc->reada_count = wc->reada_count * 2 / 3;
5835                 wc->reada_count = max(wc->reada_count, 2);
5836         } else {
5837                 wc->reada_count = wc->reada_count * 3 / 2;
5838                 wc->reada_count = min_t(int, wc->reada_count,
5839                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5840         }
5841
5842         eb = path->nodes[wc->level];
5843         nritems = btrfs_header_nritems(eb);
5844         blocksize = btrfs_level_size(root, wc->level - 1);
5845
5846         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5847                 if (nread >= wc->reada_count)
5848                         break;
5849
5850                 cond_resched();
5851                 bytenr = btrfs_node_blockptr(eb, slot);
5852                 generation = btrfs_node_ptr_generation(eb, slot);
5853
5854                 if (slot == path->slots[wc->level])
5855                         goto reada;
5856
5857                 if (wc->stage == UPDATE_BACKREF &&
5858                     generation <= root->root_key.offset)
5859                         continue;
5860
5861                 /* We don't lock the tree block, it's OK to be racy here */
5862                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5863                                                &refs, &flags);
5864                 BUG_ON(ret);
5865                 BUG_ON(refs == 0);
5866
5867                 if (wc->stage == DROP_REFERENCE) {
5868                         if (refs == 1)
5869                                 goto reada;
5870
5871                         if (wc->level == 1 &&
5872                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5873                                 continue;
5874                         if (!wc->update_ref ||
5875                             generation <= root->root_key.offset)
5876                                 continue;
5877                         btrfs_node_key_to_cpu(eb, &key, slot);
5878                         ret = btrfs_comp_cpu_keys(&key,
5879                                                   &wc->update_progress);
5880                         if (ret < 0)
5881                                 continue;
5882                 } else {
5883                         if (wc->level == 1 &&
5884                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5885                                 continue;
5886                 }
5887 reada:
5888                 ret = readahead_tree_block(root, bytenr, blocksize,
5889                                            generation);
5890                 if (ret)
5891                         break;
5892                 nread++;
5893         }
5894         wc->reada_slot = slot;
5895 }
5896
5897 /*
5898  * hepler to process tree block while walking down the tree.
5899  *
5900  * when wc->stage == UPDATE_BACKREF, this function updates
5901  * back refs for pointers in the block.
5902  *
5903  * NOTE: return value 1 means we should stop walking down.
5904  */
5905 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5906                                    struct btrfs_root *root,
5907                                    struct btrfs_path *path,
5908                                    struct walk_control *wc, int lookup_info)
5909 {
5910         int level = wc->level;
5911         struct extent_buffer *eb = path->nodes[level];
5912         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5913         int ret;
5914
5915         if (wc->stage == UPDATE_BACKREF &&
5916             btrfs_header_owner(eb) != root->root_key.objectid)
5917                 return 1;
5918
5919         /*
5920          * when reference count of tree block is 1, it won't increase
5921          * again. once full backref flag is set, we never clear it.
5922          */
5923         if (lookup_info &&
5924             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5925              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5926                 BUG_ON(!path->locks[level]);
5927                 ret = btrfs_lookup_extent_info(trans, root,
5928                                                eb->start, eb->len,
5929                                                &wc->refs[level],
5930                                                &wc->flags[level]);
5931                 BUG_ON(ret);
5932                 BUG_ON(wc->refs[level] == 0);
5933         }
5934
5935         if (wc->stage == DROP_REFERENCE) {
5936                 if (wc->refs[level] > 1)
5937                         return 1;
5938
5939                 if (path->locks[level] && !wc->keep_locks) {
5940                         btrfs_tree_unlock_rw(eb, path->locks[level]);
5941                         path->locks[level] = 0;
5942                 }
5943                 return 0;
5944         }
5945
5946         /* wc->stage == UPDATE_BACKREF */
5947         if (!(wc->flags[level] & flag)) {
5948                 BUG_ON(!path->locks[level]);
5949                 ret = btrfs_inc_ref(trans, root, eb, 1);
5950                 BUG_ON(ret);
5951                 ret = btrfs_dec_ref(trans, root, eb, 0);
5952                 BUG_ON(ret);
5953                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5954                                                   eb->len, flag, 0);
5955                 BUG_ON(ret);
5956                 wc->flags[level] |= flag;
5957         }
5958
5959         /*
5960          * the block is shared by multiple trees, so it's not good to
5961          * keep the tree lock
5962          */
5963         if (path->locks[level] && level > 0) {
5964                 btrfs_tree_unlock_rw(eb, path->locks[level]);
5965                 path->locks[level] = 0;
5966         }
5967         return 0;
5968 }
5969
5970 /*
5971  * hepler to process tree block pointer.
5972  *
5973  * when wc->stage == DROP_REFERENCE, this function checks
5974  * reference count of the block pointed to. if the block
5975  * is shared and we need update back refs for the subtree
5976  * rooted at the block, this function changes wc->stage to
5977  * UPDATE_BACKREF. if the block is shared and there is no
5978  * need to update back, this function drops the reference
5979  * to the block.
5980  *
5981  * NOTE: return value 1 means we should stop walking down.
