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 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);
 109         }
 110 }
 111
 112 /*
 113  * this adds the block group to the fs_info rb tree for the block group
 114  * cache
 115  */
 116 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 117                                 struct btrfs_block_group_cache *block_group)
 118 {
 119         struct rb_node **p;
 120         struct rb_node *parent = NULL;
 121         struct btrfs_block_group_cache *cache;
 122
 123         spin_lock(&info->block_group_cache_lock);
 124         p = &info->block_group_cache_tree.rb_node;
 125
 126         while (*p) {
 127                 parent = *p;
 128                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 129                                  cache_node);
 130                 if (block_group->key.objectid < cache->key.objectid) {
 131                         p = &(*p)->rb_left;
 132                 } else if (block_group->key.objectid > cache->key.objectid) {
 133                         p = &(*p)->rb_right;
 134                 } else {
 135                         spin_unlock(&info->block_group_cache_lock);
 136                         return -EEXIST;
 137                 }
 138         }
 139
 140         rb_link_node(&block_group->cache_node, parent, p);
 141         rb_insert_color(&block_group->cache_node,
 142                         &info->block_group_cache_tree);
 143         spin_unlock(&info->block_group_cache_lock);
 144
 145         return 0;
 146 }
 147
 148 /*
 149  * This will return the block group at or after bytenr if contains is 0, else
 150  * it will return the block group that contains the bytenr
 151  */
 152 static struct btrfs_block_group_cache *
 153 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 154                               int contains)
 155 {
 156         struct btrfs_block_group_cache *cache, *ret = NULL;
 157         struct rb_node *n;
 158         u64 end, start;
 159
 160         spin_lock(&info->block_group_cache_lock);
 161         n = info->block_group_cache_tree.rb_node;
 162
 163         while (n) {
 164                 cache = rb_entry(n, struct btrfs_block_group_cache,
 165                                  cache_node);
 166                 end = cache->key.objectid + cache->key.offset - 1;
 167                 start = cache->key.objectid;
 168
 169                 if (bytenr < start) {
 170                         if (!contains && (!ret || start < ret->key.objectid))
 171                                 ret = cache;
 172                         n = n->rb_left;
 173                 } else if (bytenr > start) {
 174                         if (contains && bytenr <= end) {
 175                                 ret = cache;
 176                                 break;
 177                         }
 178                         n = n->rb_right;
 179                 } else {
 180                         ret = cache;
 181                         break;
 182                 }
 183         }
 184         if (ret)
 185                 btrfs_get_block_group(ret);
 186         spin_unlock(&info->block_group_cache_lock);
 187
 188         return ret;
 189 }
 190
 191 static int add_excluded_extent(struct btrfs_root *root,
 192                                u64 start, u64 num_bytes)
 193 {
 194         u64 end = start + num_bytes - 1;
 195         set_extent_bits(&root->fs_info->freed_extents[0],
 196                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 197         set_extent_bits(&root->fs_info->freed_extents[1],
 198                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 199         return 0;
 200 }
 201
 202 static void free_excluded_extents(struct btrfs_root *root,
 203                                   struct btrfs_block_group_cache *cache)
 204 {
 205         u64 start, end;
 206
 207         start = cache->key.objectid;
 208         end = start + cache->key.offset - 1;
 209
 210         clear_extent_bits(&root->fs_info->freed_extents[0],
 211                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 212         clear_extent_bits(&root->fs_info->freed_extents[1],
 213                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 214 }
 215
 216 static int exclude_super_stripes(struct btrfs_root *root,
 217                                  struct btrfs_block_group_cache *cache)
 218 {
 219         u64 bytenr;
 220         u64 *logical;
 221         int stripe_len;
 222         int i, nr, ret;
 223
 224         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 225                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 226                 cache->bytes_super += stripe_len;
 227                 ret = add_excluded_extent(root, cache->key.objectid,
 228                                           stripe_len);
 229                 BUG_ON(ret);
 230         }
 231
 232         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 233                 bytenr = btrfs_sb_offset(i);
 234                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 235                                        cache->key.objectid, bytenr,
 236                                        0, &logical, &nr, &stripe_len);
 237                 BUG_ON(ret);
 238
 239                 while (nr--) {
 240                         cache->bytes_super += stripe_len;
 241                         ret = add_excluded_extent(root, logical[nr],
 242                                                   stripe_len);
 243                         BUG_ON(ret);
 244                 }
 245
 246                 kfree(logical);
 247         }
 248         return 0;
 249 }
 250
 251 static struct btrfs_caching_control *
 252 get_caching_control(struct btrfs_block_group_cache *cache)
 253 {
 254         struct btrfs_caching_control *ctl;
 255
 256         spin_lock(&cache->lock);
 257         if (cache->cached != BTRFS_CACHE_STARTED) {
 258                 spin_unlock(&cache->lock);
 259                 return NULL;
 260         }
 261
 262         /* We're loading it the fast way, so we don't have a caching_ctl. */
 263         if (!cache->caching_ctl) {
 264                 spin_unlock(&cache->lock);
 265                 return NULL;
 266         }
 267
 268         ctl = cache->caching_ctl;
 269         atomic_inc(&ctl->count);
 270         spin_unlock(&cache->lock);
 271         return ctl;
 272 }
 273
 274 static void put_caching_control(struct btrfs_caching_control *ctl)
 275 {
 276         if (atomic_dec_and_test(&ctl->count))
 277                 kfree(ctl);
 278 }
 279
 280 /*
 281  * this is only called by cache_block_group, since we could have freed extents
 282  * we need to check the pinned_extents for any extents that can't be used yet
 283  * since their free space will be released as soon as the transaction commits.
 284  */
 285 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 286                               struct btrfs_fs_info *info, u64 start, u64 end)
 287 {
 288         u64 extent_start, extent_end, size, total_added = 0;
 289         int ret;
 290
 291         while (start < end) {
 292                 ret = find_first_extent_bit(info->pinned_extents, start,
 293                                             &extent_start, &extent_end,
 294                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 295                 if (ret)
 296                         break;
 297
 298                 if (extent_start <= start) {
 299                         start = extent_end + 1;
 300                 } else if (extent_start > start && extent_start < end) {
 301                         size = extent_start - start;
 302                         total_added += size;
 303                         ret = btrfs_add_free_space(block_group, start,
 304                                                    size);
 305                         BUG_ON(ret);
 306                         start = extent_end + 1;
 307                 } else {
 308                         break;
 309                 }
 310         }
 311
 312         if (start < end) {
 313                 size = end - start;
 314                 total_added += size;
 315                 ret = btrfs_add_free_space(block_group, start, size);
 316                 BUG_ON(ret);
 317         }
 318
 319         return total_added;
 320 }
 321
 322 static int caching_kthread(void *data)
 323 {
 324         struct btrfs_block_group_cache *block_group = data;
 325         struct btrfs_fs_info *fs_info = block_group->fs_info;
 326         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 327         struct btrfs_root *extent_root = fs_info->extent_root;
 328         struct btrfs_path *path;
 329         struct extent_buffer *leaf;
 330         struct btrfs_key key;
 331         u64 total_found = 0;
 332         u64 last = 0;
 333         u32 nritems;
 334         int ret = 0;
 335
 336         path = btrfs_alloc_path();
 337         if (!path)
 338                 return -ENOMEM;
 339
 340         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 341
 342         /*
 343          * We don't want to deadlock with somebody trying to allocate a new
 344          * extent for the extent root while also trying to search the extent
 345          * root to add free space.  So we skip locking and search the commit
 346          * root, since its read-only
 347          */
 348         path->skip_locking = 1;
 349         path->search_commit_root = 1;
 350         path->reada = 2;
 351
 352         key.objectid = last;
 353         key.offset = 0;
 354         key.type = BTRFS_EXTENT_ITEM_KEY;
 355 again:
 356         mutex_lock(&caching_ctl->mutex);
 357         /* need to make sure the commit_root doesn't disappear */
 358         down_read(&fs_info->extent_commit_sem);
 359
 360         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 361         if (ret < 0)
 362                 goto err;
 363
 364         leaf = path->nodes[0];
 365         nritems = btrfs_header_nritems(leaf);
 366
 367         while (1) {
 368                 smp_mb();
 369                 if (fs_info->closing > 1) {
 370                         last = (u64)-1;
 371                         break;
 372                 }
 373
 374                 if (path->slots[0] < nritems) {
 375                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 376                 } else {
 377                         ret = find_next_key(path, 0, &key);
 378                         if (ret)
 379                                 break;
 380
 381                         caching_ctl->progress = last;
 382                         btrfs_release_path(extent_root, path);
 383                         up_read(&fs_info->extent_commit_sem);
 384                         mutex_unlock(&caching_ctl->mutex);
 385                         if (btrfs_transaction_in_commit(fs_info))
 386                                 schedule_timeout(1);
 387                         else
 388                                 cond_resched();
 389                         goto again;
 390                 }
 391
 392                 if (key.objectid < block_group->key.objectid) {
 393                         path->slots[0]++;
 394                         continue;
 395                 }
 396
 397                 if (key.objectid >= block_group->key.objectid +
 398                     block_group->key.offset)
 399                         break;
 400
 401                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 402                         total_found += add_new_free_space(block_group,
 403                                                           fs_info, last,
 404                                                           key.objectid);
 405                         last = key.objectid + key.offset;
 406
 407                         if (total_found > (1024 * 1024 * 2)) {
 408                                 total_found = 0;
 409                                 wake_up(&caching_ctl->wait);
 410                         }
 411                 }
 412                 path->slots[0]++;
 413         }
 414         ret = 0;
 415
 416         total_found += add_new_free_space(block_group, fs_info, last,
 417                                           block_group->key.objectid +
 418                                           block_group->key.offset);
 419         caching_ctl->progress = (u64)-1;
 420
 421         spin_lock(&block_group->lock);
 422         block_group->caching_ctl = NULL;
 423         block_group->cached = BTRFS_CACHE_FINISHED;
 424         spin_unlock(&block_group->lock);
 425
 426 err:
 427         btrfs_free_path(path);
 428         up_read(&fs_info->extent_commit_sem);
 429
 430         free_excluded_extents(extent_root, block_group);
 431
 432         mutex_unlock(&caching_ctl->mutex);
 433         wake_up(&caching_ctl->wait);
 434
 435         put_caching_control(caching_ctl);
 436         atomic_dec(&block_group->space_info->caching_threads);
 437         btrfs_put_block_group(block_group);
 438
 439         return 0;
 440 }
 441
 442 static int cache_block_group(struct btrfs_block_group_cache *cache,
 443                              struct btrfs_trans_handle *trans,
 444                              struct btrfs_root *root,
 445                              int load_cache_only)
 446 {
 447         struct btrfs_fs_info *fs_info = cache->fs_info;
 448         struct btrfs_caching_control *caching_ctl;
 449         struct task_struct *tsk;
 450         int ret = 0;
 451
 452         smp_mb();
 453         if (cache->cached != BTRFS_CACHE_NO)
 454                 return 0;
 455
 456         /*
 457          * We can't do the read from on-disk cache during a commit since we need
 458          * to have the normal tree locking.  Also if we are currently trying to
 459          * allocate blocks for the tree root we can't do the fast caching since
 460          * we likely hold important locks.
 461          */
 462         if (trans && (!trans->transaction->in_commit) &&
 463             (root && root != root->fs_info->tree_root)) {
 464                 spin_lock(&cache->lock);
 465                 if (cache->cached != BTRFS_CACHE_NO) {
 466                         spin_unlock(&cache->lock);
 467                         return 0;
 468                 }
 469                 cache->cached = BTRFS_CACHE_STARTED;
 470                 spin_unlock(&cache->lock);
 471
 472                 ret = load_free_space_cache(fs_info, cache);
 473
 474                 spin_lock(&cache->lock);
 475                 if (ret == 1) {
 476                         cache->cached = BTRFS_CACHE_FINISHED;
 477                         cache->last_byte_to_unpin = (u64)-1;
 478                 } else {
 479                         cache->cached = BTRFS_CACHE_NO;
 480                 }
 481                 spin_unlock(&cache->lock);
 482                 if (ret == 1) {
 483                         free_excluded_extents(fs_info->extent_root, cache);
 484                         return 0;
 485                 }
 486         }
 487
 488         if (load_cache_only)
 489                 return 0;
 490
 491         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 492         BUG_ON(!caching_ctl);
 493
 494         INIT_LIST_HEAD(&caching_ctl->list);
 495         mutex_init(&caching_ctl->mutex);
 496         init_waitqueue_head(&caching_ctl->wait);
 497         caching_ctl->block_group = cache;
 498         caching_ctl->progress = cache->key.objectid;
 499         /* one for caching kthread, one for caching block group list */
 500         atomic_set(&caching_ctl->count, 2);
 501
 502         spin_lock(&cache->lock);
 503         if (cache->cached != BTRFS_CACHE_NO) {
 504                 spin_unlock(&cache->lock);
 505                 kfree(caching_ctl);
 506                 return 0;
 507         }
 508         cache->caching_ctl = caching_ctl;
 509         cache->cached = BTRFS_CACHE_STARTED;
 510         spin_unlock(&cache->lock);
 511
 512         down_write(&fs_info->extent_commit_sem);
 513         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 514         up_write(&fs_info->extent_commit_sem);
 515
 516         atomic_inc(&cache->space_info->caching_threads);
 517         btrfs_get_block_group(cache);
 518
 519         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 520                           cache->key.objectid);
 521         if (IS_ERR(tsk)) {
 522                 ret = PTR_ERR(tsk);
 523                 printk(KERN_ERR "error running thread %d\n", ret);
 524                 BUG();
 525         }
 526
 527         return ret;
 528 }
 529
 530 /*
 531  * return the block group that starts at or after bytenr
 532  */
 533 static struct btrfs_block_group_cache *
 534 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 535 {
 536         struct btrfs_block_group_cache *cache;
 537
 538         cache = block_group_cache_tree_search(info, bytenr, 0);
 539
 540         return cache;
 541 }
 542
 543 /*
 544  * return the block group that contains the given bytenr
 545  */
 546 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 547                                                  struct btrfs_fs_info *info,
 548                                                  u64 bytenr)
 549 {
 550         struct btrfs_block_group_cache *cache;
 551
 552         cache = block_group_cache_tree_search(info, bytenr, 1);
 553
 554         return cache;
 555 }
 556
 557 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 558                                                   u64 flags)
 559 {
 560         struct list_head *head = &info->space_info;
 561         struct btrfs_space_info *found;
 562
 563         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 564                  BTRFS_BLOCK_GROUP_METADATA;
 565
 566         rcu_read_lock();
 567         list_for_each_entry_rcu(found, head, list) {
 568                 if (found->flags & flags) {
 569                         rcu_read_unlock();
 570                         return found;
 571                 }
 572         }
 573         rcu_read_unlock();
 574         return NULL;
 575 }
 576
 577 /*
 578  * after adding space to the filesystem, we need to clear the full flags
 579  * on all the space infos.
 580  */
 581 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 582 {
 583         struct list_head *head = &info->space_info;
 584         struct btrfs_space_info *found;
 585
 586         rcu_read_lock();
 587         list_for_each_entry_rcu(found, head, list)
 588                 found->full = 0;
 589         rcu_read_unlock();
 590 }
 591
 592 static u64 div_factor(u64 num, int factor)
 593 {
 594         if (factor == 10)
 595                 return num;
 596         num *= factor;
 597         do_div(num, 10);
 598         return num;
 599 }
 600
 601 static u64 div_factor_fine(u64 num, int factor)
 602 {
 603         if (factor == 100)
 604                 return num;
 605         num *= factor;
 606         do_div(num, 100);
 607         return num;
 608 }
 609
 610 u64 btrfs_find_block_group(struct btrfs_root *root,
 611                            u64 search_start, u64 search_hint, int owner)
 612 {
 613         struct btrfs_block_group_cache *cache;
 614         u64 used;
 615         u64 last = max(search_hint, search_start);
 616         u64 group_start = 0;
 617         int full_search = 0;
 618         int factor = 9;
 619         int wrapped = 0;
 620 again:
 621         while (1) {
 622                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 623                 if (!cache)
 624                         break;
 625
 626                 spin_lock(&cache->lock);
 627                 last = cache->key.objectid + cache->key.offset;
 628                 used = btrfs_block_group_used(&cache->item);
 629
 630                 if ((full_search || !cache->ro) &&
 631                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 632                         if (used + cache->pinned + cache->reserved <
 633                             div_factor(cache->key.offset, factor)) {
 634                                 group_start = cache->key.objectid;
 635                                 spin_unlock(&cache->lock);
 636                                 btrfs_put_block_group(cache);
 637                                 goto found;
 638                         }
 639                 }
 640                 spin_unlock(&cache->lock);
 641                 btrfs_put_block_group(cache);
 642                 cond_resched();
 643         }
 644         if (!wrapped) {
 645                 last = search_start;
 646                 wrapped = 1;
 647                 goto again;
 648         }
 649         if (!full_search && factor < 10) {
 650                 last = search_start;
 651                 full_search = 1;
 652                 factor = 10;
 653                 goto again;
 654         }
 655 found:
 656         return group_start;
 657 }
 658
 659 /* simple helper to search for an existing extent at a given offset */
 660 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 661 {
 662         int ret;
 663         struct btrfs_key key;
 664         struct btrfs_path *path;
 665
 666         path = btrfs_alloc_path();
 667         BUG_ON(!path);
 668         key.objectid = start;
 669         key.offset = len;
 670         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 671         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 672                                 0, 0);
 673         btrfs_free_path(path);
 674         return ret;
 675 }
 676
 677 /*
 678  * helper function to lookup reference count and flags of extent.
 679  *
 680  * the head node for delayed ref is used to store the sum of all the
 681  * reference count modifications queued up in the rbtree. the head
 682  * node may also store the extent flags to set. This way you can check
 683  * to see what the reference count and extent flags would be if all of
 684  * the delayed refs are not processed.
 685  */
 686 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 687                              struct btrfs_root *root, u64 bytenr,
 688                              u64 num_bytes, u64 *refs, u64 *flags)
 689 {
 690         struct btrfs_delayed_ref_head *head;
 691         struct btrfs_delayed_ref_root *delayed_refs;
 692         struct btrfs_path *path;
 693         struct btrfs_extent_item *ei;
 694         struct extent_buffer *leaf;
 695         struct btrfs_key key;
 696         u32 item_size;
 697         u64 num_refs;
 698         u64 extent_flags;
 699         int ret;
 700
 701         path = btrfs_alloc_path();
 702         if (!path)
 703                 return -ENOMEM;
 704
 705         key.objectid = bytenr;
 706         key.type = BTRFS_EXTENT_ITEM_KEY;
 707         key.offset = num_bytes;
 708         if (!trans) {
 709                 path->skip_locking = 1;
 710                 path->search_commit_root = 1;
 711         }
 712 again:
 713         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 714                                 &key, path, 0, 0);
 715         if (ret < 0)
 716                 goto out_free;
 717
 718         if (ret == 0) {
 719                 leaf = path->nodes[0];
 720                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 721                 if (item_size >= sizeof(*ei)) {
 722                         ei = btrfs_item_ptr(leaf, path->slots[0],
 723                                             struct btrfs_extent_item);
 724                         num_refs = btrfs_extent_refs(leaf, ei);
 725                         extent_flags = btrfs_extent_flags(leaf, ei);
 726                 } else {
 727 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 728                         struct btrfs_extent_item_v0 *ei0;
 729                         BUG_ON(item_size != sizeof(*ei0));
 730                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 731                                              struct btrfs_extent_item_v0);
 732                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 733                         /* FIXME: this isn't correct for data */
 734                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 735 #else
 736                         BUG();
 737 #endif
 738                 }
 739                 BUG_ON(num_refs == 0);
 740         } else {
 741                 num_refs = 0;
 742                 extent_flags = 0;
 743                 ret = 0;
 744         }
 745
 746         if (!trans)
 747                 goto out;
 748
 749         delayed_refs = &trans->transaction->delayed_refs;
 750         spin_lock(&delayed_refs->lock);
 751         head = btrfs_find_delayed_ref_head(trans, bytenr);
 752         if (head) {
 753                 if (!mutex_trylock(&head->mutex)) {
 754                         atomic_inc(&head->node.refs);
 755                         spin_unlock(&delayed_refs->lock);
 756
 757                         btrfs_release_path(root->fs_info->extent_root, path);
 758
 759                         mutex_lock(&head->mutex);
 760                         mutex_unlock(&head->mutex);
 761                         btrfs_put_delayed_ref(&head->node);
 762                         goto again;
 763                 }
 764                 if (head->extent_op && head->extent_op->update_flags)
 765                         extent_flags |= head->extent_op->flags_to_set;
 766                 else
 767                         BUG_ON(num_refs == 0);
 768
 769                 num_refs += head->node.ref_mod;
 770                 mutex_unlock(&head->mutex);
 771         }
 772         spin_unlock(&delayed_refs->lock);
 773 out:
 774         WARN_ON(num_refs == 0);
 775         if (refs)
 776                 *refs = num_refs;
 777         if (flags)
 778                 *flags = extent_flags;
 779 out_free:
 780         btrfs_free_path(path);
 781         return ret;
 782 }
 783
 784 /*
 785  * Back reference rules.  Back refs have three main goals:
 786  *
 787  * 1) differentiate between all holders of references to an extent so that
 788  *    when a reference is dropped we can make sure it was a valid reference
 789  *    before freeing the extent.
 790  *
 791  * 2) Provide enough information to quickly find the holders of an extent
 792  *    if we notice a given block is corrupted or bad.
 793  *
 794  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 795  *    maintenance.  This is actually the same as #2, but with a slightly
 796  *    different use case.
 797  *
 798  * There are two kinds of back refs. The implicit back refs is optimized
 799  * for pointers in non-shared tree blocks. For a given pointer in a block,
 800  * back refs of this kind provide information about the block's owner tree
 801  * and the pointer's key. These information allow us to find the block by
 802  * b-tree searching. The full back refs is for pointers in tree blocks not
 803  * referenced by their owner trees. The location of tree block is recorded
 804  * in the back refs. Actually the full back refs is generic, and can be
 805  * used in all cases the implicit back refs is used. The major shortcoming
 806  * of the full back refs is its overhead. Every time a tree block gets
 807  * COWed, we have to update back refs entry for all pointers in it.
 808  *
 809  * For a newly allocated tree block, we use implicit back refs for
 810  * pointers in it. This means most tree related operations only involve
 811  * implicit back refs. For a tree block created in old transaction, the
 812  * only way to drop a reference to it is COW it. So we can detect the
 813  * event that tree block loses its owner tree's reference and do the
 814  * back refs conversion.
 815  *
 816  * When a tree block is COW'd through a tree, there are four cases:
 817  *
 818  * The reference count of the block is one and the tree is the block's
 819  * owner tree. Nothing to do in this case.
 820  *
 821  * The reference count of the block is one and the tree is not the
 822  * block's owner tree. In this case, full back refs is used for pointers
 823  * in the block. Remove these full back refs, add implicit back refs for
 824  * every pointers in the new block.
 825  *
 826  * The reference count of the block is greater than one and the tree is
 827  * the block's owner tree. In this case, implicit back refs is used for
 828  * pointers in the block. Add full back refs for every pointers in the
 829  * block, increase lower level extents' reference counts. The original
 830  * implicit back refs are entailed to the new block.
 831  *
 832  * The reference count of the block is greater than one and the tree is
 833  * not the block's owner tree. Add implicit back refs for every pointer in
 834  * the new block, increase lower level extents' reference count.
 835  *
 836  * Back Reference Key composing:
 837  *
 838  * The key objectid corresponds to the first byte in the extent,
 839  * The key type is used to differentiate between types of back refs.
 840  * There are different meanings of the key offset for different types
 841  * of back refs.
 842  *
 843  * File extents can be referenced by:
 844  *
 845  * - multiple snapshots, subvolumes, or different generations in one subvol
 846  * - different files inside a single subvolume
 847  * - different offsets inside a file (bookend extents in file.c)
 848  *
 849  * The extent ref structure for the implicit back refs has fields for:
 850  *
 851  * - Objectid of the subvolume root
 852  * - objectid of the file holding the reference
 853  * - original offset in the file
 854  * - how many bookend extents
 855  *
 856  * The key offset for the implicit back refs is hash of the first
 857  * three fields.
 858  *
 859  * The extent ref structure for the full back refs has field for:
 860  *
 861  * - number of pointers in the tree leaf
 862  *
 863  * The key offset for the implicit back refs is the first byte of
 864  * the tree leaf
 865  *
 866  * When a file extent is allocated, The implicit back refs is used.
 867  * the fields are filled in:
 868  *
 869  *     (root_key.objectid, inode objectid, offset in file, 1)
 870  *
 871  * When a file extent is removed file truncation, we find the
 872  * corresponding implicit back refs and check the following fields:
 873  *
 874  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 875  *
 876  * Btree extents can be referenced by:
 877  *
 878  * - Different subvolumes
 879  *
 880  * Both the implicit back refs and the full back refs for tree blocks
 881  * only consist of key. The key offset for the implicit back refs is
 882  * objectid of block's owner tree. The key offset for the full back refs
 883  * is the first byte of parent block.
 884  *
 885  * When implicit back refs is used, information about the lowest key and
 886  * level of the tree block are required. These information are stored in
 887  * tree block info structure.
 888  */
 889
 890 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 891 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 892                                   struct btrfs_root *root,
 893                                   struct btrfs_path *path,
 894                                   u64 owner, u32 extra_size)
 895 {
 896         struct btrfs_extent_item *item;
 897         struct btrfs_extent_item_v0 *ei0;
 898         struct btrfs_extent_ref_v0 *ref0;
 899         struct btrfs_tree_block_info *bi;
 900         struct extent_buffer *leaf;
 901         struct btrfs_key key;
 902         struct btrfs_key found_key;
 903         u32 new_size = sizeof(*item);
 904         u64 refs;
 905         int ret;
 906
 907         leaf = path->nodes[0];
 908         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 909
 910         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 911         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 912                              struct btrfs_extent_item_v0);
 913         refs = btrfs_extent_refs_v0(leaf, ei0);
 914
 915         if (owner == (u64)-1) {
 916                 while (1) {
 917                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 918                                 ret = btrfs_next_leaf(root, path);
 919                                 if (ret < 0)
 920                                         return ret;
 921                                 BUG_ON(ret > 0);
 922                                 leaf = path->nodes[0];
 923                         }
 924                         btrfs_item_key_to_cpu(leaf, &found_key,
 925                                               path->slots[0]);
 926                         BUG_ON(key.objectid != found_key.objectid);
 927                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 928                                 path->slots[0]++;
 929                                 continue;
 930                         }
 931                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 932                                               struct btrfs_extent_ref_v0);
 933                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 934                         break;
 935                 }
 936         }
 937         btrfs_release_path(root, path);
 938
 939         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 940                 new_size += sizeof(*bi);
 941
 942         new_size -= sizeof(*ei0);
 943         ret = btrfs_search_slot(trans, root, &key, path,
 944                                 new_size + extra_size, 1);
 945         if (ret < 0)
 946                 return ret;
 947         BUG_ON(ret);
 948
 949         ret = btrfs_extend_item(trans, root, path, new_size);
 950         BUG_ON(ret);
 951
 952         leaf = path->nodes[0];
 953         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 954         btrfs_set_extent_refs(leaf, item, refs);
 955         /* FIXME: get real generation */
 956         btrfs_set_extent_generation(leaf, item, 0);
 957         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 958                 btrfs_set_extent_flags(leaf, item,
 959                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 960                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 961                 bi = (struct btrfs_tree_block_info *)(item + 1);
 962                 /* FIXME: get first key of the block */
 963                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 964                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 965         } else {
 966                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 967         }
 968         btrfs_mark_buffer_dirty(leaf);
 969         return 0;
 970 }
 971 #endif
 972
 973 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 974 {
 975         u32 high_crc = ~(u32)0;
 976         u32 low_crc = ~(u32)0;
 977         __le64 lenum;
 978
 979         lenum = cpu_to_le64(root_objectid);
 980         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 981         lenum = cpu_to_le64(owner);
 982         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 983         lenum = cpu_to_le64(offset);
 984         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 985
 986         return ((u64)high_crc << 31) ^ (u64)low_crc;
 987 }
 988
 989 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 990                                      struct btrfs_extent_data_ref *ref)
 991 {
 992         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 993                                     btrfs_extent_data_ref_objectid(leaf, ref),
 994                                     btrfs_extent_data_ref_offset(leaf, ref));
 995 }
 996
 997 static int match_extent_data_ref(struct extent_buffer *leaf,
 998                                  struct btrfs_extent_data_ref *ref,
 999                                  u64 root_objectid, u64 owner, u64 offset)
1000 {
1001         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1002             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1003             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1004                 return 0;
1005         return 1;
1006 }
1007
1008 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1009                                            struct btrfs_root *root,
1010                                            struct btrfs_path *path,
1011                                            u64 bytenr, u64 parent,
1012                                            u64 root_objectid,
1013                                            u64 owner, u64 offset)
1014 {
1015         struct btrfs_key key;
1016         struct btrfs_extent_data_ref *ref;
1017         struct extent_buffer *leaf;
1018         u32 nritems;
1019         int ret;
1020         int recow;
1021         int err = -ENOENT;
1022
1023         key.objectid = bytenr;
1024         if (parent) {
1025                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1026                 key.offset = parent;
1027         } else {
1028                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1029                 key.offset = hash_extent_data_ref(root_objectid,
1030                                                   owner, offset);
1031         }
1032 again:
1033         recow = 0;
1034         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1035         if (ret < 0) {
1036                 err = ret;
1037                 goto fail;
1038         }
1039
1040         if (parent) {
1041                 if (!ret)
1042                         return 0;
1043 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1044                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1045                 btrfs_release_path(root, path);
1046                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1047                 if (ret < 0) {
1048                         err = ret;
1049                         goto fail;
1050                 }
1051                 if (!ret)
1052                         return 0;
1053 #endif
1054                 goto fail;
1055         }
1056
1057         leaf = path->nodes[0];