5982  */
5983 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5984                                  struct btrfs_root *root,
5985                                  struct btrfs_path *path,
5986                                  struct walk_control *wc, int *lookup_info)
5987 {
5988         u64 bytenr;
5989         u64 generation;
5990         u64 parent;
5991         u32 blocksize;
5992         struct btrfs_key key;
5993         struct extent_buffer *next;
5994         int level = wc->level;
5995         int reada = 0;
5996         int ret = 0;
5997
5998         generation = btrfs_node_ptr_generation(path->nodes[level],
5999                                                path->slots[level]);
6000         /*
6001          * if the lower level block was created before the snapshot
6002          * was created, we know there is no need to update back refs
6003          * for the subtree
6004          */
6005         if (wc->stage == UPDATE_BACKREF &&
6006             generation <= root->root_key.offset) {
6007                 *lookup_info = 1;
6008                 return 1;
6009         }
6010
6011         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6012         blocksize = btrfs_level_size(root, level - 1);
6013
6014         next = btrfs_find_tree_block(root, bytenr, blocksize);
6015         if (!next) {
6016                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6017                 if (!next)
6018                         return -ENOMEM;
6019                 reada = 1;
6020         }
6021         btrfs_tree_lock(next);
6022         btrfs_set_lock_blocking(next);
6023
6024         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6025                                        &wc->refs[level - 1],
6026                                        &wc->flags[level - 1]);
6027         BUG_ON(ret);
6028         BUG_ON(wc->refs[level - 1] == 0);
6029         *lookup_info = 0;
6030
6031         if (wc->stage == DROP_REFERENCE) {
6032                 if (wc->refs[level - 1] > 1) {
6033                         if (level == 1 &&
6034                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6035                                 goto skip;
6036
6037                         if (!wc->update_ref ||
6038                             generation <= root->root_key.offset)
6039                                 goto skip;
6040
6041                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6042                                               path->slots[level]);
6043                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6044                         if (ret < 0)
6045                                 goto skip;
6046
6047                         wc->stage = UPDATE_BACKREF;
6048                         wc->shared_level = level - 1;
6049                 }
6050         } else {
6051                 if (level == 1 &&
6052                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6053                         goto skip;
6054         }
6055
6056         if (!btrfs_buffer_uptodate(next, generation)) {
6057                 btrfs_tree_unlock(next);
6058                 free_extent_buffer(next);
6059                 next = NULL;
6060                 *lookup_info = 1;
6061         }
6062
6063         if (!next) {
6064                 if (reada && level == 1)
6065                         reada_walk_down(trans, root, wc, path);
6066                 next = read_tree_block(root, bytenr, blocksize, generation);
6067                 if (!next)
6068                         return -EIO;
6069                 btrfs_tree_lock(next);
6070                 btrfs_set_lock_blocking(next);
6071         }
6072
6073         level--;
6074         BUG_ON(level != btrfs_header_level(next));
6075         path->nodes[level] = next;
6076         path->slots[level] = 0;
6077         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6078         wc->level = level;
6079         if (wc->level == 1)
6080                 wc->reada_slot = 0;
6081         return 0;
6082 skip:
6083         wc->refs[level - 1] = 0;
6084         wc->flags[level - 1] = 0;
6085         if (wc->stage == DROP_REFERENCE) {
6086                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6087                         parent = path->nodes[level]->start;
6088                 } else {
6089                         BUG_ON(root->root_key.objectid !=
6090                                btrfs_header_owner(path->nodes[level]));
6091                         parent = 0;
6092                 }
6093
6094                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6095                                         root->root_key.objectid, level - 1, 0);
6096                 BUG_ON(ret);
6097         }
6098         btrfs_tree_unlock(next);
6099         free_extent_buffer(next);
6100         *lookup_info = 1;
6101         return 1;
6102 }
6103
6104 /*
6105  * hepler to process tree block while walking up the tree.
6106  *
6107  * when wc->stage == DROP_REFERENCE, this function drops
6108  * reference count on the block.
6109  *
6110  * when wc->stage == UPDATE_BACKREF, this function changes
6111  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6112  * to UPDATE_BACKREF previously while processing the block.
6113  *
6114  * NOTE: return value 1 means we should stop walking up.
6115  */
6116 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6117                                  struct btrfs_root *root,
6118                                  struct btrfs_path *path,
6119                                  struct walk_control *wc)
6120 {
6121         int ret;
6122         int level = wc->level;
6123         struct extent_buffer *eb = path->nodes[level];
6124         u64 parent = 0;
6125
6126         if (wc->stage == UPDATE_BACKREF) {
6127                 BUG_ON(wc->shared_level < level);
6128                 if (level < wc->shared_level)
6129                         goto out;
6130
6131                 ret = find_next_key(path, level + 1, &wc->update_progress);
6132                 if (ret > 0)
6133                         wc->update_ref = 0;
6134
6135                 wc->stage = DROP_REFERENCE;
6136                 wc->shared_level = -1;
6137                 path->slots[level] = 0;
6138
6139                 /*
6140                  * check reference count again if the block isn't locked.
6141                  * we should start walking down the tree again if reference
6142                  * count is one.
6143                  */
6144                 if (!path->locks[level]) {
6145                         BUG_ON(level == 0);
6146                         btrfs_tree_lock(eb);
6147                         btrfs_set_lock_blocking(eb);
6148                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6149
6150                         ret = btrfs_lookup_extent_info(trans, root,
6151                                                        eb->start, eb->len,
6152                                                        &wc->refs[level],
6153                                                        &wc->flags[level]);
6154                         BUG_ON(ret);
6155                         BUG_ON(wc->refs[level] == 0);
6156                         if (wc->refs[level] == 1) {
6157                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6158                                 return 1;
6159                         }
6160                 }
6161         }
6162
6163         /* wc->stage == DROP_REFERENCE */
6164         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6165
6166         if (wc->refs[level] == 1) {
6167                 if (level == 0) {
6168                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6169                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6170                         else
6171                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6172                         BUG_ON(ret);
6173                 }
6174                 /* make block locked assertion in clean_tree_block happy */
6175                 if (!path->locks[level] &&
6176                     btrfs_header_generation(eb) == trans->transid) {
6177                         btrfs_tree_lock(eb);
6178                         btrfs_set_lock_blocking(eb);
6179                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6180                 }
6181                 clean_tree_block(trans, root, eb);
6182         }
6183
6184         if (eb == root->node) {
6185                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6186                         parent = eb->start;
6187                 else
6188                         BUG_ON(root->root_key.objectid !=
6189                                btrfs_header_owner(eb));
6190         } else {
6191                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6192                         parent = path->nodes[level + 1]->start;
6193                 else
6194                         BUG_ON(root->root_key.objectid !=
6195                                btrfs_header_owner(path->nodes[level + 1]));
6196         }
6197
6198         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6199 out:
6200         wc->refs[level] = 0;
6201         wc->flags[level] = 0;
6202         return 0;
6203 }
6204
6205 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6206                                    struct btrfs_root *root,
6207                                    struct btrfs_path *path,
6208                                    struct walk_control *wc)
6209 {
6210         int level = wc->level;
6211         int lookup_info = 1;
6212         int ret;
6213
6214         while (level >= 0) {
6215                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6216                 if (ret > 0)
6217                         break;
6218
6219                 if (level == 0)
6220                         break;
6221
6222                 if (path->slots[level] >=
6223                     btrfs_header_nritems(path->nodes[level]))
6224                         break;
6225
6226                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6227                 if (ret > 0) {
6228                         path->slots[level]++;
6229                         continue;
6230                 } else if (ret < 0)
6231                         return ret;
6232                 level = wc->level;
6233         }
6234         return 0;
6235 }
6236
6237 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6238                                  struct btrfs_root *root,
6239                                  struct btrfs_path *path,
6240                                  struct walk_control *wc, int max_level)
6241 {
6242         int level = wc->level;
6243         int ret;
6244
6245         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6246         while (level < max_level && path->nodes[level]) {
6247                 wc->level = level;
6248                 if (path->slots[level] + 1 <
6249                     btrfs_header_nritems(path->nodes[level])) {
6250                         path->slots[level]++;
6251                         return 0;
6252                 } else {
6253                         ret = walk_up_proc(trans, root, path, wc);
6254                         if (ret > 0)
6255                                 return 0;
6256
6257                         if (path->locks[level]) {
6258                                 btrfs_tree_unlock_rw(path->nodes[level],
6259                                                      path->locks[level]);
6260                                 path->locks[level] = 0;
6261                         }
6262                         free_extent_buffer(path->nodes[level]);
6263                         path->nodes[level] = NULL;
6264                         level++;
6265                 }
6266         }
6267         return 1;
6268 }
6269
6270 /*
6271  * drop a subvolume tree.