1058         nritems = btrfs_header_nritems(leaf);
1059         while (1) {
1060                 if (path->slots[0] >= nritems) {
1061                         ret = btrfs_next_leaf(root, path);
1062                         if (ret < 0)
1063                                 err = ret;
1064                         if (ret)
1065                                 goto fail;
1066
1067                         leaf = path->nodes[0];
1068                         nritems = btrfs_header_nritems(leaf);
1069                         recow = 1;
1070                 }
1071
1072                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1073                 if (key.objectid != bytenr ||
1074                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1075                         goto fail;
1076
1077                 ref = btrfs_item_ptr(leaf, path->slots[0],
1078                                      struct btrfs_extent_data_ref);
1079
1080                 if (match_extent_data_ref(leaf, ref, root_objectid,
1081                                           owner, offset)) {
1082                         if (recow) {
1083                                 btrfs_release_path(root, path);
1084                                 goto again;
1085                         }
1086                         err = 0;
1087                         break;
1088                 }
1089                 path->slots[0]++;
1090         }
1091 fail:
1092         return err;
1093 }
1094
1095 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1096                                            struct btrfs_root *root,
1097                                            struct btrfs_path *path,
1098                                            u64 bytenr, u64 parent,
1099                                            u64 root_objectid, u64 owner,
1100                                            u64 offset, int refs_to_add)
1101 {
1102         struct btrfs_key key;
1103         struct extent_buffer *leaf;
1104         u32 size;
1105         u32 num_refs;
1106         int ret;
1107
1108         key.objectid = bytenr;
1109         if (parent) {
1110                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1111                 key.offset = parent;
1112                 size = sizeof(struct btrfs_shared_data_ref);
1113         } else {
1114                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1115                 key.offset = hash_extent_data_ref(root_objectid,
1116                                                   owner, offset);
1117                 size = sizeof(struct btrfs_extent_data_ref);
1118         }
1119
1120         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1121         if (ret && ret != -EEXIST)
1122                 goto fail;
1123
1124         leaf = path->nodes[0];
1125         if (parent) {
1126                 struct btrfs_shared_data_ref *ref;
1127                 ref = btrfs_item_ptr(leaf, path->slots[0],
1128                                      struct btrfs_shared_data_ref);
1129                 if (ret == 0) {
1130                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1131                 } else {
1132                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1133                         num_refs += refs_to_add;
1134                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1135                 }
1136         } else {
1137                 struct btrfs_extent_data_ref *ref;
1138                 while (ret == -EEXIST) {
1139                         ref = btrfs_item_ptr(leaf, path->slots[0],
1140                                              struct btrfs_extent_data_ref);
1141                         if (match_extent_data_ref(leaf, ref, root_objectid,
1142                                                   owner, offset))
1143                                 break;
1144                         btrfs_release_path(root, path);
1145                         key.offset++;
1146                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1147                                                       size);
1148                         if (ret && ret != -EEXIST)
1149                                 goto fail;
1150
1151                         leaf = path->nodes[0];
1152                 }
1153                 ref = btrfs_item_ptr(leaf, path->slots[0],
1154                                      struct btrfs_extent_data_ref);
1155                 if (ret == 0) {
1156                         btrfs_set_extent_data_ref_root(leaf, ref,
1157                                                        root_objectid);
1158                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1159                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1160                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1161                 } else {
1162                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1163                         num_refs += refs_to_add;
1164                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1165                 }
1166         }
1167         btrfs_mark_buffer_dirty(leaf);
1168         ret = 0;
1169 fail:
1170         btrfs_release_path(root, path);
1171         return ret;
1172 }
1173
1174 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1175                                            struct btrfs_root *root,
1176                                            struct btrfs_path *path,
1177                                            int refs_to_drop)
1178 {
1179         struct btrfs_key key;
1180         struct btrfs_extent_data_ref *ref1 = NULL;
1181         struct btrfs_shared_data_ref *ref2 = NULL;
1182         struct extent_buffer *leaf;
1183         u32 num_refs = 0;
1184         int ret = 0;
1185
1186         leaf = path->nodes[0];
1187         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1188
1189         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1190                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1191                                       struct btrfs_extent_data_ref);
1192                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1193         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1194                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1195                                       struct btrfs_shared_data_ref);
1196                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1197 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1198         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1199                 struct btrfs_extent_ref_v0 *ref0;
1200                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1201                                       struct btrfs_extent_ref_v0);
1202                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1203 #endif
1204         } else {
1205                 BUG();
1206         }
1207
1208         BUG_ON(num_refs < refs_to_drop);
1209         num_refs -= refs_to_drop;
1210
1211         if (num_refs == 0) {
1212                 ret = btrfs_del_item(trans, root, path);
1213         } else {
1214                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1215                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1216                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1217                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1218 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1219                 else {
1220                         struct btrfs_extent_ref_v0 *ref0;
1221                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1222                                         struct btrfs_extent_ref_v0);
1223                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1224                 }
1225 #endif
1226                 btrfs_mark_buffer_dirty(leaf);
1227         }
1228         return ret;
1229 }
1230
1231 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1232                                           struct btrfs_path *path,
1233                                           struct btrfs_extent_inline_ref *iref)
1234 {
1235         struct btrfs_key key;
1236         struct extent_buffer *leaf;
1237         struct btrfs_extent_data_ref *ref1;
1238         struct btrfs_shared_data_ref *ref2;
1239         u32 num_refs = 0;
1240
1241         leaf = path->nodes[0];
1242         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1243         if (iref) {
1244                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1245                     BTRFS_EXTENT_DATA_REF_KEY) {
1246                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1247                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1248                 } else {
1249                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1250                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1251                 }
1252         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1253                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1254                                       struct btrfs_extent_data_ref);
1255                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1256         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1257                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1258                                       struct btrfs_shared_data_ref);
1259                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1260 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1261         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1262                 struct btrfs_extent_ref_v0 *ref0;
1263                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1264                                       struct btrfs_extent_ref_v0);
1265                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1266 #endif
1267         } else {
1268                 WARN_ON(1);
1269         }
1270         return num_refs;
1271 }
1272
1273 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1274                                           struct btrfs_root *root,
1275                                           struct btrfs_path *path,
1276                                           u64 bytenr, u64 parent,
1277                                           u64 root_objectid)
1278 {
1279         struct btrfs_key key;
1280         int ret;
1281
1282         key.objectid = bytenr;
1283         if (parent) {
1284                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1285                 key.offset = parent;
1286         } else {
1287                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1288                 key.offset = root_objectid;
1289         }
1290
1291         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1292         if (ret > 0)
1293                 ret = -ENOENT;
1294 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1295         if (ret == -ENOENT && parent) {
1296                 btrfs_release_path(root, path);
1297                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1298                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1299                 if (ret > 0)
1300                         ret = -ENOENT;
1301         }
1302 #endif
1303         return ret;
1304 }
1305
1306 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1307                                           struct btrfs_root *root,
1308                                           struct btrfs_path *path,
1309                                           u64 bytenr, u64 parent,
1310                                           u64 root_objectid)
1311 {
1312         struct btrfs_key key;
1313         int ret;
1314
1315         key.objectid = bytenr;
1316         if (parent) {
1317                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1318                 key.offset = parent;
1319         } else {
1320                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1321                 key.offset = root_objectid;
1322         }
1323
1324         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1325         btrfs_release_path(root, path);
1326         return ret;
1327 }
1328
1329 static inline int extent_ref_type(u64 parent, u64 owner)
1330 {
1331         int type;
1332         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1333                 if (parent > 0)
1334                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1335                 else
1336                         type = BTRFS_TREE_BLOCK_REF_KEY;
1337         } else {
1338                 if (parent > 0)
1339                         type = BTRFS_SHARED_DATA_REF_KEY;
1340                 else
1341                         type = BTRFS_EXTENT_DATA_REF_KEY;
1342         }
1343         return type;
1344 }
1345
1346 static int find_next_key(struct btrfs_path *path, int level,
1347                          struct btrfs_key *key)
1348
1349 {
1350         for (; level < BTRFS_MAX_LEVEL; level++) {
1351                 if (!path->nodes[level])
1352                         break;
1353                 if (path->slots[level] + 1 >=
1354                     btrfs_header_nritems(path->nodes[level]))
1355                         continue;
1356                 if (level == 0)
1357                         btrfs_item_key_to_cpu(path->nodes[level], key,
1358                                               path->slots[level] + 1);
1359                 else
1360                         btrfs_node_key_to_cpu(path->nodes[level], key,
1361                                               path->slots[level] + 1);
1362                 return 0;
1363         }
1364         return 1;
1365 }
1366
1367 /*
1368  * look for inline back ref. if back ref is found, *ref_ret is set
1369  * to the address of inline back ref, and 0 is returned.
1370  *
1371  * if back ref isn't found, *ref_ret is set to the address where it
1372  * should be inserted, and -ENOENT is returned.
1373  *
1374  * if insert is true and there are too many inline back refs, the path
1375  * points to the extent item, and -EAGAIN is returned.
1376  *
1377  * NOTE: inline back refs are ordered in the same way that back ref
1378  *       items in the tree are ordered.
1379  */
1380 static noinline_for_stack
1381 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1382                                  struct btrfs_root *root,
1383                                  struct btrfs_path *path,
1384                                  struct btrfs_extent_inline_ref **ref_ret,
1385                                  u64 bytenr, u64 num_bytes,
1386                                  u64 parent, u64 root_objectid,
1387                                  u64 owner, u64 offset, int insert)
1388 {
1389         struct btrfs_key key;
1390         struct extent_buffer *leaf;
1391         struct btrfs_extent_item *ei;
1392         struct btrfs_extent_inline_ref *iref;
1393         u64 flags;
1394         u64 item_size;
1395         unsigned long ptr;
1396         unsigned long end;
1397         int extra_size;
1398         int type;
1399         int want;
1400         int ret;
1401         int err = 0;
1402
1403         key.objectid = bytenr;
1404         key.type = BTRFS_EXTENT_ITEM_KEY;
1405         key.offset = num_bytes;
1406
1407         want = extent_ref_type(parent, owner);
1408         if (insert) {
1409                 extra_size = btrfs_extent_inline_ref_size(want);
1410                 path->keep_locks = 1;
1411         } else
1412                 extra_size = -1;
1413         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1414         if (ret < 0) {
1415                 err = ret;
1416                 goto out;
1417         }
1418         BUG_ON(ret);
1419
1420         leaf = path->nodes[0];
1421         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1422 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1423         if (item_size < sizeof(*ei)) {
1424                 if (!insert) {
1425                         err = -ENOENT;
1426                         goto out;
1427                 }
1428                 ret = convert_extent_item_v0(trans, root, path, owner,
1429                                              extra_size);
1430                 if (ret < 0) {
1431                         err = ret;
1432                         goto out;
1433                 }
1434                 leaf = path->nodes[0];
1435                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1436         }
1437 #endif
1438         BUG_ON(item_size < sizeof(*ei));
1439
1440         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1441         flags = btrfs_extent_flags(leaf, ei);
1442
1443         ptr = (unsigned long)(ei + 1);
1444         end = (unsigned long)ei + item_size;
1445
1446         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1447                 ptr += sizeof(struct btrfs_tree_block_info);
1448                 BUG_ON(ptr > end);
1449         } else {
1450                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1451         }
1452
1453         err = -ENOENT;
1454         while (1) {
1455                 if (ptr >= end) {
1456                         WARN_ON(ptr > end);
1457                         break;
1458                 }
1459                 iref = (struct btrfs_extent_inline_ref *)ptr;
1460                 type = btrfs_extent_inline_ref_type(leaf, iref);
1461                 if (want < type)
1462                         break;
1463                 if (want > type) {
1464                         ptr += btrfs_extent_inline_ref_size(type);
1465                         continue;
1466                 }
1467
1468                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1469                         struct btrfs_extent_data_ref *dref;
1470                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1471                         if (match_extent_data_ref(leaf, dref, root_objectid,
1472                                                   owner, offset)) {
1473                                 err = 0;
1474                                 break;
1475                         }
1476                         if (hash_extent_data_ref_item(leaf, dref) <
1477                             hash_extent_data_ref(root_objectid, owner, offset))
1478                                 break;
1479                 } else {
1480                         u64 ref_offset;
1481                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1482                         if (parent > 0) {
1483                                 if (parent == ref_offset) {
1484                                         err = 0;
1485                                         break;
1486                                 }
1487                                 if (ref_offset < parent)
1488                                         break;
1489                         } else {
1490                                 if (root_objectid == ref_offset) {
1491                                         err = 0;
1492                                         break;
1493                                 }
1494                                 if (ref_offset < root_objectid)
1495                                         break;
1496                         }
1497                 }
1498                 ptr += btrfs_extent_inline_ref_size(type);
1499         }
1500         if (err == -ENOENT && insert) {
1501                 if (item_size + extra_size >=
1502                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1503                         err = -EAGAIN;
1504                         goto out;
1505                 }
1506                 /*
1507                  * To add new inline back ref, we have to make sure
1508                  * there is no corresponding back ref item.
1509                  * For simplicity, we just do not add new inline back
1510                  * ref if there is any kind of item for this block
1511                  */
1512                 if (find_next_key(path, 0, &key) == 0 &&
1513                     key.objectid == bytenr &&
1514                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1515                         err = -EAGAIN;
1516                         goto out;
1517                 }
1518         }
1519         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1520 out:
1521         if (insert) {
1522                 path->keep_locks = 0;
1523                 btrfs_unlock_up_safe(path, 1);
1524         }
1525         return err;
1526 }
1527
1528 /*
1529  * helper to add new inline back ref
1530  */
1531 static noinline_for_stack
1532 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1533                                 struct btrfs_root *root,
1534                                 struct btrfs_path *path,
1535                                 struct btrfs_extent_inline_ref *iref,
1536                                 u64 parent, u64 root_objectid,
1537                                 u64 owner, u64 offset, int refs_to_add,
1538                                 struct btrfs_delayed_extent_op *extent_op)
1539 {
1540         struct extent_buffer *leaf;
1541         struct btrfs_extent_item *ei;
1542         unsigned long ptr;
1543         unsigned long end;
1544         unsigned long item_offset;
1545         u64 refs;
1546         int size;
1547         int type;
1548         int ret;
1549
1550         leaf = path->nodes[0];
1551         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1552         item_offset = (unsigned long)iref - (unsigned long)ei;
1553
1554         type = extent_ref_type(parent, owner);
1555         size = btrfs_extent_inline_ref_size(type);
1556
1557         ret = btrfs_extend_item(trans, root, path, size);
1558         BUG_ON(ret);
1559
1560         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1561         refs = btrfs_extent_refs(leaf, ei);
1562         refs += refs_to_add;
1563         btrfs_set_extent_refs(leaf, ei, refs);
1564         if (extent_op)
1565                 __run_delayed_extent_op(extent_op, leaf, ei);
1566
1567         ptr = (unsigned long)ei + item_offset;
1568         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1569         if (ptr < end - size)
1570                 memmove_extent_buffer(leaf, ptr + size, ptr,
1571                                       end - size - ptr);
1572
1573         iref = (struct btrfs_extent_inline_ref *)ptr;
1574         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1575         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1576                 struct btrfs_extent_data_ref *dref;
1577                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1578                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1579                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1580                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1581                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1582         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1583                 struct btrfs_shared_data_ref *sref;
1584                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1585                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1586                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1587         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1588                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1589         } else {
1590                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1591         }
1592         btrfs_mark_buffer_dirty(leaf);
1593         return 0;
1594 }
1595
1596 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1597                                  struct btrfs_root *root,
1598                                  struct btrfs_path *path,
1599                                  struct btrfs_extent_inline_ref **ref_ret,
1600                                  u64 bytenr, u64 num_bytes, u64 parent,
1601                                  u64 root_objectid, u64 owner, u64 offset)
1602 {
1603         int ret;
1604
1605         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1606                                            bytenr, num_bytes, parent,
1607                                            root_objectid, owner, offset, 0);
1608         if (ret != -ENOENT)
1609                 return ret;
1610
1611         btrfs_release_path(root, path);
1612         *ref_ret = NULL;
1613
1614         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1615                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1616                                             root_objectid);
1617         } else {
1618                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1619                                              root_objectid, owner, offset);
1620         }
1621         return ret;
1622 }
1623
1624 /*
1625  * helper to update/remove inline back ref
1626  */
1627 static noinline_for_stack
1628 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1629                                  struct btrfs_root *root,
1630                                  struct btrfs_path *path,
1631                                  struct btrfs_extent_inline_ref *iref,
1632                                  int refs_to_mod,
1633                                  struct btrfs_delayed_extent_op *extent_op)
1634 {
1635         struct extent_buffer *leaf;
1636         struct btrfs_extent_item *ei;
1637         struct btrfs_extent_data_ref *dref = NULL;
1638         struct btrfs_shared_data_ref *sref = NULL;
1639         unsigned long ptr;
1640         unsigned long end;
1641         u32 item_size;
1642         int size;
1643         int type;
1644         int ret;
1645         u64 refs;
1646
1647         leaf = path->nodes[0];
1648         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1649         refs = btrfs_extent_refs(leaf, ei);
1650         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1651         refs += refs_to_mod;
1652         btrfs_set_extent_refs(leaf, ei, refs);
1653         if (extent_op)
1654                 __run_delayed_extent_op(extent_op, leaf, ei);
1655
1656         type = btrfs_extent_inline_ref_type(leaf, iref);
1657
1658         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1659                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1660                 refs = btrfs_extent_data_ref_count(leaf, dref);
1661         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1662                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1663                 refs = btrfs_shared_data_ref_count(leaf, sref);
1664         } else {
1665                 refs = 1;
1666                 BUG_ON(refs_to_mod != -1);
1667         }
1668
1669         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1670         refs += refs_to_mod;
1671
1672         if (refs > 0) {
1673                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1674                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1675                 else
1676                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1677         } else {
1678                 size =  btrfs_extent_inline_ref_size(type);
1679                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1680                 ptr = (unsigned long)iref;
1681                 end = (unsigned long)ei + item_size;
1682                 if (ptr + size < end)
1683                         memmove_extent_buffer(leaf, ptr, ptr + size,
1684                                               end - ptr - size);
1685                 item_size -= size;
1686                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1687                 BUG_ON(ret);
1688         }
1689         btrfs_mark_buffer_dirty(leaf);
1690         return 0;
1691 }
1692
1693 static noinline_for_stack
1694 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1695                                  struct btrfs_root *root,
1696                                  struct btrfs_path *path,
1697                                  u64 bytenr, u64 num_bytes, u64 parent,
1698                                  u64 root_objectid, u64 owner,
1699                                  u64 offset, int refs_to_add,
1700                                  struct btrfs_delayed_extent_op *extent_op)
1701 {
1702         struct btrfs_extent_inline_ref *iref;
1703         int ret;
1704
1705         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1706                                            bytenr, num_bytes, parent,
1707                                            root_objectid, owner, offset, 1);
1708         if (ret == 0) {
1709                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1710                 ret = update_inline_extent_backref(trans, root, path, iref,
1711                                                    refs_to_add, extent_op);
1712         } else if (ret == -ENOENT) {
1713                 ret = setup_inline_extent_backref(trans, root, path, iref,
1714                                                   parent, root_objectid,
1715                                                   owner, offset, refs_to_add,
1716                                                   extent_op);
1717         }
1718         return ret;
1719 }
1720
1721 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1722                                  struct btrfs_root *root,
1723                                  struct btrfs_path *path,
1724                                  u64 bytenr, u64 parent, u64 root_objectid,
1725                                  u64 owner, u64 offset, int refs_to_add)
1726 {
1727         int ret;
1728         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1729                 BUG_ON(refs_to_add != 1);
1730                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1731                                             parent, root_objectid);
1732         } else {
1733                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1734                                              parent, root_objectid,
1735                                              owner, offset, refs_to_add);
1736         }
1737         return ret;
1738 }
1739
1740 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1741                                  struct btrfs_root *root,
1742                                  struct btrfs_path *path,
1743                                  struct btrfs_extent_inline_ref *iref,
1744                                  int refs_to_drop, int is_data)
1745 {
1746         int ret;
1747
1748         BUG_ON(!is_data && refs_to_drop != 1);
1749         if (iref) {
1750                 ret = update_inline_extent_backref(trans, root, path, iref,
1751                                                    -refs_to_drop, NULL);
1752         } else if (is_data) {
1753                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1754         } else {
1755                 ret = btrfs_del_item(trans, root, path);
1756         }
1757         return ret;
1758 }
1759
1760 static int btrfs_issue_discard(struct block_device *bdev,
1761                                 u64 start, u64 len)
1762 {
1763         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1764 }
1765
1766 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1767                                 u64 num_bytes, u64 *actual_bytes)
1768 {
1769         int ret;
1770         u64 discarded_bytes = 0;
1771         struct btrfs_multi_bio *multi = NULL;
1772
1773
1774         /* Tell the block device(s) that the sectors can be discarded */
1775         ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1776                               bytenr, &num_bytes, &multi, 0);
1777         if (!ret) {
1778                 struct btrfs_bio_stripe *stripe = multi->stripes;
1779                 int i;
1780
1781
1782                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1783                         ret = btrfs_issue_discard(stripe->dev->bdev,
1784                                                   stripe->physical,
1785                                                   stripe->length);
1786                         if (!ret)
1787                                 discarded_bytes += stripe->length;
1788                         else if (ret != -EOPNOTSUPP)
1789                                 break;
1790                 }
1791                 kfree(multi);
1792         }
1793         if (discarded_bytes && ret == -EOPNOTSUPP)
1794                 ret = 0;
1795
1796         if (actual_bytes)
1797                 *actual_bytes = discarded_bytes;
1798
1799
1800         return ret;
1801 }
1802
1803 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1804                          struct btrfs_root *root,
1805                          u64 bytenr, u64 num_bytes, u64 parent,
1806                          u64 root_objectid, u64 owner, u64 offset)
1807 {
1808         int ret;
1809         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1810                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1811
1812         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1813                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1814                                         parent, root_objectid, (int)owner,
1815                                         BTRFS_ADD_DELAYED_REF, NULL);
1816         } else {
1817                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1818                                         parent, root_objectid, owner, offset,
1819                                         BTRFS_ADD_DELAYED_REF, NULL);
1820         }
1821         return ret;
1822 }
1823
1824 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1825                                   struct btrfs_root *root,
1826                                   u64 bytenr, u64 num_bytes,
1827                                   u64 parent, u64 root_objectid,
1828                                   u64 owner, u64 offset, int refs_to_add,
1829                                   struct btrfs_delayed_extent_op *extent_op)
1830 {
1831         struct btrfs_path *path;
1832         struct extent_buffer *leaf;
1833         struct btrfs_extent_item *item;
1834         u64 refs;
1835         int ret;
1836         int err = 0;
1837
1838         path = btrfs_alloc_path();
1839         if (!path)
1840                 return -ENOMEM;
1841
1842         path->reada = 1;
1843         path->leave_spinning = 1;
1844         /* this will setup the path even if it fails to insert the back ref */
1845         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1846                                            path, bytenr, num_bytes, parent,
1847                                            root_objectid, owner, offset,
1848                                            refs_to_add, extent_op);
1849         if (ret == 0)
1850                 goto out;
1851
1852         if (ret != -EAGAIN) {
1853                 err = ret;
1854                 goto out;
1855         }
1856
1857         leaf = path->nodes[0];
1858         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1859         refs = btrfs_extent_refs(leaf, item);
1860         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1861         if (extent_op)
1862                 __run_delayed_extent_op(extent_op, leaf, item);
1863
1864         btrfs_mark_buffer_dirty(leaf);
1865         btrfs_release_path(root->fs_info->extent_root, path);
1866
1867         path->reada = 1;
1868         path->leave_spinning = 1;
1869
1870         /* now insert the actual backref */
1871         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1872                                     path, bytenr, parent, root_objectid,
1873                                     owner, offset, refs_to_add);
1874         BUG_ON(ret);
1875 out:
1876         btrfs_free_path(path);
1877         return err;
1878 }
1879
1880 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1881                                 struct btrfs_root *root,
1882                                 struct btrfs_delayed_ref_node *node,
1883                                 struct btrfs_delayed_extent_op *extent_op,
1884                                 int insert_reserved)
1885 {
1886         int ret = 0;
1887         struct btrfs_delayed_data_ref *ref;
1888         struct btrfs_key ins;
1889         u64 parent = 0;
1890         u64 ref_root = 0;
1891         u64 flags = 0;
1892
1893         ins.objectid = node->bytenr;
1894         ins.offset = node->num_bytes;
1895         ins.type = BTRFS_EXTENT_ITEM_KEY;
1896
1897         ref = btrfs_delayed_node_to_data_ref(node);
1898         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1899                 parent = ref->parent;
1900         else
1901                 ref_root = ref->root;
1902
1903         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1904                 if (extent_op) {
1905                         BUG_ON(extent_op->update_key);
1906                         flags |= extent_op->flags_to_set;
1907                 }
1908                 ret = alloc_reserved_file_extent(trans, root,
1909                                                  parent, ref_root, flags,
1910                                                  ref->objectid, ref->offset,
1911                                                  &ins, node->ref_mod);
1912         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1913                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1914                                              node->num_bytes, parent,
1915                                              ref_root, ref->objectid,
1916                                              ref->offset, node->ref_mod,
1917                                              extent_op);
1918         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1919                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1920                                           node->num_bytes, parent,
1921                                           ref_root, ref->objectid,
1922                                           ref->offset, node->ref_mod,
1923                                           extent_op);
1924         } else {
1925                 BUG();
1926         }
1927         return ret;
1928 }
1929
1930 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1931                                     struct extent_buffer *leaf,
1932                                     struct btrfs_extent_item *ei)
1933 {
1934         u64 flags = btrfs_extent_flags(leaf, ei);
1935         if (extent_op->update_flags) {
1936                 flags |= extent_op->flags_to_set;
1937                 btrfs_set_extent_flags(leaf, ei, flags);
1938         }
1939
1940         if (extent_op->update_key) {
1941                 struct btrfs_tree_block_info *bi;
1942                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1943                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1944                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1945         }
1946 }
1947
1948 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1949                                  struct btrfs_root *root,
1950                                  struct btrfs_delayed_ref_node *node,
1951                                  struct btrfs_delayed_extent_op *extent_op)
1952 {
1953         struct btrfs_key key;
1954         struct btrfs_path *path;
1955         struct btrfs_extent_item *ei;
1956         struct extent_buffer *leaf;
1957         u32 item_size;
1958         int ret;
1959         int err = 0;
1960
1961         path = btrfs_alloc_path();
1962         if (!path)
1963                 return -ENOMEM;
1964
1965         key.objectid = node->bytenr;
1966         key.type = BTRFS_EXTENT_ITEM_KEY;
1967         key.offset = node->num_bytes;
1968
1969         path->reada = 1;
1970         path->leave_spinning = 1;
1971         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1972                                 path, 0, 1);
1973         if (ret < 0) {
1974                 err = ret;
1975                 goto out;
1976         }
1977         if (ret > 0) {
1978                 err = -EIO;
1979                 goto out;
1980         }
1981
1982         leaf = path->nodes[0];
1983         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1984 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1985         if (item_size < sizeof(*ei)) {
1986                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1987                                              path, (u64)-1, 0);
1988                 if (ret < 0) {
1989                         err = ret;
1990                         goto out;
1991                 }
1992                 leaf = path->nodes[0];
1993                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1994         }
1995 #endif
1996         BUG_ON(item_size < sizeof(*ei));
1997         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1998         __run_delayed_extent_op(extent_op, leaf, ei);
1999
2000         btrfs_mark_buffer_dirty(leaf);
2001 out:
2002         btrfs_free_path(path);
2003         return err;
2004 }
2005
2006 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2007                                 struct btrfs_root *root,
2008                                 struct btrfs_delayed_ref_node *node,
2009                                 struct btrfs_delayed_extent_op *extent_op,
2010                                 int insert_reserved)
2011 {
2012         int ret = 0;
2013         struct btrfs_delayed_tree_ref *ref;
2014         struct btrfs_key ins;
2015         u64 parent = 0;
2016         u64 ref_root = 0;
2017
2018         ins.objectid = node->bytenr;
2019         ins.offset = node->num_bytes;
2020         ins.type = BTRFS_EXTENT_ITEM_KEY;
2021
2022         ref = btrfs_delayed_node_to_tree_ref(node);
2023         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2024                 parent = ref->parent;
2025         else
2026                 ref_root = ref->root;
2027
2028         BUG_ON(node->ref_mod != 1);
2029         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2030                 BUG_ON(!extent_op || !extent_op->update_flags ||
2031                        !extent_op->update_key);
2032                 ret = alloc_reserved_tree_block(trans, root,
2033                                                 parent, ref_root,
2034                                                 extent_op->flags_to_set,
2035                                                 &extent_op->key,
2036                                                 ref->level, &ins);
2037         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2038                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2039                                              node->num_bytes, parent, ref_root,
2040                                              ref->level, 0, 1, extent_op);
2041         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2042                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2043                                           node->num_bytes, parent, ref_root,
2044                                           ref->level, 0, 1, extent_op);
2045         } else {
2046                 BUG();
2047         }
2048         return ret;
2049 }
2050
2051 /* helper function to actually process a single delayed ref entry */
2052 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2053                                struct btrfs_root *root,
2054                                struct btrfs_delayed_ref_node *node,
2055                                struct btrfs_delayed_extent_op *extent_op,
2056                                int insert_reserved)
2057 {
2058         int ret;
2059         if (btrfs_delayed_ref_is_head(node)) {
2060                 struct btrfs_delayed_ref_head *head;
2061                 /*
2062                  * we've hit the end of the chain and we were supposed
2063                  * to insert this extent into the tree.  But, it got
2064                  * deleted before we ever needed to insert it, so all
2065                  * we have to do is clean up the accounting
2066                  */
2067                 BUG_ON(extent_op);
2068                 head = btrfs_delayed_node_to_head(node);
2069                 if (insert_reserved) {
2070                         btrfs_pin_extent(root, node->bytenr,
2071                                          node->num_bytes, 1);
2072                         if (head->is_data) {
2073                                 ret = btrfs_del_csums(trans, root,
2074                                                       node->bytenr,
2075                                                       node->num_bytes);
2076                                 BUG_ON(ret);
2077                         }
2078                 }
2079                 mutex_unlock(&head->mutex);
2080                 return 0;
2081         }
2082
2083         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2084             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2085                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2086                                            insert_reserved);
2087         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2088                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2089                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2090                                            insert_reserved);
2091         else
2092                 BUG();
2093         return ret;
2094 }
2095
2096 static noinline struct btrfs_delayed_ref_node *
2097 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2098 {
2099         struct rb_node *node;
2100         struct btrfs_delayed_ref_node *ref;
2101         int action = BTRFS_ADD_DELAYED_REF;
2102 again:
2103         /*
2104          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2105          * this prevents ref count from going down to zero when
2106          * there still are pending delayed ref.
2107          */
2108         node = rb_prev(&head->node.rb_node);
2109         while (1) {
2110                 if (!node)
2111                         break;
2112                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2113                                 rb_node);
2114                 if (ref->bytenr != head->node.bytenr)
2115                         break;
2116                 if (ref->action == action)
2117                         return ref;
2118                 node = rb_prev(node);
2119         }
2120         if (action == BTRFS_ADD_DELAYED_REF) {
2121                 action = BTRFS_DROP_DELAYED_REF;
2122                 goto again;
2123         }
2124         return NULL;
2125 }
2126
2127 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2128                                        struct btrfs_root *root,
2129                                        struct list_head *cluster)
2130 {
2131         struct btrfs_delayed_ref_root *delayed_refs;
2132         struct btrfs_delayed_ref_node *ref;
2133         struct btrfs_delayed_ref_head *locked_ref = NULL;
2134         struct btrfs_delayed_extent_op *extent_op;
2135         int ret;
2136         int count = 0;
2137         int must_insert_reserved = 0;
2138
2139         delayed_refs = &trans->transaction->delayed_refs;
2140         while (1) {
2141                 if (!locked_ref) {
2142                         /* pick a new head ref from the cluster list */
2143                         if (list_empty(cluster))
2144                                 break;
2145
2146                         locked_ref = list_entry(cluster->next,
2147                                      struct btrfs_delayed_ref_head, cluster);
2148
2149                         /* grab the lock that says we are going to process
2150                          * all the refs for this head */
2151                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2152
2153                         /*
2154                          * we may have dropped the spin lock to get the head
2155                          * mutex lock, and that might have given someone else
2156                          * time to free the head.  If that's true, it has been
2157                          * removed from our list and we can move on.
2158                          */
2159                         if (ret == -EAGAIN) {
2160                                 locked_ref = NULL;
2161                                 count++;
2162                                 continue;
2163                         }
2164                 }
2165
2166                 /*
2167                  * record the must insert reserved flag before we
2168                  * drop the spin lock.