6272  *
6273  * this function traverses the tree freeing any blocks that only
6274  * referenced by the tree.
6275  *
6276  * when a shared tree block is found. this function decreases its
6277  * reference count by one. if update_ref is true, this function
6278  * also make sure backrefs for the shared block and all lower level
6279  * blocks are properly updated.
6280  */
6281 void btrfs_drop_snapshot(struct btrfs_root *root,
6282                          struct btrfs_block_rsv *block_rsv, int update_ref)
6283 {
6284         struct btrfs_path *path;
6285         struct btrfs_trans_handle *trans;
6286         struct btrfs_root *tree_root = root->fs_info->tree_root;
6287         struct btrfs_root_item *root_item = &root->root_item;
6288         struct walk_control *wc;
6289         struct btrfs_key key;
6290         int err = 0;
6291         int ret;
6292         int level;
6293
6294         path = btrfs_alloc_path();
6295         if (!path) {
6296                 err = -ENOMEM;
6297                 goto out;
6298         }
6299
6300         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6301         if (!wc) {
6302                 btrfs_free_path(path);
6303                 err = -ENOMEM;
6304                 goto out;
6305         }
6306
6307         trans = btrfs_start_transaction(tree_root, 0);
6308         BUG_ON(IS_ERR(trans));
6309
6310         if (block_rsv)
6311                 trans->block_rsv = block_rsv;
6312
6313         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6314                 level = btrfs_header_level(root->node);
6315                 path->nodes[level] = btrfs_lock_root_node(root);
6316                 btrfs_set_lock_blocking(path->nodes[level]);
6317                 path->slots[level] = 0;
6318                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6319                 memset(&wc->update_progress, 0,
6320                        sizeof(wc->update_progress));
6321         } else {
6322                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6323                 memcpy(&wc->update_progress, &key,
6324                        sizeof(wc->update_progress));
6325
6326                 level = root_item->drop_level;
6327                 BUG_ON(level == 0);
6328                 path->lowest_level = level;
6329                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6330                 path->lowest_level = 0;
6331                 if (ret < 0) {
6332                         err = ret;
6333                         goto out_free;
6334                 }
6335                 WARN_ON(ret > 0);
6336
6337                 /*
6338                  * unlock our path, this is safe because only this
6339                  * function is allowed to delete this snapshot
6340                  */
6341                 btrfs_unlock_up_safe(path, 0);
6342
6343                 level = btrfs_header_level(root->node);
6344                 while (1) {
6345                         btrfs_tree_lock(path->nodes[level]);
6346                         btrfs_set_lock_blocking(path->nodes[level]);
6347
6348                         ret = btrfs_lookup_extent_info(trans, root,
6349                                                 path->nodes[level]->start,
6350                                                 path->nodes[level]->len,
6351                                                 &wc->refs[level],
6352                                                 &wc->flags[level]);
6353                         BUG_ON(ret);
6354                         BUG_ON(wc->refs[level] == 0);
6355
6356                         if (level == root_item->drop_level)
6357                                 break;
6358
6359                         btrfs_tree_unlock(path->nodes[level]);
6360                         WARN_ON(wc->refs[level] != 1);
6361                         level--;
6362                 }
6363         }
6364
6365         wc->level = level;
6366         wc->shared_level = -1;
6367         wc->stage = DROP_REFERENCE;
6368         wc->update_ref = update_ref;
6369         wc->keep_locks = 0;
6370         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6371
6372         while (1) {
6373                 ret = walk_down_tree(trans, root, path, wc);
6374                 if (ret < 0) {
6375                         err = ret;
6376                         break;
6377                 }
6378
6379                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6380                 if (ret < 0) {
6381                         err = ret;
6382                         break;
6383                 }
6384
6385                 if (ret > 0) {
6386                         BUG_ON(wc->stage != DROP_REFERENCE);
6387                         break;
6388                 }
6389
6390                 if (wc->stage == DROP_REFERENCE) {
6391                         level = wc->level;
6392                         btrfs_node_key(path->nodes[level],
6393                                        &root_item->drop_progress,
6394                                        path->slots[level]);
6395                         root_item->drop_level = level;
6396                 }
6397
6398                 BUG_ON(wc->level == 0);
6399                 if (btrfs_should_end_transaction(trans, tree_root)) {
6400                         ret = btrfs_update_root(trans, tree_root,
6401                                                 &root->root_key,
6402                                                 root_item);
6403                         BUG_ON(ret);
6404
6405                         btrfs_end_transaction_throttle(trans, tree_root);
6406                         trans = btrfs_start_transaction(tree_root, 0);
6407                         BUG_ON(IS_ERR(trans));
6408                         if (block_rsv)
6409                                 trans->block_rsv = block_rsv;
6410                 }
6411         }
6412         btrfs_release_path(path);
6413         BUG_ON(err);
6414
6415         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6416         BUG_ON(ret);
6417
6418         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6419                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6420                                            NULL, NULL);
6421                 BUG_ON(ret < 0);
6422                 if (ret > 0) {
6423                         /* if we fail to delete the orphan item this time
6424                          * around, it'll get picked up the next time.
6425                          *
6426                          * The most common failure here is just -ENOENT.