2169                  */
2170                 must_insert_reserved = locked_ref->must_insert_reserved;
2171                 locked_ref->must_insert_reserved = 0;
2172
2173                 extent_op = locked_ref->extent_op;
2174                 locked_ref->extent_op = NULL;
2175
2176                 /*
2177                  * locked_ref is the head node, so we have to go one
2178                  * node back for any delayed ref updates
2179                  */
2180                 ref = select_delayed_ref(locked_ref);
2181                 if (!ref) {
2182                         /* All delayed refs have been processed, Go ahead
2183                          * and send the head node to run_one_delayed_ref,
2184                          * so that any accounting fixes can happen
2185                          */
2186                         ref = &locked_ref->node;
2187
2188                         if (extent_op && must_insert_reserved) {
2189                                 kfree(extent_op);
2190                                 extent_op = NULL;
2191                         }
2192
2193                         if (extent_op) {
2194                                 spin_unlock(&delayed_refs->lock);
2195
2196                                 ret = run_delayed_extent_op(trans, root,
2197                                                             ref, extent_op);
2198                                 BUG_ON(ret);
2199                                 kfree(extent_op);
2200
2201                                 cond_resched();
2202                                 spin_lock(&delayed_refs->lock);
2203                                 continue;
2204                         }
2205
2206                         list_del_init(&locked_ref->cluster);
2207                         locked_ref = NULL;
2208                 }
2209
2210                 ref->in_tree = 0;
2211                 rb_erase(&ref->rb_node, &delayed_refs->root);
2212                 delayed_refs->num_entries--;
2213
2214                 spin_unlock(&delayed_refs->lock);
2215
2216                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2217                                           must_insert_reserved);
2218                 BUG_ON(ret);
2219
2220                 btrfs_put_delayed_ref(ref);
2221                 kfree(extent_op);
2222                 count++;
2223
2224                 cond_resched();
2225                 spin_lock(&delayed_refs->lock);
2226         }
2227         return count;
2228 }
2229
2230 /*
2231  * this starts processing the delayed reference count updates and
2232  * extent insertions we have queued up so far.  count can be
2233  * 0, which means to process everything in the tree at the start
2234  * of the run (but not newly added entries), or it can be some target
2235  * number you'd like to process.
2236  */
2237 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2238                            struct btrfs_root *root, unsigned long count)
2239 {
2240         struct rb_node *node;
2241         struct btrfs_delayed_ref_root *delayed_refs;
2242         struct btrfs_delayed_ref_node *ref;
2243         struct list_head cluster;
2244         int ret;
2245         int run_all = count == (unsigned long)-1;
2246         int run_most = 0;
2247
2248         if (root == root->fs_info->extent_root)
2249                 root = root->fs_info->tree_root;
2250
2251         delayed_refs = &trans->transaction->delayed_refs;
2252         INIT_LIST_HEAD(&cluster);
2253 again:
2254         spin_lock(&delayed_refs->lock);
2255         if (count == 0) {
2256                 count = delayed_refs->num_entries * 2;
2257                 run_most = 1;
2258         }
2259         while (1) {
2260                 if (!(run_all || run_most) &&
2261                     delayed_refs->num_heads_ready < 64)
2262                         break;
2263
2264                 /*
2265                  * go find something we can process in the rbtree.  We start at
2266                  * the beginning of the tree, and then build a cluster
2267                  * of refs to process starting at the first one we are able to
2268                  * lock
2269                  */
2270                 ret = btrfs_find_ref_cluster(trans, &cluster,
2271                                              delayed_refs->run_delayed_start);
2272                 if (ret)
2273                         break;
2274
2275                 ret = run_clustered_refs(trans, root, &cluster);
2276                 BUG_ON(ret < 0);
2277
2278                 count -= min_t(unsigned long, ret, count);
2279
2280                 if (count == 0)
2281                         break;
2282         }
2283
2284         if (run_all) {
2285                 node = rb_first(&delayed_refs->root);
2286                 if (!node)
2287                         goto out;
2288                 count = (unsigned long)-1;
2289
2290                 while (node) {
2291                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2292                                        rb_node);
2293                         if (btrfs_delayed_ref_is_head(ref)) {
2294                                 struct btrfs_delayed_ref_head *head;
2295
2296                                 head = btrfs_delayed_node_to_head(ref);
2297                                 atomic_inc(&ref->refs);
2298
2299                                 spin_unlock(&delayed_refs->lock);
2300                                 mutex_lock(&head->mutex);
2301                                 mutex_unlock(&head->mutex);
2302
2303                                 btrfs_put_delayed_ref(ref);
2304                                 cond_resched();
2305                                 goto again;
2306                         }
2307                         node = rb_next(node);
2308                 }
2309                 spin_unlock(&delayed_refs->lock);
2310                 schedule_timeout(1);
2311                 goto again;
2312         }
2313 out:
2314         spin_unlock(&delayed_refs->lock);
2315         return 0;
2316 }
2317
2318 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2319                                 struct btrfs_root *root,
2320                                 u64 bytenr, u64 num_bytes, u64 flags,
2321                                 int is_data)
2322 {
2323         struct btrfs_delayed_extent_op *extent_op;
2324         int ret;
2325
2326         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2327         if (!extent_op)
2328                 return -ENOMEM;
2329
2330         extent_op->flags_to_set = flags;
2331         extent_op->update_flags = 1;
2332         extent_op->update_key = 0;
2333         extent_op->is_data = is_data ? 1 : 0;
2334
2335         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2336         if (ret)
2337                 kfree(extent_op);
2338         return ret;
2339 }
2340
2341 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2342                                       struct btrfs_root *root,
2343                                       struct btrfs_path *path,
2344                                       u64 objectid, u64 offset, u64 bytenr)
2345 {
2346         struct btrfs_delayed_ref_head *head;
2347         struct btrfs_delayed_ref_node *ref;
2348         struct btrfs_delayed_data_ref *data_ref;
2349         struct btrfs_delayed_ref_root *delayed_refs;
2350         struct rb_node *node;
2351         int ret = 0;
2352
2353         ret = -ENOENT;
2354         delayed_refs = &trans->transaction->delayed_refs;
2355         spin_lock(&delayed_refs->lock);
2356         head = btrfs_find_delayed_ref_head(trans, bytenr);
2357         if (!head)
2358                 goto out;
2359
2360         if (!mutex_trylock(&head->mutex)) {
2361                 atomic_inc(&head->node.refs);
2362                 spin_unlock(&delayed_refs->lock);
2363
2364                 btrfs_release_path(root->fs_info->extent_root, path);
2365
2366                 mutex_lock(&head->mutex);
2367                 mutex_unlock(&head->mutex);
2368                 btrfs_put_delayed_ref(&head->node);
2369                 return -EAGAIN;
2370         }
2371
2372         node = rb_prev(&head->node.rb_node);
2373         if (!node)
2374                 goto out_unlock;
2375
2376         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2377
2378         if (ref->bytenr != bytenr)
2379                 goto out_unlock;
2380
2381         ret = 1;
2382         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2383                 goto out_unlock;
2384
2385         data_ref = btrfs_delayed_node_to_data_ref(ref);
2386
2387         node = rb_prev(node);
2388         if (node) {
2389                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2390                 if (ref->bytenr == bytenr)
2391                         goto out_unlock;
2392         }
2393
2394         if (data_ref->root != root->root_key.objectid ||
2395             data_ref->objectid != objectid || data_ref->offset != offset)
2396                 goto out_unlock;
2397
2398         ret = 0;
2399 out_unlock:
2400         mutex_unlock(&head->mutex);
2401 out:
2402         spin_unlock(&delayed_refs->lock);
2403         return ret;
2404 }
2405
2406 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2407                                         struct btrfs_root *root,
2408                                         struct btrfs_path *path,
2409                                         u64 objectid, u64 offset, u64 bytenr)
2410 {
2411         struct btrfs_root *extent_root = root->fs_info->extent_root;
2412         struct extent_buffer *leaf;
2413         struct btrfs_extent_data_ref *ref;
2414         struct btrfs_extent_inline_ref *iref;
2415         struct btrfs_extent_item *ei;
2416         struct btrfs_key key;
2417         u32 item_size;
2418         int ret;
2419
2420         key.objectid = bytenr;
2421         key.offset = (u64)-1;
2422         key.type = BTRFS_EXTENT_ITEM_KEY;
2423
2424         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2425         if (ret < 0)
2426                 goto out;
2427         BUG_ON(ret == 0);
2428
2429         ret = -ENOENT;
2430         if (path->slots[0] == 0)
2431                 goto out;
2432
2433         path->slots[0]--;
2434         leaf = path->nodes[0];
2435         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2436
2437         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2438                 goto out;
2439
2440         ret = 1;
2441         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2442 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2443         if (item_size < sizeof(*ei)) {
2444                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2445                 goto out;
2446         }
2447 #endif
2448         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2449
2450         if (item_size != sizeof(*ei) +
2451             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2452                 goto out;
2453
2454         if (btrfs_extent_generation(leaf, ei) <=
2455             btrfs_root_last_snapshot(&root->root_item))
2456                 goto out;
2457
2458         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2459         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2460             BTRFS_EXTENT_DATA_REF_KEY)
2461                 goto out;
2462
2463         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2464         if (btrfs_extent_refs(leaf, ei) !=
2465             btrfs_extent_data_ref_count(leaf, ref) ||
2466             btrfs_extent_data_ref_root(leaf, ref) !=
2467             root->root_key.objectid ||
2468             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2469             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2470                 goto out;
2471
2472         ret = 0;
2473 out:
2474         return ret;
2475 }
2476
2477 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2478                           struct btrfs_root *root,
2479                           u64 objectid, u64 offset, u64 bytenr)
2480 {
2481         struct btrfs_path *path;
2482         int ret;
2483         int ret2;
2484
2485         path = btrfs_alloc_path();
2486         if (!path)
2487                 return -ENOENT;
2488
2489         do {
2490                 ret = check_committed_ref(trans, root, path, objectid,
2491                                           offset, bytenr);
2492                 if (ret && ret != -ENOENT)
2493                         goto out;
2494
2495                 ret2 = check_delayed_ref(trans, root, path, objectid,
2496                                          offset, bytenr);
2497         } while (ret2 == -EAGAIN);
2498
2499         if (ret2 && ret2 != -ENOENT) {
2500                 ret = ret2;
2501                 goto out;
2502         }
2503
2504         if (ret != -ENOENT || ret2 != -ENOENT)
2505                 ret = 0;
2506 out:
2507         btrfs_free_path(path);
2508         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2509                 WARN_ON(ret > 0);
2510         return ret;
2511 }
2512
2513 #if 0
2514 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2515                     struct extent_buffer *buf, u32 nr_extents)
2516 {
2517         struct btrfs_key key;
2518         struct btrfs_file_extent_item *fi;
2519         u64 root_gen;
2520         u32 nritems;
2521         int i;
2522         int level;
2523         int ret = 0;
2524         int shared = 0;
2525
2526         if (!root->ref_cows)
2527                 return 0;
2528
2529         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2530                 shared = 0;
2531                 root_gen = root->root_key.offset;
2532         } else {
2533                 shared = 1;
2534                 root_gen = trans->transid - 1;
2535         }
2536
2537         level = btrfs_header_level(buf);
2538         nritems = btrfs_header_nritems(buf);
2539
2540         if (level == 0) {
2541                 struct btrfs_leaf_ref *ref;
2542                 struct btrfs_extent_info *info;
2543
2544                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2545                 if (!ref) {
2546                         ret = -ENOMEM;
2547                         goto out;
2548                 }
2549
2550                 ref->root_gen = root_gen;
2551                 ref->bytenr = buf->start;
2552                 ref->owner = btrfs_header_owner(buf);
2553                 ref->generation = btrfs_header_generation(buf);
2554                 ref->nritems = nr_extents;
2555                 info = ref->extents;
2556
2557                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2558                         u64 disk_bytenr;
2559                         btrfs_item_key_to_cpu(buf, &key, i);
2560                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2561                                 continue;
2562                         fi = btrfs_item_ptr(buf, i,
2563                                             struct btrfs_file_extent_item);
2564                         if (btrfs_file_extent_type(buf, fi) ==
2565                             BTRFS_FILE_EXTENT_INLINE)
2566                                 continue;
2567                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2568                         if (disk_bytenr == 0)
2569                                 continue;
2570
2571                         info->bytenr = disk_bytenr;
2572                         info->num_bytes =
2573                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2574                         info->objectid = key.objectid;
2575                         info->offset = key.offset;
2576                         info++;
2577                 }
2578
2579                 ret = btrfs_add_leaf_ref(root, ref, shared);
2580                 if (ret == -EEXIST && shared) {
2581                         struct btrfs_leaf_ref *old;
2582                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2583                         BUG_ON(!old);
2584                         btrfs_remove_leaf_ref(root, old);
2585                         btrfs_free_leaf_ref(root, old);
2586                         ret = btrfs_add_leaf_ref(root, ref, shared);
2587                 }
2588                 WARN_ON(ret);
2589                 btrfs_free_leaf_ref(root, ref);
2590         }
2591 out:
2592         return ret;
2593 }
2594
2595 /* when a block goes through cow, we update the reference counts of
2596  * everything that block points to.  The internal pointers of the block
2597  * can be in just about any order, and it is likely to have clusters of
2598  * things that are close together and clusters of things that are not.
2599  *
2600  * To help reduce the seeks that come with updating all of these reference
2601  * counts, sort them by byte number before actual updates are done.
2602  *
2603  * struct refsort is used to match byte number to slot in the btree block.
2604  * we sort based on the byte number and then use the slot to actually
2605  * find the item.
2606  *
2607  * struct refsort is smaller than strcut btrfs_item and smaller than
2608  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2609  * for a btree block, there's no way for a kmalloc of refsorts for a
2610  * single node to be bigger than a page.