6427                          */
6428                         btrfs_del_orphan_item(trans, tree_root,
6429                                               root->root_key.objectid);
6430                 }
6431         }
6432
6433         if (root->in_radix) {
6434                 btrfs_free_fs_root(tree_root->fs_info, root);
6435         } else {
6436                 free_extent_buffer(root->node);
6437                 free_extent_buffer(root->commit_root);
6438                 kfree(root);
6439         }
6440 out_free:
6441         btrfs_end_transaction_throttle(trans, tree_root);
6442         kfree(wc);
6443         btrfs_free_path(path);
6444 out:
6445         if (err)
6446                 btrfs_std_error(root->fs_info, err);
6447         return;
6448 }
6449
6450 /*
6451  * drop subtree rooted at tree block 'node'.
6452  *
6453  * NOTE: this function will unlock and release tree block 'node'
6454  */
6455 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6456                         struct btrfs_root *root,
6457                         struct extent_buffer *node,
6458                         struct extent_buffer *parent)
6459 {
6460         struct btrfs_path *path;
6461         struct walk_control *wc;
6462         int level;
6463         int parent_level;
6464         int ret = 0;
6465         int wret;
6466
6467         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6468
6469         path = btrfs_alloc_path();
6470         if (!path)
6471                 return -ENOMEM;
6472
6473         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6474         if (!wc) {
6475                 btrfs_free_path(path);
6476                 return -ENOMEM;
6477         }
6478
6479         btrfs_assert_tree_locked(parent);
6480         parent_level = btrfs_header_level(parent);
6481         extent_buffer_get(parent);
6482         path->nodes[parent_level] = parent;
6483         path->slots[parent_level] = btrfs_header_nritems(parent);
6484
6485         btrfs_assert_tree_locked(node);
6486         level = btrfs_header_level(node);
6487         path->nodes[level] = node;
6488         path->slots[level] = 0;
6489         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6490
6491         wc->refs[parent_level] = 1;
6492         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6493         wc->level = level;
6494         wc->shared_level = -1;
6495         wc->stage = DROP_REFERENCE;
6496         wc->update_ref = 0;
6497         wc->keep_locks = 1;
6498         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6499
6500         while (1) {
6501                 wret = walk_down_tree(trans, root, path, wc);
6502                 if (wret < 0) {
6503                         ret = wret;
6504                         break;
6505                 }
6506
6507                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6508                 if (wret < 0)
6509                         ret = wret;
6510                 if (wret != 0)
6511                         break;
6512         }
6513
6514         kfree(wc);
6515         btrfs_free_path(path);
6516         return ret;
6517 }
6518
6519 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6520 {
6521         u64 num_devices;
6522         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6523                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6524
6525         /*
6526          * we add in the count of missing devices because we want
6527          * to make sure that any RAID levels on a degraded FS
6528          * continue to be honored.
6529          */
6530         num_devices = root->fs_info->fs_devices->rw_devices +
6531                 root->fs_info->fs_devices->missing_devices;
6532
6533         if (num_devices == 1) {
6534                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6535                 stripped = flags & ~stripped;
6536
6537                 /* turn raid0 into single device chunks */
6538                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6539                         return stripped;
6540
6541                 /* turn mirroring into duplication */
6542                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6543                              BTRFS_BLOCK_GROUP_RAID10))
6544                         return stripped | BTRFS_BLOCK_GROUP_DUP;
6545                 return flags;
6546         } else {
6547                 /* they already had raid on here, just return */
6548                 if (flags & stripped)
6549                         return flags;
6550
6551                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6552                 stripped = flags & ~stripped;
6553
6554                 /* switch duplicated blocks with raid1 */
6555                 if (flags & BTRFS_BLOCK_GROUP_DUP)
6556                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
6557
6558                 /* turn single device chunks into raid0 */
6559                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6560         }
6561         return flags;
6562 }
6563
6564 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6565 {
6566         struct btrfs_space_info *sinfo = cache->space_info;
6567         u64 num_bytes;
6568         u64 min_allocable_bytes;
6569         int ret = -ENOSPC;
6570
6571
6572         /*
6573          * We need some metadata space and system metadata space for
6574          * allocating chunks in some corner cases until we force to set
6575          * it to be readonly.
6576          */
6577         if ((sinfo->flags &
6578              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6579             !force)
6580                 min_allocable_bytes = 1 * 1024 * 1024;
6581         else
6582                 min_allocable_bytes = 0;
6583
6584         spin_lock(&sinfo->lock);
6585         spin_lock(&cache->lock);
6586
6587         if (cache->ro) {
6588                 ret = 0;
6589                 goto out;
6590         }
6591
6592         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6593                     cache->bytes_super - btrfs_block_group_used(&cache->item);
6594
6595         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6596             sinfo->bytes_may_use + sinfo->bytes_readonly +
6597             cache->reserved_pinned + num_bytes + min_allocable_bytes <=
6598             sinfo->total_bytes) {
6599                 sinfo->bytes_readonly += num_bytes;
6600                 sinfo->bytes_reserved += cache->reserved_pinned;
6601                 cache->reserved_pinned = 0;
6602                 cache->ro = 1;
6603                 ret = 0;
6604         }
6605 out:
6606         spin_unlock(&cache->lock);
6607         spin_unlock(&sinfo->lock);
6608         return ret;
6609 }
6610
6611 int btrfs_set_block_group_ro(struct btrfs_root *root,
6612                              struct btrfs_block_group_cache *cache)
6613
6614 {
6615         struct btrfs_trans_handle *trans;
6616         u64 alloc_flags;
6617         int ret;
6618
6619         BUG_ON(cache->ro);
6620
6621         trans = btrfs_join_transaction(root);
6622         BUG_ON(IS_ERR(trans));
6623
6624         alloc_flags = update_block_group_flags(root, cache->flags);
6625         if (alloc_flags != cache->flags)
6626                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6627                                CHUNK_ALLOC_FORCE);
6628
6629         ret = set_block_group_ro(cache, 0);
6630         if (!ret)
6631                 goto out;
6632         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6633         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6634                              CHUNK_ALLOC_FORCE);
6635         if (ret < 0)
6636                 goto out;
6637         ret = set_block_group_ro(cache, 0);
6638 out:
6639         btrfs_end_transaction(trans, root);
6640         return ret;
6641 }
6642
6643 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6644                             struct btrfs_root *root, u64 type)
6645 {
6646         u64 alloc_flags = get_alloc_profile(root, type);
6647         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6648                               CHUNK_ALLOC_FORCE);
6649 }
6650
6651 /*
6652  * helper to account the unused space of all the readonly block group in the
6653  * list. takes mirrors into account.
6654  */
6655 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6656 {
6657         struct btrfs_block_group_cache *block_group;
6658         u64 free_bytes = 0;
6659         int factor;
6660
6661         list_for_each_entry(block_group, groups_list, list) {
6662                 spin_lock(&block_group->lock);
6663
6664                 if (!block_group->ro) {
6665                         spin_unlock(&block_group->lock);
6666                         continue;
6667                 }
6668
6669                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6670                                           BTRFS_BLOCK_GROUP_RAID10 |
6671                                           BTRFS_BLOCK_GROUP_DUP))
6672                         factor = 2;
6673                 else
6674                         factor = 1;
6675
6676                 free_bytes += (block_group->key.offset -
6677                                btrfs_block_group_used(&block_group->item)) *
6678                                factor;
6679
6680                 spin_unlock(&block_group->lock);
6681         }
6682
6683         return free_bytes;
6684 }
6685
6686 /*
6687  * helper to account the unused space of all the readonly block group in the
6688  * space_info. takes mirrors into account.
6689  */
6690 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6691 {
6692         int i;
6693         u64 free_bytes = 0;
6694
6695         spin_lock(&sinfo->lock);
6696
6697         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6698                 if (!list_empty(&sinfo->block_groups[i]))
6699                         free_bytes += __btrfs_get_ro_block_group_free_space(
6700                                                 &sinfo->block_groups[i]);
6701
6702         spin_unlock(&sinfo->lock);
6703
6704         return free_bytes;
6705 }
6706
6707 int btrfs_set_block_group_rw(struct btrfs_root *root,
6708                               struct btrfs_block_group_cache *cache)
6709 {
6710         struct btrfs_space_info *sinfo = cache->space_info;
6711         u64 num_bytes;
6712
6713         BUG_ON(!cache->ro);
6714
6715         spin_lock(&sinfo->lock);
6716         spin_lock(&cache->lock);
6717         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6718                     cache->bytes_super - btrfs_block_group_used(&cache->item);
6719         sinfo->bytes_readonly -= num_bytes;
6720         cache->ro = 0;
6721         spin_unlock(&cache->lock);
6722         spin_unlock(&sinfo->lock);
6723         return 0;
6724 }
6725
6726 /*
6727  * checks to see if its even possible to relocate this block group.
6728  *
6729  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6730  * ok to go ahead and try.
6731  */
6732 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6733 {
6734         struct btrfs_block_group_cache *block_group;
6735         struct btrfs_space_info *space_info;
6736         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6737         struct btrfs_device *device;
6738         u64 min_free;
6739         u64 dev_min = 1;
6740         u64 dev_nr = 0;
6741         int index;
6742         int full = 0;
6743         int ret = 0;
6744
6745         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6746
6747         /* odd, couldn't find the block group, leave it alone */
6748         if (!block_group)
6749                 return -1;
6750
6751         min_free = btrfs_block_group_used(&block_group->item);
6752
6753         /* no bytes used, we're good */
6754         if (!min_free)
6755                 goto out;
6756
6757         space_info = block_group->space_info;
6758         spin_lock(&space_info->lock);
6759
6760         full = space_info->full;
6761
6762         /*
6763          * if this is the last block group we have in this space, we can't
6764          * relocate it unless we're able to allocate a new chunk below.
6765          *
6766          * Otherwise, we need to make sure we have room in the space to handle
6767          * all of the extents from this block group.  If we can, we're good
6768          */
6769         if ((space_info->total_bytes != block_group->key.offset) &&
6770             (space_info->bytes_used + space_info->bytes_reserved +
6771              space_info->bytes_pinned + space_info->bytes_readonly +
6772              min_free < space_info->total_bytes)) {
6773                 spin_unlock(&space_info->lock);
6774                 goto out;
6775         }
6776         spin_unlock(&space_info->lock);
6777
6778         /*
6779          * ok we don't have enough space, but maybe we have free space on our
6780          * devices to allocate new chunks for relocation, so loop through our
6781          * alloc devices and guess if we have enough space.  However, if we
6782          * were marked as full, then we know there aren't enough chunks, and we
6783          * can just return.
6784          */
6785         ret = -1;
6786         if (full)
6787                 goto out;
6788
6789         /*
6790          * index:
6791          *      0: raid10
6792          *      1: raid1
6793          *      2: dup
6794          *      3: raid0
6795          *      4: single
6796          */
6797         index = get_block_group_index(block_group);
6798         if (index == 0) {
6799                 dev_min = 4;
6800                 /* Divide by 2 */
6801                 min_free >>= 1;
6802         } else if (index == 1) {
6803                 dev_min = 2;
6804         } else if (index == 2) {
6805                 /* Multiply by 2 */
6806                 min_free <<= 1;
6807         } else if (index == 3) {
6808                 dev_min = fs_devices->rw_devices;
6809                 do_div(min_free, dev_min);
6810         }
6811
6812         mutex_lock(&root->fs_info->chunk_mutex);
6813         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
6814                 u64 dev_offset;
6815
6816                 /*
6817                  * check to make sure we can actually find a chunk with enough
6818                  * space to fit our block group in.