2611  */
2612 struct refsort {
2613         u64 bytenr;
2614         u32 slot;
2615 };
2616
2617 /*
2618  * for passing into sort()
2619  */
2620 static int refsort_cmp(const void *a_void, const void *b_void)
2621 {
2622         const struct refsort *a = a_void;
2623         const struct refsort *b = b_void;
2624
2625         if (a->bytenr < b->bytenr)
2626                 return -1;
2627         if (a->bytenr > b->bytenr)
2628                 return 1;
2629         return 0;
2630 }
2631 #endif
2632
2633 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2634                            struct btrfs_root *root,
2635                            struct extent_buffer *buf,
2636                            int full_backref, int inc)
2637 {
2638         u64 bytenr;
2639         u64 num_bytes;
2640         u64 parent;
2641         u64 ref_root;
2642         u32 nritems;
2643         struct btrfs_key key;
2644         struct btrfs_file_extent_item *fi;
2645         int i;
2646         int level;
2647         int ret = 0;
2648         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2649                             u64, u64, u64, u64, u64, u64);
2650
2651         ref_root = btrfs_header_owner(buf);
2652         nritems = btrfs_header_nritems(buf);
2653         level = btrfs_header_level(buf);
2654
2655         if (!root->ref_cows && level == 0)
2656                 return 0;
2657
2658         if (inc)
2659                 process_func = btrfs_inc_extent_ref;
2660         else
2661                 process_func = btrfs_free_extent;
2662
2663         if (full_backref)
2664                 parent = buf->start;
2665         else
2666                 parent = 0;
2667
2668         for (i = 0; i < nritems; i++) {
2669                 if (level == 0) {
2670                         btrfs_item_key_to_cpu(buf, &key, i);
2671                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2672                                 continue;
2673                         fi = btrfs_item_ptr(buf, i,
2674                                             struct btrfs_file_extent_item);
2675                         if (btrfs_file_extent_type(buf, fi) ==
2676                             BTRFS_FILE_EXTENT_INLINE)
2677                                 continue;
2678                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2679                         if (bytenr == 0)
2680                                 continue;
2681
2682                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2683                         key.offset -= btrfs_file_extent_offset(buf, fi);
2684                         ret = process_func(trans, root, bytenr, num_bytes,
2685                                            parent, ref_root, key.objectid,
2686                                            key.offset);
2687                         if (ret)
2688                                 goto fail;
2689                 } else {
2690                         bytenr = btrfs_node_blockptr(buf, i);
2691                         num_bytes = btrfs_level_size(root, level - 1);
2692                         ret = process_func(trans, root, bytenr, num_bytes,
2693                                            parent, ref_root, level - 1, 0);
2694                         if (ret)
2695                                 goto fail;
2696                 }
2697         }
2698         return 0;
2699 fail:
2700         BUG();
2701         return ret;
2702 }
2703
2704 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2705                   struct extent_buffer *buf, int full_backref)
2706 {
2707         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2708 }
2709
2710 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2711                   struct extent_buffer *buf, int full_backref)
2712 {
2713         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2714 }
2715
2716 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2717                                  struct btrfs_root *root,
2718                                  struct btrfs_path *path,
2719                                  struct btrfs_block_group_cache *cache)
2720 {
2721         int ret;
2722         struct btrfs_root *extent_root = root->fs_info->extent_root;
2723         unsigned long bi;
2724         struct extent_buffer *leaf;
2725
2726         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2727         if (ret < 0)
2728                 goto fail;
2729         BUG_ON(ret);
2730
2731         leaf = path->nodes[0];
2732         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2733         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2734         btrfs_mark_buffer_dirty(leaf);
2735         btrfs_release_path(extent_root, path);
2736 fail:
2737         if (ret)
2738                 return ret;
2739         return 0;
2740
2741 }
2742
2743 static struct btrfs_block_group_cache *
2744 next_block_group(struct btrfs_root *root,
2745                  struct btrfs_block_group_cache *cache)
2746 {
2747         struct rb_node *node;
2748         spin_lock(&root->fs_info->block_group_cache_lock);
2749         node = rb_next(&cache->cache_node);
2750         btrfs_put_block_group(cache);
2751         if (node) {
2752                 cache = rb_entry(node, struct btrfs_block_group_cache,
2753                                  cache_node);
2754                 btrfs_get_block_group(cache);
2755         } else
2756                 cache = NULL;
2757         spin_unlock(&root->fs_info->block_group_cache_lock);
2758         return cache;
2759 }
2760
2761 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2762                             struct btrfs_trans_handle *trans,
2763                             struct btrfs_path *path)
2764 {
2765         struct btrfs_root *root = block_group->fs_info->tree_root;
2766         struct inode *inode = NULL;
2767         u64 alloc_hint = 0;
2768         int dcs = BTRFS_DC_ERROR;
2769         int num_pages = 0;
2770         int retries = 0;
2771         int ret = 0;
2772
2773         /*
2774          * If this block group is smaller than 100 megs don't bother caching the
2775          * block group.
2776          */
2777         if (block_group->key.offset < (100 * 1024 * 1024)) {
2778                 spin_lock(&block_group->lock);
2779                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2780                 spin_unlock(&block_group->lock);
2781                 return 0;
2782         }
2783
2784 again:
2785         inode = lookup_free_space_inode(root, block_group, path);
2786         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2787                 ret = PTR_ERR(inode);
2788                 btrfs_release_path(root, path);
2789                 goto out;
2790         }
2791
2792         if (IS_ERR(inode)) {
2793                 BUG_ON(retries);
2794                 retries++;
2795
2796                 if (block_group->ro)
2797                         goto out_free;
2798
2799                 ret = create_free_space_inode(root, trans, block_group, path);
2800                 if (ret)
2801                         goto out_free;
2802                 goto again;
2803         }
2804
2805         /*
2806          * We want to set the generation to 0, that way if anything goes wrong
2807          * from here on out we know not to trust this cache when we load up next
2808          * time.
2809          */
2810         BTRFS_I(inode)->generation = 0;
2811         ret = btrfs_update_inode(trans, root, inode);
2812         WARN_ON(ret);
2813
2814         if (i_size_read(inode) > 0) {
2815                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2816                                                       inode);
2817                 if (ret)
2818                         goto out_put;
2819         }
2820
2821         spin_lock(&block_group->lock);
2822         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2823                 /* We're not cached, don't bother trying to write stuff out */
2824                 dcs = BTRFS_DC_WRITTEN;
2825                 spin_unlock(&block_group->lock);
2826                 goto out_put;
2827         }
2828         spin_unlock(&block_group->lock);
2829
2830         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2831         if (!num_pages)
2832                 num_pages = 1;
2833
2834         /*
2835          * Just to make absolutely sure we have enough space, we're going to
2836          * preallocate 12 pages worth of space for each block group.  In
2837          * practice we ought to use at most 8, but we need extra space so we can
2838          * add our header and have a terminator between the extents and the
2839          * bitmaps.
2840          */
2841         num_pages *= 16;
2842         num_pages *= PAGE_CACHE_SIZE;
2843
2844         ret = btrfs_check_data_free_space(inode, num_pages);
2845         if (ret)
2846                 goto out_put;
2847
2848         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2849                                               num_pages, num_pages,
2850                                               &alloc_hint);
2851         if (!ret)
2852                 dcs = BTRFS_DC_SETUP;
2853         btrfs_free_reserved_data_space(inode, num_pages);
2854 out_put:
2855         iput(inode);
2856 out_free:
2857         btrfs_release_path(root, path);
2858 out:
2859         spin_lock(&block_group->lock);
2860         block_group->disk_cache_state = dcs;
2861         spin_unlock(&block_group->lock);
2862
2863         return ret;
2864 }
2865
2866 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2867                                    struct btrfs_root *root)
2868 {
2869         struct btrfs_block_group_cache *cache;
2870         int err = 0;
2871         struct btrfs_path *path;
2872         u64 last = 0;
2873
2874         path = btrfs_alloc_path();
2875         if (!path)
2876                 return -ENOMEM;
2877
2878 again:
2879         while (1) {
2880                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2881                 while (cache) {
2882                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2883                                 break;
2884                         cache = next_block_group(root, cache);
2885                 }
2886                 if (!cache) {
2887                         if (last == 0)
2888                                 break;
2889                         last = 0;
2890                         continue;
2891                 }
2892                 err = cache_save_setup(cache, trans, path);
2893                 last = cache->key.objectid + cache->key.offset;
2894                 btrfs_put_block_group(cache);
2895         }
2896
2897         while (1) {
2898                 if (last == 0) {
2899                         err = btrfs_run_delayed_refs(trans, root,
2900                                                      (unsigned long)-1);
2901                         BUG_ON(err);
2902                 }
2903
2904                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2905                 while (cache) {
2906                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2907                                 btrfs_put_block_group(cache);
2908                                 goto again;
2909                         }
2910
2911                         if (cache->dirty)
2912                                 break;
2913                         cache = next_block_group(root, cache);
2914                 }
2915                 if (!cache) {
2916                         if (last == 0)
2917                                 break;
2918                         last = 0;
2919                         continue;
2920                 }
2921
2922                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2923                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2924                 cache->dirty = 0;
2925                 last = cache->key.objectid + cache->key.offset;
2926
2927                 err = write_one_cache_group(trans, root, path, cache);
2928                 BUG_ON(err);
2929                 btrfs_put_block_group(cache);
2930         }
2931
2932         while (1) {
2933                 /*
2934                  * I don't think this is needed since we're just marking our
2935                  * preallocated extent as written, but just in case it can't
2936                  * hurt.
2937                  */
2938                 if (last == 0) {
2939                         err = btrfs_run_delayed_refs(trans, root,
2940                                                      (unsigned long)-1);
2941                         BUG_ON(err);
2942                 }
2943
2944                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2945                 while (cache) {
2946                         /*
2947                          * Really this shouldn't happen, but it could if we
2948                          * couldn't write the entire preallocated extent and
2949                          * splitting the extent resulted in a new block.
2950                          */
2951                         if (cache->dirty) {
2952                                 btrfs_put_block_group(cache);
2953                                 goto again;
2954                         }
2955                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2956                                 break;
2957                         cache = next_block_group(root, cache);
2958                 }
2959                 if (!cache) {
2960                         if (last == 0)
2961                                 break;
2962                         last = 0;
2963                         continue;
2964                 }
2965
2966                 btrfs_write_out_cache(root, trans, cache, path);
2967
2968                 /*
2969                  * If we didn't have an error then the cache state is still
2970                  * NEED_WRITE, so we can set it to WRITTEN.
2971                  */
2972                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2973                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2974                 last = cache->key.objectid + cache->key.offset;
2975                 btrfs_put_block_group(cache);
2976         }
2977
2978         btrfs_free_path(path);
2979         return 0;
2980 }
2981
2982 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2983 {
2984         struct btrfs_block_group_cache *block_group;
2985         int readonly = 0;
2986
2987         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2988         if (!block_group || block_group->ro)
2989                 readonly = 1;
2990         if (block_group)
2991                 btrfs_put_block_group(block_group);
2992         return readonly;
2993 }
2994
2995 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2996                              u64 total_bytes, u64 bytes_used,
2997                              struct btrfs_space_info **space_info)
2998 {
2999         struct btrfs_space_info *found;
3000         int i;
3001         int factor;
3002
3003         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3004                      BTRFS_BLOCK_GROUP_RAID10))
3005                 factor = 2;
3006         else
3007                 factor = 1;
3008
3009         found = __find_space_info(info, flags);
3010         if (found) {
3011                 spin_lock(&found->lock);
3012                 found->total_bytes += total_bytes;
3013                 found->disk_total += total_bytes * factor;
3014                 found->bytes_used += bytes_used;
3015                 found->disk_used += bytes_used * factor;
3016                 found->full = 0;
3017                 spin_unlock(&found->lock);
3018                 *space_info = found;
3019                 return 0;
3020         }
3021         found = kzalloc(sizeof(*found), GFP_NOFS);
3022         if (!found)
3023                 return -ENOMEM;
3024
3025         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3026                 INIT_LIST_HEAD(&found->block_groups[i]);
3027         init_rwsem(&found->groups_sem);
3028         spin_lock_init(&found->lock);
3029         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3030                                 BTRFS_BLOCK_GROUP_SYSTEM |
3031                                 BTRFS_BLOCK_GROUP_METADATA);
3032         found->total_bytes = total_bytes;
3033         found->disk_total = total_bytes * factor;
3034         found->bytes_used = bytes_used;
3035         found->disk_used = bytes_used * factor;
3036         found->bytes_pinned = 0;
3037         found->bytes_reserved = 0;
3038         found->bytes_readonly = 0;
3039         found->bytes_may_use = 0;
3040         found->full = 0;
3041         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3042         found->chunk_alloc = 0;
3043         *space_info = found;
3044         list_add_rcu(&found->list, &info->space_info);
3045         atomic_set(&found->caching_threads, 0);
3046         return 0;
3047 }
3048
3049 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3050 {
3051         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3052                                    BTRFS_BLOCK_GROUP_RAID1 |
3053                                    BTRFS_BLOCK_GROUP_RAID10 |
3054                                    BTRFS_BLOCK_GROUP_DUP);
3055         if (extra_flags) {
3056                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3057                         fs_info->avail_data_alloc_bits |= extra_flags;
3058                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3059                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3060                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3061                         fs_info->avail_system_alloc_bits |= extra_flags;
3062         }
3063 }
3064
3065 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3066 {
3067         /*
3068          * we add in the count of missing devices because we want
3069          * to make sure that any RAID levels on a degraded FS
3070          * continue to be honored.
3071          */
3072         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3073                 root->fs_info->fs_devices->missing_devices;
3074
3075         if (num_devices == 1)
3076                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3077         if (num_devices < 4)
3078                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3079
3080         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3081             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3082                       BTRFS_BLOCK_GROUP_RAID10))) {
3083                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3084         }
3085
3086         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3087             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3088                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3089         }
3090
3091         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3092             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3093              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3094              (flags & BTRFS_BLOCK_GROUP_DUP)))
3095                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3096         return flags;
3097 }
3098
3099 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3100 {
3101         if (flags & BTRFS_BLOCK_GROUP_DATA)
3102                 flags |= root->fs_info->avail_data_alloc_bits &
3103                          root->fs_info->data_alloc_profile;
3104         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3105                 flags |= root->fs_info->avail_system_alloc_bits &
3106                          root->fs_info->system_alloc_profile;
3107         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3108                 flags |= root->fs_info->avail_metadata_alloc_bits &
3109                          root->fs_info->metadata_alloc_profile;
3110         return btrfs_reduce_alloc_profile(root, flags);
3111 }
3112
3113 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3114 {
3115         u64 flags;
3116
3117         if (data)
3118                 flags = BTRFS_BLOCK_GROUP_DATA;
3119         else if (root == root->fs_info->chunk_root)
3120                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3121         else
3122                 flags = BTRFS_BLOCK_GROUP_METADATA;
3123
3124         return get_alloc_profile(root, flags);
3125 }
3126
3127 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3128 {
3129         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3130                                                        BTRFS_BLOCK_GROUP_DATA);
3131 }
3132
3133 /*
3134  * This will check the space that the inode allocates from to make sure we have
3135  * enough space for bytes.
3136  */
3137 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3138 {
3139         struct btrfs_space_info *data_sinfo;
3140         struct btrfs_root *root = BTRFS_I(inode)->root;
3141         u64 used;
3142         int ret = 0, committed = 0, alloc_chunk = 1;
3143
3144         /* make sure bytes are sectorsize aligned */
3145         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3146
3147         if (root == root->fs_info->tree_root) {
3148                 alloc_chunk = 0;
3149                 committed = 1;
3150         }
3151
3152         data_sinfo = BTRFS_I(inode)->space_info;
3153         if (!data_sinfo)
3154                 goto alloc;
3155
3156 again:
3157         /* make sure we have enough space to handle the data first */
3158         spin_lock(&data_sinfo->lock);
3159         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3160                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3161                 data_sinfo->bytes_may_use;
3162
3163         if (used + bytes > data_sinfo->total_bytes) {
3164                 struct btrfs_trans_handle *trans;
3165
3166                 /*
3167                  * if we don't have enough free bytes in this space then we need
3168                  * to alloc a new chunk.
3169                  */
3170                 if (!data_sinfo->full && alloc_chunk) {
3171                         u64 alloc_target;
3172
3173                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3174                         spin_unlock(&data_sinfo->lock);
3175 alloc:
3176                         alloc_target = btrfs_get_alloc_profile(root, 1);
3177                         trans = btrfs_join_transaction(root);
3178                         if (IS_ERR(trans))
3179                                 return PTR_ERR(trans);
3180
3181                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3182                                              bytes + 2 * 1024 * 1024,
3183                                              alloc_target,
3184                                              CHUNK_ALLOC_NO_FORCE);
3185                         btrfs_end_transaction(trans, root);
3186                         if (ret < 0) {
3187                                 if (ret != -ENOSPC)
3188                                         return ret;
3189                                 else
3190                                         goto commit_trans;
3191                         }
3192
3193                         if (!data_sinfo) {
3194                                 btrfs_set_inode_space_info(root, inode);
3195                                 data_sinfo = BTRFS_I(inode)->space_info;
3196                         }
3197                         goto again;
3198                 }
3199                 spin_unlock(&data_sinfo->lock);
3200
3201                 /* commit the current transaction and try again */
3202 commit_trans:
3203                 if (!committed &&
3204                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3205                         committed = 1;
3206                         trans = btrfs_join_transaction(root);
3207                         if (IS_ERR(trans))
3208                                 return PTR_ERR(trans);
3209                         ret = btrfs_commit_transaction(trans, root);
3210                         if (ret)
3211                                 return ret;
3212                         goto again;
3213                 }
3214
3215 #if 0 /* I hope we never need this code again, just in case */
3216                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3217                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3218                        "%llu bytes_readonly, %llu may use %llu total\n",
3219                        (unsigned long long)bytes,
3220                        (unsigned long long)data_sinfo->bytes_used,
3221                        (unsigned long long)data_sinfo->bytes_reserved,
3222                        (unsigned long long)data_sinfo->bytes_pinned,
3223                        (unsigned long long)data_sinfo->bytes_readonly,
3224                        (unsigned long long)data_sinfo->bytes_may_use,
3225                        (unsigned long long)data_sinfo->total_bytes);
3226 #endif
3227                 return -ENOSPC;
3228         }
3229         data_sinfo->bytes_may_use += bytes;
3230         BTRFS_I(inode)->reserved_bytes += bytes;
3231         spin_unlock(&data_sinfo->lock);
3232
3233         return 0;
3234 }
3235
3236 /*
3237  * called when we are clearing an delalloc extent from the
3238  * inode's io_tree or there was an error for whatever reason
3239  * after calling btrfs_check_data_free_space
3240  */
3241 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3242 {
3243         struct btrfs_root *root = BTRFS_I(inode)->root;
3244         struct btrfs_space_info *data_sinfo;
3245
3246         /* make sure bytes are sectorsize aligned */
3247         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3248
3249         data_sinfo = BTRFS_I(inode)->space_info;
3250         spin_lock(&data_sinfo->lock);
3251         data_sinfo->bytes_may_use -= bytes;
3252         BTRFS_I(inode)->reserved_bytes -= bytes;
3253         spin_unlock(&data_sinfo->lock);
3254 }
3255
3256 static void force_metadata_allocation(struct btrfs_fs_info *info)
3257 {
3258         struct list_head *head = &info->space_info;
3259         struct btrfs_space_info *found;
3260
3261         rcu_read_lock();
3262         list_for_each_entry_rcu(found, head, list) {
3263                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3264                         found->force_alloc = CHUNK_ALLOC_FORCE;
3265         }
3266         rcu_read_unlock();
3267 }
3268
3269 static int should_alloc_chunk(struct btrfs_root *root,
3270                               struct btrfs_space_info *sinfo, u64 alloc_bytes,
3271                               int force)
3272 {
3273         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3274         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3275         u64 thresh;
3276
3277         if (force == CHUNK_ALLOC_FORCE)
3278                 return 1;
3279
3280         /*
3281          * in limited mode, we want to have some free space up to
3282          * about 1% of the FS size.
3283          */
3284         if (force == CHUNK_ALLOC_LIMITED) {
3285                 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3286                 thresh = max_t(u64, 64 * 1024 * 1024,
3287                                div_factor_fine(thresh, 1));
3288
3289                 if (num_bytes - num_allocated < thresh)
3290                         return 1;
3291         }
3292
3293         /*
3294          * we have two similar checks here, one based on percentage
3295          * and once based on a hard number of 256MB.  The idea
3296          * is that if we have a good amount of free
3297          * room, don't allocate a chunk.  A good mount is
3298          * less than 80% utilized of the chunks we have allocated,
3299          * or more than 256MB free
3300          */
3301         if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3302                 return 0;
3303
3304         if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3305                 return 0;
3306
3307         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3308
3309         /* 256MB or 5% of the FS */
3310         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3311
3312         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3313                 return 0;
3314         return 1;
3315 }
3316
3317 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3318                           struct btrfs_root *extent_root, u64 alloc_bytes,
3319                           u64 flags, int force)
3320 {
3321         struct btrfs_space_info *space_info;
3322         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3323         int wait_for_alloc = 0;
3324         int ret = 0;
3325
3326         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3327
3328         space_info = __find_space_info(extent_root->fs_info, flags);
3329         if (!space_info) {
3330                 ret = update_space_info(extent_root->fs_info, flags,
3331                                         0, 0, &space_info);
3332                 BUG_ON(ret);
3333         }
3334         BUG_ON(!space_info);
3335
3336 again:
3337         spin_lock(&space_info->lock);
3338         if (space_info->force_alloc)
3339                 force = space_info->force_alloc;
3340         if (space_info->full) {
3341                 spin_unlock(&space_info->lock);
3342                 return 0;
3343         }
3344
3345         if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3346                 spin_unlock(&space_info->lock);
3347                 return 0;
3348         } else if (space_info->chunk_alloc) {
3349                 wait_for_alloc = 1;
3350         } else {
3351                 space_info->chunk_alloc = 1;
3352         }
3353
3354         spin_unlock(&space_info->lock);
3355
3356         mutex_lock(&fs_info->chunk_mutex);
3357
3358         /*
3359          * The chunk_mutex is held throughout the entirety of a chunk
3360          * allocation, so once we've acquired the chunk_mutex we know that the
3361          * other guy is done and we need to recheck and see if we should
3362          * allocate.
3363          */
3364         if (wait_for_alloc) {
3365                 mutex_unlock(&fs_info->chunk_mutex);
3366                 wait_for_alloc = 0;
3367                 goto again;
3368         }
3369
3370         /*
3371          * If we have mixed data/metadata chunks we want to make sure we keep
3372          * allocating mixed chunks instead of individual chunks.
3373          */
3374         if (btrfs_mixed_space_info(space_info))
3375                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3376
3377         /*
3378          * if we're doing a data chunk, go ahead and make sure that
3379          * we keep a reasonable number of metadata chunks allocated in the
3380          * FS as well.
3381          */
3382         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3383                 fs_info->data_chunk_allocations++;
3384                 if (!(fs_info->data_chunk_allocations %
3385                       fs_info->metadata_ratio))
3386                         force_metadata_allocation(fs_info);
3387         }
3388
3389         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3390         spin_lock(&space_info->lock);
3391         if (ret)
3392                 space_info->full = 1;
3393         else
3394                 ret = 1;
3395
3396         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3397         space_info->chunk_alloc = 0;
3398         spin_unlock(&space_info->lock);
3399         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3400         return ret;
3401 }
3402
3403 /*
3404  * shrink metadata reservation for delalloc
3405  */
3406 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3407                            struct btrfs_root *root, u64 to_reclaim, int sync)
3408 {
3409         struct btrfs_block_rsv *block_rsv;
3410         struct btrfs_space_info *space_info;
3411         u64 reserved;
3412         u64 max_reclaim;
3413         u64 reclaimed = 0;
3414         long time_left;
3415         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3416         int loops = 0;
3417         unsigned long progress;
3418
3419         block_rsv = &root->fs_info->delalloc_block_rsv;
3420         space_info = block_rsv->space_info;
3421
3422         smp_mb();
3423         reserved = space_info->bytes_reserved;
3424         progress = space_info->reservation_progress;
3425
3426         if (reserved == 0)
3427                 return 0;
3428
3429         max_reclaim = min(reserved, to_reclaim);
3430
3431         while (loops < 1024) {
3432                 /* have the flusher threads jump in and do some IO */
3433                 smp_mb();
3434                 nr_pages = min_t(unsigned long, nr_pages,
3435                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3436                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3437
3438                 spin_lock(&space_info->lock);
3439                 if (reserved > space_info->bytes_reserved)
3440                         reclaimed += reserved - space_info->bytes_reserved;
3441                 reserved = space_info->bytes_reserved;
3442                 spin_unlock(&space_info->lock);
3443
3444                 loops++;
3445
3446                 if (reserved == 0 || reclaimed >= max_reclaim)
3447                         break;
3448
3449                 if (trans && trans->transaction->blocked)
3450                         return -EAGAIN;
3451
3452                 time_left = schedule_timeout_interruptible(1);
3453
3454                 /* We were interrupted, exit */
3455                 if (time_left)
3456                         break;
3457
3458                 /* we've kicked the IO a few times, if anything has been freed,
3459                  * exit.  There is no sense in looping here for a long time
3460                  * when we really need to commit the transaction, or there are
3461                  * just too many writers without enough free space
3462                  */
3463
3464                 if (loops > 3) {
3465                         smp_mb();
3466                         if (progress != space_info->reservation_progress)
3467                                 break;
3468                 }
3469
3470         }
3471         return reclaimed >= to_reclaim;
3472 }
3473
3474 /*
3475  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3476  * idea is if this is the first call (retries == 0) then we will add to our
3477  * reserved count if we can't make the allocation in order to hold our place
3478  * while we go and try and free up space.  That way for retries > 1 we don't try
3479  * and add space, we just check to see if the amount of unused space is >= the
3480  * total space, meaning that our reservation is valid.
3481  *
3482  * However if we don't intend to retry this reservation, pass -1 as retries so
3483  * that it short circuits this logic.
3484  */
3485 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3486                                   struct btrfs_root *root,
3487                                   struct btrfs_block_rsv *block_rsv,
3488                                   u64 orig_bytes, int flush)
3489 {
3490         struct btrfs_space_info *space_info = block_rsv->space_info;
3491         u64 unused;
3492         u64 num_bytes = orig_bytes;
3493         int retries = 0;
3494         int ret = 0;
3495         bool reserved = false;
3496         bool committed = false;
3497
3498 again:
3499         ret = -ENOSPC;
3500         if (reserved)
3501                 num_bytes = 0;
3502
3503         spin_lock(&space_info->lock);
3504         unused = space_info->bytes_used + space_info->bytes_reserved +
3505                  space_info->bytes_pinned + space_info->bytes_readonly +
3506                  space_info->bytes_may_use;
3507
3508         /*
3509          * The idea here is that we've not already over-reserved the block group
3510          * then we can go ahead and save our reservation first and then start
3511          * flushing if we need to.  Otherwise if we've already overcommitted
3512          * lets start flushing stuff first and then come back and try to make
3513          * our reservation.
3514          */
3515         if (unused <= space_info->total_bytes) {
3516                 unused = space_info->total_bytes - unused;
3517                 if (unused >= num_bytes) {
3518                         if (!reserved)
3519                                 space_info->bytes_reserved += orig_bytes;
3520                         ret = 0;
3521                 } else {
3522                         /*
3523                          * Ok set num_bytes to orig_bytes since we aren't
3524                          * overocmmitted, this way we only try and reclaim what
3525                          * we need.
3526                          */
3527                         num_bytes = orig_bytes;
3528                 }
3529         } else {
3530                 /*
3531                  * Ok we're over committed, set num_bytes to the overcommitted
3532                  * amount plus the amount of bytes that we need for this
3533                  * reservation.
3534                  */
3535                 num_bytes = unused - space_info->total_bytes +
3536                         (orig_bytes * (retries + 1));
3537         }
3538
3539         /*
3540          * Couldn't make our reservation, save our place so while we're trying
3541          * to reclaim space we can actually use it instead of somebody else
3542          * stealing it from us.
3543          */
3544         if (ret && !reserved) {
3545                 space_info->bytes_reserved += orig_bytes;
3546                 reserved = true;
3547         }
3548
3549         spin_unlock(&space_info->lock);
3550
3551         if (!ret)
3552                 return 0;
3553
3554         if (!flush)
3555                 goto out;
3556
3557         /*
3558          * We do synchronous shrinking since we don't actually unreserve
3559          * metadata until after the IO is completed.
3560          */
3561         ret = shrink_delalloc(trans, root, num_bytes, 1);
3562         if (ret > 0)
3563                 return 0;
3564         else if (ret < 0)
3565                 goto out;
3566
3567         /*
3568          * So if we were overcommitted it's possible that somebody else flushed
3569          * out enough space and we simply didn't have enough space to reclaim,
3570          * so go back around and try again.
3571          */
3572         if (retries < 2) {
3573                 retries++;
3574                 goto again;
3575         }
3576
3577         spin_lock(&space_info->lock);
3578         /*
3579          * Not enough space to be reclaimed, don't bother committing the
3580          * transaction.
3581          */
3582         if (space_info->bytes_pinned < orig_bytes)
3583                 ret = -ENOSPC;
3584         spin_unlock(&space_info->lock);
3585         if (ret)
3586                 goto out;
3587
3588         ret = -EAGAIN;
3589         if (trans || committed)
3590                 goto out;
3591
3592         ret = -ENOSPC;
3593         trans = btrfs_join_transaction(root);
3594         if (IS_ERR(trans))
3595                 goto out;
3596         ret = btrfs_commit_transaction(trans, root);
3597         if (!ret) {
3598                 trans = NULL;
3599                 committed = true;
3600                 goto again;
3601         }
3602
3603 out:
3604         if (reserved) {
3605                 spin_lock(&space_info->lock);
3606                 space_info->bytes_reserved -= orig_bytes;
3607                 spin_unlock(&space_info->lock);
3608         }
3609
3610         return ret;
3611 }
3612
3613 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3614                                              struct btrfs_root *root)
3615 {
3616         struct btrfs_block_rsv *block_rsv;
3617         if (root->ref_cows)
3618                 block_rsv = trans->block_rsv;
3619         else
3620                 block_rsv = root->block_rsv;
3621
3622         if (!block_rsv)
3623                 block_rsv = &root->fs_info->empty_block_rsv;
3624
3625         return block_rsv;
3626 }
3627
3628 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3629                                u64 num_bytes)
3630 {
3631         int ret = -ENOSPC;
3632         spin_lock(&block_rsv->lock);
3633         if (block_rsv->reserved >= num_bytes) {
3634                 block_rsv->reserved -= num_bytes;
3635                 if (block_rsv->reserved < block_rsv->size)
3636                         block_rsv->full = 0;
3637                 ret = 0;
3638         }
3639         spin_unlock(&block_rsv->lock);
3640         return ret;
3641 }
3642
3643 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3644                                 u64 num_bytes, int update_size)
3645 {
3646         spin_lock(&block_rsv->lock);
3647         block_rsv->reserved += num_bytes;
3648         if (update_size)
3649                 block_rsv->size += num_bytes;
3650         else if (block_rsv->reserved >= block_rsv->size)
3651                 block_rsv->full = 1;
3652         spin_unlock(&block_rsv->lock);
3653 }
3654
3655 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3656                              struct btrfs_block_rsv *dest, u64 num_bytes)
3657 {
3658         struct btrfs_space_info *space_info = block_rsv->space_info;
3659
3660         spin_lock(&block_rsv->lock);
3661         if (num_bytes == (u64)-1)
3662                 num_bytes = block_rsv->size;
3663         block_rsv->size -= num_bytes;
3664         if (block_rsv->reserved >= block_rsv->size) {
3665                 num_bytes = block_rsv->reserved - block_rsv->size;
3666                 block_rsv->reserved = block_rsv->size;
3667                 block_rsv->full = 1;
3668         } else {
3669                 num_bytes = 0;
3670         }
3671         spin_unlock(&block_rsv->lock);
3672
3673         if (num_bytes > 0) {
3674                 if (dest) {
3675                         spin_lock(&dest->lock);
3676                         if (!dest->full) {
3677                                 u64 bytes_to_add;
3678
3679                                 bytes_to_add = dest->size - dest->reserved;
3680                                 bytes_to_add = min(num_bytes, bytes_to_add);
3681                                 dest->reserved += bytes_to_add;
3682                                 if (dest->reserved >= dest->size)
3683                                         dest->full = 1;
3684                                 num_bytes -= bytes_to_add;
3685                         }
3686                         spin_unlock(&dest->lock);
3687                 }
3688                 if (num_bytes) {
3689                         spin_lock(&space_info->lock);
3690                         space_info->bytes_reserved -= num_bytes;
3691                         space_info->reservation_progress++;
3692                         spin_unlock(&space_info->lock);
3693                 }
3694         }
3695 }
3696
3697 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3698                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3699 {
3700         int ret;
3701
3702         ret = block_rsv_use_bytes(src, num_bytes);
3703         if (ret)
3704                 return ret;
3705
3706         block_rsv_add_bytes(dst, num_bytes, 1);
3707         return 0;
3708 }
3709
3710 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3711 {
3712         memset(rsv, 0, sizeof(*rsv));
3713         spin_lock_init(&rsv->lock);
3714         atomic_set(&rsv->usage, 1);
3715         rsv->priority = 6;
3716         INIT_LIST_HEAD(&rsv->list);
3717 }
3718
3719 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3720 {
3721         struct btrfs_block_rsv *block_rsv;
3722         struct btrfs_fs_info *fs_info = root->fs_info;
3723
3724         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3725         if (!block_rsv)
3726                 return NULL;
3727
3728         btrfs_init_block_rsv(block_rsv);
3729         block_rsv->space_info = __find_space_info(fs_info,
3730                                                   BTRFS_BLOCK_GROUP_METADATA);
3731         return block_rsv;
3732 }
3733
3734 void btrfs_free_block_rsv(struct btrfs_root *root,
3735                           struct btrfs_block_rsv *rsv)
3736 {
3737         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3738                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3739                 if (!rsv->durable)
3740                         kfree(rsv);
3741         }
3742 }
3743
3744 /*
3745  * make the block_rsv struct be able to capture freed space.