6819                  */
6820                 if (device->total_bytes > device->bytes_used + min_free) {
6821                         ret = find_free_dev_extent(NULL, device, min_free,
6822                                                    &dev_offset, NULL);
6823                         if (!ret)
6824                                 dev_nr++;
6825
6826                         if (dev_nr >= dev_min)
6827                                 break;
6828
6829                         ret = -1;
6830                 }
6831         }
6832         mutex_unlock(&root->fs_info->chunk_mutex);
6833 out:
6834         btrfs_put_block_group(block_group);
6835         return ret;
6836 }
6837
6838 static int find_first_block_group(struct btrfs_root *root,
6839                 struct btrfs_path *path, struct btrfs_key *key)
6840 {
6841         int ret = 0;
6842         struct btrfs_key found_key;
6843         struct extent_buffer *leaf;
6844         int slot;
6845
6846         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6847         if (ret < 0)
6848                 goto out;
6849
6850         while (1) {
6851                 slot = path->slots[0];
6852                 leaf = path->nodes[0];
6853                 if (slot >= btrfs_header_nritems(leaf)) {
6854                         ret = btrfs_next_leaf(root, path);
6855                         if (ret == 0)
6856                                 continue;
6857                         if (ret < 0)
6858                                 goto out;
6859                         break;
6860                 }
6861                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6862
6863                 if (found_key.objectid >= key->objectid &&
6864                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6865                         ret = 0;
6866                         goto out;
6867                 }
6868                 path->slots[0]++;
6869         }
6870 out:
6871         return ret;
6872 }
6873
6874 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
6875 {
6876         struct btrfs_block_group_cache *block_group;
6877         u64 last = 0;
6878
6879         while (1) {
6880                 struct inode *inode;
6881
6882                 block_group = btrfs_lookup_first_block_group(info, last);
6883                 while (block_group) {
6884                         spin_lock(&block_group->lock);
6885                         if (block_group->iref)
6886                                 break;
6887                         spin_unlock(&block_group->lock);
6888                         block_group = next_block_group(info->tree_root,
6889                                                        block_group);
6890                 }
6891                 if (!block_group) {
6892                         if (last == 0)
6893                                 break;
6894                         last = 0;
6895                         continue;
6896                 }
6897
6898                 inode = block_group->inode;
6899                 block_group->iref = 0;
6900                 block_group->inode = NULL;
6901                 spin_unlock(&block_group->lock);
6902                 iput(inode);
6903                 last = block_group->key.objectid + block_group->key.offset;
6904                 btrfs_put_block_group(block_group);
6905         }
6906 }
6907
6908 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6909 {
6910         struct btrfs_block_group_cache *block_group;
6911         struct btrfs_space_info *space_info;
6912         struct btrfs_caching_control *caching_ctl;
6913         struct rb_node *n;
6914
6915         down_write(&info->extent_commit_sem);
6916         while (!list_empty(&info->caching_block_groups)) {
6917                 caching_ctl = list_entry(info->caching_block_groups.next,
6918                                          struct btrfs_caching_control, list);
6919                 list_del(&caching_ctl->list);
6920                 put_caching_control(caching_ctl);
6921         }
6922         up_write(&info->extent_commit_sem);
6923
6924         spin_lock(&info->block_group_cache_lock);
6925         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6926                 block_group = rb_entry(n, struct btrfs_block_group_cache,
6927                                        cache_node);
6928                 rb_erase(&block_group->cache_node,
6929                          &info->block_group_cache_tree);
6930                 spin_unlock(&info->block_group_cache_lock);
6931
6932                 down_write(&block_group->space_info->groups_sem);
6933                 list_del(&block_group->list);
6934                 up_write(&block_group->space_info->groups_sem);
6935
6936                 if (block_group->cached == BTRFS_CACHE_STARTED)
6937                         wait_block_group_cache_done(block_group);
6938
6939                 /*
6940                  * We haven't cached this block group, which means we could
6941                  * possibly have excluded extents on this block group.
6942                  */
6943                 if (block_group->cached == BTRFS_CACHE_NO)
6944                         free_excluded_extents(info->extent_root, block_group);
6945
6946                 btrfs_remove_free_space_cache(block_group);
6947                 btrfs_put_block_group(block_group);
6948
6949                 spin_lock(&info->block_group_cache_lock);
6950         }
6951         spin_unlock(&info->block_group_cache_lock);
6952
6953         /* now that all the block groups are freed, go through and
6954          * free all the space_info structs.  This is only called during
6955          * the final stages of unmount, and so we know nobody is
6956          * using them.  We call synchronize_rcu() once before we start,
6957          * just to be on the safe side.
6958          */
6959         synchronize_rcu();
6960
6961         release_global_block_rsv(info);
6962
6963         while(!list_empty(&info->space_info)) {
6964                 space_info = list_entry(info->space_info.next,
6965                                         struct btrfs_space_info,
6966                                         list);
6967                 if (space_info->bytes_pinned > 0 ||
6968                     space_info->bytes_reserved > 0) {
6969                         WARN_ON(1);
6970                         dump_space_info(space_info, 0, 0);
6971                 }
6972                 list_del(&space_info->list);
6973                 kfree(space_info);
6974         }
6975         return 0;
6976 }
6977
6978 static void __link_block_group(struct btrfs_space_info *space_info,
6979                                struct btrfs_block_group_cache *cache)
6980 {
6981         int index = get_block_group_index(cache);
6982
6983         down_write(&space_info->groups_sem);
6984         list_add_tail(&cache->list, &space_info->block_groups[index]);
6985         up_write(&space_info->groups_sem);
6986 }
6987
6988 int btrfs_read_block_groups(struct btrfs_root *root)
6989 {
6990         struct btrfs_path *path;
6991         int ret;
6992         struct btrfs_block_group_cache *cache;
6993         struct btrfs_fs_info *info = root->fs_info;
6994         struct btrfs_space_info *space_info;
6995         struct btrfs_key key;
6996         struct btrfs_key found_key;
6997         struct extent_buffer *leaf;
6998         int need_clear = 0;
6999         u64 cache_gen;
7000
7001         root = info->extent_root;
7002         key.objectid = 0;
7003         key.offset = 0;
7004         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7005         path = btrfs_alloc_path();
7006         if (!path)
7007                 return -ENOMEM;
7008         path->reada = 1;
7009
7010         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
7011         if (cache_gen != 0 &&
7012             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
7013                 need_clear = 1;
7014         if (btrfs_test_opt(root, CLEAR_CACHE))
7015                 need_clear = 1;
7016         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
7017                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
7018
7019         while (1) {
7020                 ret = find_first_block_group(root, path, &key);
7021                 if (ret > 0)
7022                         break;
7023                 if (ret != 0)
7024                         goto error;
7025                 leaf = path->nodes[0];
7026                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7027                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7028                 if (!cache) {
7029                         ret = -ENOMEM;
7030                         goto error;
7031                 }
7032                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7033                                                 GFP_NOFS);
7034                 if (!cache->free_space_ctl) {
7035                         kfree(cache);
7036                         ret = -ENOMEM;
7037                         goto error;
7038                 }
7039
7040                 atomic_set(&cache->count, 1);
7041                 spin_lock_init(&cache->lock);
7042                 cache->fs_info = info;
7043                 INIT_LIST_HEAD(&cache->list);
7044                 INIT_LIST_HEAD(&cache->cluster_list);
7045
7046                 if (need_clear)
7047                         cache->disk_cache_state = BTRFS_DC_CLEAR;
7048
7049                 read_extent_buffer(leaf, &cache->item,
7050                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7051                                    sizeof(cache->item));
7052                 memcpy(&cache->key, &found_key, sizeof(found_key));
7053
7054                 key.objectid = found_key.objectid + found_key.offset;
7055                 btrfs_release_path(path);
7056                 cache->flags = btrfs_block_group_flags(&cache->item);
7057                 cache->sectorsize = root->sectorsize;
7058
7059                 btrfs_init_free_space_ctl(cache);
7060
7061                 /*
7062                  * We need to exclude the super stripes now so that the space
7063                  * info has super bytes accounted for, otherwise we'll think
7064                  * we have more space than we actually do.
7065                  */
7066                 exclude_super_stripes(root, cache);
7067
7068                 /*
7069                  * check for two cases, either we are full, and therefore
7070                  * don't need to bother with the caching work since we won't
7071                  * find any space, or we are empty, and we can just add all
7072                  * the space in and be done with it.  This saves us _alot_ of
7073                  * time, particularly in the full case.
7074                  */
7075                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7076                         cache->last_byte_to_unpin = (u64)-1;
7077                         cache->cached = BTRFS_CACHE_FINISHED;
7078                         free_excluded_extents(root, cache);
7079                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7080                         cache->last_byte_to_unpin = (u64)-1;
7081                         cache->cached = BTRFS_CACHE_FINISHED;
7082                         add_new_free_space(cache, root->fs_info,
7083                                            found_key.objectid,
7084                                            found_key.objectid +
7085                                            found_key.offset);
7086                         free_excluded_extents(root, cache);
7087                 }
7088
7089                 ret = update_space_info(info, cache->flags, found_key.offset,
7090                                         btrfs_block_group_used(&cache->item),
7091                                         &space_info);
7092                 BUG_ON(ret);
7093                 cache->space_info = space_info;
7094                 spin_lock(&cache->space_info->lock);
7095                 cache->space_info->bytes_readonly += cache->bytes_super;
7096                 spin_unlock(&cache->space_info->lock);
7097
7098                 __link_block_group(space_info, cache);
7099
7100                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7101                 BUG_ON(ret);
7102
7103                 set_avail_alloc_bits(root->fs_info, cache->flags);
7104                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7105                         set_block_group_ro(cache, 1);
7106         }
7107
7108         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7109                 if (!(get_alloc_profile(root, space_info->flags) &
7110                       (BTRFS_BLOCK_GROUP_RAID10 |
7111                        BTRFS_BLOCK_GROUP_RAID1 |
7112                        BTRFS_BLOCK_GROUP_DUP)))
7113                         continue;
7114                 /*
7115                  * avoid allocating from un-mirrored block group if there are
7116                  * mirrored block groups.
7117                  */
7118                 list_for_each_entry(cache, &space_info->block_groups[3], list)
7119                         set_block_group_ro(cache, 1);
7120                 list_for_each_entry(cache, &space_info->block_groups[4], list)
7121                         set_block_group_ro(cache, 1);
7122         }
7123
7124         init_global_block_rsv(info);
7125         ret = 0;
7126 error:
7127         btrfs_free_path(path);
7128         return ret;
7129 }
7130
7131 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7132                            struct btrfs_root *root, u64 bytes_used,
7133                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7134                            u64 size)
7135 {
7136         int ret;
7137         struct btrfs_root *extent_root;
7138         struct btrfs_block_group_cache *cache;
7139
7140         extent_root = root->fs_info->extent_root;
7141
7142         root->fs_info->last_trans_log_full_commit = trans->transid;
7143
7144         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7145         if (!cache)
7146                 return -ENOMEM;
7147         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7148                                         GFP_NOFS);
7149         if (!cache->free_space_ctl) {
7150                 kfree(cache);
7151                 return -ENOMEM;
7152         }
7153
7154         cache->key.objectid = chunk_offset;
7155         cache->key.offset = size;
7156         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7157         cache->sectorsize = root->sectorsize;
7158         cache->fs_info = root->fs_info;
7159
7160         atomic_set(&cache->count, 1);
7161         spin_lock_init(&cache->lock);
7162         INIT_LIST_HEAD(&cache->list);
7163         INIT_LIST_HEAD(&cache->cluster_list);
7164
7165         btrfs_init_free_space_ctl(cache);
7166
7167         btrfs_set_block_group_used(&cache->item, bytes_used);
7168         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7169         cache->flags = type;
7170         btrfs_set_block_group_flags(&cache->item, type);
7171
7172         cache->last_byte_to_unpin = (u64)-1;
7173         cache->cached = BTRFS_CACHE_FINISHED;
7174         exclude_super_stripes(root, cache);
7175
7176         add_new_free_space(cache, root->fs_info, chunk_offset,
7177                            chunk_offset + size);
7178
7179         free_excluded_extents(root, cache);
7180
7181         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7182                                 &cache->space_info);
7183         BUG_ON(ret);
7184
7185         spin_lock(&cache->space_info->lock);
7186         cache->space_info->bytes_readonly += cache->bytes_super;
7187         spin_unlock(&cache->space_info->lock);
7188
7189         __link_block_group(cache->space_info, cache);
7190
7191         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7192         BUG_ON(ret);
7193
7194         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7195                                 sizeof(cache->item));
7196         BUG_ON(ret);
7197
7198         set_avail_alloc_bits(extent_root->fs_info, type);
7199
7200         return 0;
7201 }
7202
7203 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7204                              struct btrfs_root *root, u64 group_start)
7205 {
7206         struct btrfs_path *path;
7207         struct btrfs_block_group_cache *block_group;
7208         struct btrfs_free_cluster *cluster;
7209         struct btrfs_root *tree_root = root->fs_info->tree_root;
7210         struct btrfs_key key;
7211         struct inode *inode;
7212         int ret;
7213         int factor;
7214
7215         root = root->fs_info->extent_root;
7216
7217         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7218         BUG_ON(!block_group);
7219         BUG_ON(!block_group->ro);
7220
7221         /*
7222          * Free the reserved super bytes from this block group before
7223          * remove it.
7224          */
7225         free_excluded_extents(root, block_group);
7226
7227         memcpy(&key, &block_group->key, sizeof(key));
7228         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7229                                   BTRFS_BLOCK_GROUP_RAID1 |
7230                                   BTRFS_BLOCK_GROUP_RAID10))
7231                 factor = 2;
7232         else
7233                 factor = 1;
7234
7235         /* make sure this block group isn't part of an allocation cluster */
7236         cluster = &root->fs_info->data_alloc_cluster;
7237         spin_lock(&cluster->refill_lock);
7238         btrfs_return_cluster_to_free_space(block_group, cluster);
7239         spin_unlock(&cluster->refill_lock);
7240
7241         /*
7242          * make sure this block group isn't part of a metadata
7243          * allocation cluster
7244          */
7245         cluster = &root->fs_info->meta_alloc_cluster;
7246         spin_lock(&cluster->refill_lock);
7247         btrfs_return_cluster_to_free_space(block_group, cluster);
7248         spin_unlock(&cluster->refill_lock);
7249
7250         path = btrfs_alloc_path();
7251         if (!path) {
7252                 ret = -ENOMEM;
7253                 goto out;
7254         }
7255
7256         inode = lookup_free_space_inode(root, block_group, path);
7257         if (!IS_ERR(inode)) {
7258                 ret = btrfs_orphan_add(trans, inode);
7259                 BUG_ON(ret);
7260                 clear_nlink(inode);
7261                 /* One for the block groups ref */
7262                 spin_lock(&block_group->lock);
7263                 if (block_group->iref) {
7264                         block_group->iref = 0;
7265                         block_group->inode = NULL;
7266                         spin_unlock(&block_group->lock);
7267                         iput(inode);
7268                 } else {
7269                         spin_unlock(&block_group->lock);
7270                 }
7271                 /* One for our lookup ref */
7272                 iput(inode);
7273         }
7274
7275         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7276         key.offset = block_group->key.objectid;
7277         key.type = 0;
7278
7279         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7280         if (ret < 0)
7281                 goto out;
7282         if (ret > 0)
7283                 btrfs_release_path(path);
7284         if (ret == 0) {
7285                 ret = btrfs_del_item(trans, tree_root, path);
7286                 if (ret)
7287                         goto out;
7288                 btrfs_release_path(path);
7289         }
7290
7291         spin_lock(&root->fs_info->block_group_cache_lock);
7292         rb_erase(&block_group->cache_node,
7293                  &root->fs_info->block_group_cache_tree);
7294         spin_unlock(&root->fs_info->block_group_cache_lock);
7295
7296         down_write(&block_group->space_info->groups_sem);
7297         /*
7298          * we must use list_del_init so people can check to see if they
7299          * are still on the list after taking the semaphore
7300          */
7301         list_del_init(&block_group->list);
7302         up_write(&block_group->space_info->groups_sem);
7303
7304         if (block_group->cached == BTRFS_CACHE_STARTED)
7305                 wait_block_group_cache_done(block_group);
7306
7307         btrfs_remove_free_space_cache(block_group);
7308
7309         spin_lock(&block_group->space_info->lock);
7310         block_group->space_info->total_bytes -= block_group->key.offset;
7311         block_group->space_info->bytes_readonly -= block_group->key.offset;
7312         block_group->space_info->disk_total -= block_group->key.offset * factor;
7313         spin_unlock(&block_group->space_info->lock);
7314
7315         memcpy(&key, &block_group->key, sizeof(key));
7316
7317         btrfs_clear_space_info_full(root->fs_info);
7318
7319         btrfs_put_block_group(block_group);
7320         btrfs_put_block_group(block_group);
7321
7322         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7323         if (ret > 0)
7324                 ret = -EIO;
7325         if (ret < 0)
7326                 goto out;
7327
7328         ret = btrfs_del_item(trans, root, path);
7329 out:
7330         btrfs_free_path(path);
7331         return ret;
7332 }
7333
7334 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7335 {
7336         struct btrfs_space_info *space_info;
7337         struct btrfs_super_block *disk_super;
7338         u64 features;
7339         u64 flags;
7340         int mixed = 0;
7341         int ret;
7342
7343         disk_super = &fs_info->super_copy;
7344         if (!btrfs_super_root(disk_super))
7345                 return 1;
7346
7347         features = btrfs_super_incompat_flags(disk_super);
7348         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7349                 mixed = 1;
7350
7351         flags = BTRFS_BLOCK_GROUP_SYSTEM;
7352         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7353         if (ret)
7354                 goto out;
7355
7356         if (mixed) {
7357                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7358                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7359         } else {
7360                 flags = BTRFS_BLOCK_GROUP_METADATA;
7361                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7362                 if (ret)
7363                         goto out;
7364
7365                 flags = BTRFS_BLOCK_GROUP_DATA;
7366                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7367         }
7368 out:
7369         return ret;
7370 }
7371
7372 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7373 {
7374         return unpin_extent_range(root, start, end);
7375 }
7376
7377 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7378                                u64 num_bytes, u64 *actual_bytes)
7379 {
7380         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7381 }
7382
7383 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7384 {
7385         struct btrfs_fs_info *fs_info = root->fs_info;
7386         struct btrfs_block_group_cache *cache = NULL;
7387         u64 group_trimmed;
7388         u64 start;
7389         u64 end;
7390         u64 trimmed = 0;
7391         int ret = 0;
7392
7393         cache = btrfs_lookup_block_group(fs_info, range->start);
7394
7395         while (cache) {
7396                 if (cache->key.objectid >= (range->start + range->len)) {
7397                         btrfs_put_block_group(cache);
7398                         break;
7399                 }
7400
7401                 start = max(range->start, cache->key.objectid);
7402                 end = min(range->start + range->len,
7403                                 cache->key.objectid + cache->key.offset);
7404
7405                 if (end - start >= range->minlen) {
7406                         if (!block_group_cache_done(cache)) {
7407                                 ret = cache_block_group(cache, NULL, root, 0);
7408                                 if (!ret)
7409                                         wait_block_group_cache_done(cache);
7410                         }
7411                         ret = btrfs_trim_block_group(cache,
7412                                                      &group_trimmed,
7413                                                      start,
7414                                                      end,
7415                                                      range->minlen);
7416
7417                         trimmed += group_trimmed;
7418                         if (ret) {
7419                                 btrfs_put_block_group(cache);
7420                                 break;
7421                         }
7422                 }
7423
7424                 cache = next_block_group(fs_info->tree_root, cache);
7425         }
7426
7427         range->len = trimmed;
7428         return ret;
7429 }