3746  * the captured space will re-add to the the block_rsv struct
3747  * after transaction commit
3748  */
3749 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3750                                  struct btrfs_block_rsv *block_rsv)
3751 {
3752         block_rsv->durable = 1;
3753         mutex_lock(&fs_info->durable_block_rsv_mutex);
3754         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3755         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3756 }
3757
3758 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3759                         struct btrfs_root *root,
3760                         struct btrfs_block_rsv *block_rsv,
3761                         u64 num_bytes)
3762 {
3763         int ret;
3764
3765         if (num_bytes == 0)
3766                 return 0;
3767
3768         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3769         if (!ret) {
3770                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3771                 return 0;
3772         }
3773
3774         return ret;
3775 }
3776
3777 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3778                           struct btrfs_root *root,
3779                           struct btrfs_block_rsv *block_rsv,
3780                           u64 min_reserved, int min_factor)
3781 {
3782         u64 num_bytes = 0;
3783         int commit_trans = 0;
3784         int ret = -ENOSPC;
3785
3786         if (!block_rsv)
3787                 return 0;
3788
3789         spin_lock(&block_rsv->lock);
3790         if (min_factor > 0)
3791                 num_bytes = div_factor(block_rsv->size, min_factor);
3792         if (min_reserved > num_bytes)
3793                 num_bytes = min_reserved;
3794
3795         if (block_rsv->reserved >= num_bytes) {
3796                 ret = 0;
3797         } else {
3798                 num_bytes -= block_rsv->reserved;
3799                 if (block_rsv->durable &&
3800                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3801                         commit_trans = 1;
3802         }
3803         spin_unlock(&block_rsv->lock);
3804         if (!ret)
3805                 return 0;
3806
3807         if (block_rsv->refill_used) {
3808                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3809                                              num_bytes, 0);
3810                 if (!ret) {
3811                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3812                         return 0;
3813                 }
3814         }
3815
3816         if (commit_trans) {
3817                 if (trans)
3818                         return -EAGAIN;
3819
3820                 trans = btrfs_join_transaction(root);
3821                 BUG_ON(IS_ERR(trans));
3822                 ret = btrfs_commit_transaction(trans, root);
3823                 return 0;
3824         }
3825
3826         return -ENOSPC;
3827 }
3828
3829 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3830                             struct btrfs_block_rsv *dst_rsv,
3831                             u64 num_bytes)
3832 {
3833         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3834 }
3835
3836 void btrfs_block_rsv_release(struct btrfs_root *root,
3837                              struct btrfs_block_rsv *block_rsv,
3838                              u64 num_bytes)
3839 {
3840         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3841         if (global_rsv->full || global_rsv == block_rsv ||
3842             block_rsv->space_info != global_rsv->space_info)
3843                 global_rsv = NULL;
3844         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3845 }
3846
3847 /*
3848  * helper to calculate size of global block reservation.
3849  * the desired value is sum of space used by extent tree,
3850  * checksum tree and root tree
3851  */
3852 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3853 {
3854         struct btrfs_space_info *sinfo;
3855         u64 num_bytes;
3856         u64 meta_used;
3857         u64 data_used;
3858         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3859 #if 0
3860         /*
3861          * per tree used space accounting can be inaccuracy, so we
3862          * can't rely on it.
3863          */
3864         spin_lock(&fs_info->extent_root->accounting_lock);
3865         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3866         spin_unlock(&fs_info->extent_root->accounting_lock);
3867
3868         spin_lock(&fs_info->csum_root->accounting_lock);
3869         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3870         spin_unlock(&fs_info->csum_root->accounting_lock);
3871
3872         spin_lock(&fs_info->tree_root->accounting_lock);
3873         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3874         spin_unlock(&fs_info->tree_root->accounting_lock);
3875 #endif
3876         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3877         spin_lock(&sinfo->lock);
3878         data_used = sinfo->bytes_used;
3879         spin_unlock(&sinfo->lock);
3880
3881         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3882         spin_lock(&sinfo->lock);
3883         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3884                 data_used = 0;
3885         meta_used = sinfo->bytes_used;
3886         spin_unlock(&sinfo->lock);
3887
3888         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3889                     csum_size * 2;
3890         num_bytes += div64_u64(data_used + meta_used, 50);
3891
3892         if (num_bytes * 3 > meta_used)
3893                 num_bytes = div64_u64(meta_used, 3);
3894
3895         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3896 }
3897
3898 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3899 {
3900         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3901         struct btrfs_space_info *sinfo = block_rsv->space_info;
3902         u64 num_bytes;
3903
3904         num_bytes = calc_global_metadata_size(fs_info);
3905
3906         spin_lock(&block_rsv->lock);
3907         spin_lock(&sinfo->lock);
3908
3909         block_rsv->size = num_bytes;
3910
3911         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3912                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3913                     sinfo->bytes_may_use;
3914
3915         if (sinfo->total_bytes > num_bytes) {
3916                 num_bytes = sinfo->total_bytes - num_bytes;
3917                 block_rsv->reserved += num_bytes;
3918                 sinfo->bytes_reserved += num_bytes;
3919         }
3920
3921         if (block_rsv->reserved >= block_rsv->size) {
3922                 num_bytes = block_rsv->reserved - block_rsv->size;
3923                 sinfo->bytes_reserved -= num_bytes;
3924                 sinfo->reservation_progress++;
3925                 block_rsv->reserved = block_rsv->size;
3926                 block_rsv->full = 1;
3927         }
3928 #if 0
3929         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3930                 block_rsv->size, block_rsv->reserved);
3931 #endif
3932         spin_unlock(&sinfo->lock);
3933         spin_unlock(&block_rsv->lock);
3934 }
3935
3936 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3937 {
3938         struct btrfs_space_info *space_info;
3939
3940         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3941         fs_info->chunk_block_rsv.space_info = space_info;
3942         fs_info->chunk_block_rsv.priority = 10;
3943
3944         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3945         fs_info->global_block_rsv.space_info = space_info;
3946         fs_info->global_block_rsv.priority = 10;
3947         fs_info->global_block_rsv.refill_used = 1;
3948         fs_info->delalloc_block_rsv.space_info = space_info;
3949         fs_info->trans_block_rsv.space_info = space_info;
3950         fs_info->empty_block_rsv.space_info = space_info;
3951         fs_info->empty_block_rsv.priority = 10;
3952
3953         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3954         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3955         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3956         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3957         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3958
3959         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3960
3961         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3962
3963         update_global_block_rsv(fs_info);
3964 }
3965
3966 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3967 {
3968         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3969         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3970         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3971         WARN_ON(fs_info->trans_block_rsv.size > 0);
3972         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3973         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3974         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3975 }
3976
3977 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3978 {
3979         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3980                 3 * num_items;
3981 }
3982
3983 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3984                                  struct btrfs_root *root,
3985                                  int num_items)
3986 {
3987         u64 num_bytes;
3988         int ret;
3989
3990         if (num_items == 0 || root->fs_info->chunk_root == root)
3991                 return 0;
3992
3993         num_bytes = calc_trans_metadata_size(root, num_items);
3994         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3995                                   num_bytes);
3996         if (!ret) {
3997                 trans->bytes_reserved += num_bytes;
3998                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3999         }
4000         return ret;
4001 }
4002
4003 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4004                                   struct btrfs_root *root)
4005 {
4006         if (!trans->bytes_reserved)
4007                 return;
4008
4009         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
4010         btrfs_block_rsv_release(root, trans->block_rsv,
4011                                 trans->bytes_reserved);
4012         trans->bytes_reserved = 0;
4013 }
4014
4015 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4016                                   struct inode *inode)
4017 {
4018         struct btrfs_root *root = BTRFS_I(inode)->root;
4019         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4020         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4021
4022         /*
4023          * one for deleting orphan item, one for updating inode and
4024          * two for calling btrfs_truncate_inode_items.
4025          *
4026          * btrfs_truncate_inode_items is a delete operation, it frees
4027          * more space than it uses in most cases. So two units of
4028          * metadata space should be enough for calling it many times.
4029          * If all of the metadata space is used, we can commit
4030          * transaction and use space it freed.
4031          */
4032         u64 num_bytes = calc_trans_metadata_size(root, 4);
4033         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4034 }
4035
4036 void btrfs_orphan_release_metadata(struct inode *inode)
4037 {
4038         struct btrfs_root *root = BTRFS_I(inode)->root;
4039         u64 num_bytes = calc_trans_metadata_size(root, 4);
4040         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4041 }
4042
4043 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4044                                 struct btrfs_pending_snapshot *pending)
4045 {
4046         struct btrfs_root *root = pending->root;
4047         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4048         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4049         /*
4050          * two for root back/forward refs, two for directory entries
4051          * and one for root of the snapshot.
4052          */
4053         u64 num_bytes = calc_trans_metadata_size(root, 5);
4054         dst_rsv->space_info = src_rsv->space_info;
4055         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4056 }
4057
4058 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
4059 {
4060         return num_bytes >>= 3;
4061 }
4062
4063 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4064 {
4065         struct btrfs_root *root = BTRFS_I(inode)->root;
4066         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4067         u64 to_reserve;
4068         int nr_extents;
4069         int reserved_extents;
4070         int ret;
4071
4072         if (btrfs_transaction_in_commit(root->fs_info))
4073                 schedule_timeout(1);
4074
4075         num_bytes = ALIGN(num_bytes, root->sectorsize);
4076
4077         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
4078         reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4079
4080         if (nr_extents > reserved_extents) {
4081                 nr_extents -= reserved_extents;
4082                 to_reserve = calc_trans_metadata_size(root, nr_extents);
4083         } else {
4084                 nr_extents = 0;
4085                 to_reserve = 0;
4086         }
4087
4088         to_reserve += calc_csum_metadata_size(inode, num_bytes);
4089         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4090         if (ret)
4091                 return ret;
4092
4093         atomic_add(nr_extents, &BTRFS_I(inode)->reserved_extents);
4094         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4095
4096         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4097
4098         if (block_rsv->size > 512 * 1024 * 1024)
4099                 shrink_delalloc(NULL, root, to_reserve, 0);
4100
4101         return 0;
4102 }
4103
4104 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4105 {
4106         struct btrfs_root *root = BTRFS_I(inode)->root;
4107         u64 to_free;
4108         int nr_extents;
4109         int reserved_extents;
4110
4111         num_bytes = ALIGN(num_bytes, root->sectorsize);
4112         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4113         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
4114
4115         reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4116         do {
4117                 int old, new;
4118
4119                 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4120                 if (nr_extents >= reserved_extents) {
4121                         nr_extents = 0;
4122                         break;
4123                 }
4124                 old = reserved_extents;
4125                 nr_extents = reserved_extents - nr_extents;
4126                 new = reserved_extents - nr_extents;
4127                 old = atomic_cmpxchg(&BTRFS_I(inode)->reserved_extents,
4128                                      reserved_extents, new);
4129                 if (likely(old == reserved_extents))
4130                         break;
4131                 reserved_extents = old;
4132         } while (1);
4133
4134         to_free = calc_csum_metadata_size(inode, num_bytes);
4135         if (nr_extents > 0)
4136                 to_free += calc_trans_metadata_size(root, nr_extents);
4137
4138         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4139                                 to_free);
4140 }
4141
4142 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4143 {
4144         int ret;
4145
4146         ret = btrfs_check_data_free_space(inode, num_bytes);
4147         if (ret)
4148                 return ret;
4149
4150         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4151         if (ret) {
4152                 btrfs_free_reserved_data_space(inode, num_bytes);
4153                 return ret;
4154         }
4155
4156         return 0;
4157 }
4158
4159 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4160 {
4161         btrfs_delalloc_release_metadata(inode, num_bytes);
4162         btrfs_free_reserved_data_space(inode, num_bytes);
4163 }
4164
4165 static int update_block_group(struct btrfs_trans_handle *trans,
4166                               struct btrfs_root *root,
4167                               u64 bytenr, u64 num_bytes, int alloc)
4168 {
4169         struct btrfs_block_group_cache *cache = NULL;
4170         struct btrfs_fs_info *info = root->fs_info;
4171         u64 total = num_bytes;
4172         u64 old_val;
4173         u64 byte_in_group;
4174         int factor;
4175
4176         /* block accounting for super block */
4177         spin_lock(&info->delalloc_lock);
4178         old_val = btrfs_super_bytes_used(&info->super_copy);
4179         if (alloc)
4180                 old_val += num_bytes;
4181         else
4182                 old_val -= num_bytes;
4183         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4184         spin_unlock(&info->delalloc_lock);
4185
4186         while (total) {
4187                 cache = btrfs_lookup_block_group(info, bytenr);
4188                 if (!cache)
4189                         return -1;
4190                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4191                                     BTRFS_BLOCK_GROUP_RAID1 |
4192                                     BTRFS_BLOCK_GROUP_RAID10))
4193                         factor = 2;
4194                 else
4195                         factor = 1;
4196                 /*
4197                  * If this block group has free space cache written out, we
4198                  * need to make sure to load it if we are removing space.  This
4199                  * is because we need the unpinning stage to actually add the
4200                  * space back to the block group, otherwise we will leak space.
4201                  */
4202                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4203                         cache_block_group(cache, trans, NULL, 1);
4204
4205                 byte_in_group = bytenr - cache->key.objectid;
4206                 WARN_ON(byte_in_group > cache->key.offset);
4207
4208                 spin_lock(&cache->space_info->lock);
4209                 spin_lock(&cache->lock);
4210
4211                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4212                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4213                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4214
4215                 cache->dirty = 1;
4216                 old_val = btrfs_block_group_used(&cache->item);
4217                 num_bytes = min(total, cache->key.offset - byte_in_group);
4218                 if (alloc) {
4219                         old_val += num_bytes;
4220                         btrfs_set_block_group_used(&cache->item, old_val);
4221                         cache->reserved -= num_bytes;
4222                         cache->space_info->bytes_reserved -= num_bytes;
4223                         cache->space_info->reservation_progress++;
4224                         cache->space_info->bytes_used += num_bytes;
4225                         cache->space_info->disk_used += num_bytes * factor;
4226                         spin_unlock(&cache->lock);
4227                         spin_unlock(&cache->space_info->lock);
4228                 } else {
4229                         old_val -= num_bytes;
4230                         btrfs_set_block_group_used(&cache->item, old_val);
4231                         cache->pinned += num_bytes;
4232                         cache->space_info->bytes_pinned += num_bytes;
4233                         cache->space_info->bytes_used -= num_bytes;
4234                         cache->space_info->disk_used -= num_bytes * factor;
4235                         spin_unlock(&cache->lock);
4236                         spin_unlock(&cache->space_info->lock);
4237
4238                         set_extent_dirty(info->pinned_extents,
4239                                          bytenr, bytenr + num_bytes - 1,
4240                                          GFP_NOFS | __GFP_NOFAIL);
4241                 }
4242                 btrfs_put_block_group(cache);
4243                 total -= num_bytes;
4244                 bytenr += num_bytes;
4245         }
4246         return 0;
4247 }
4248
4249 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4250 {
4251         struct btrfs_block_group_cache *cache;
4252         u64 bytenr;
4253
4254         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4255         if (!cache)
4256                 return 0;
4257
4258         bytenr = cache->key.objectid;
4259         btrfs_put_block_group(cache);
4260
4261         return bytenr;
4262 }
4263
4264 static int pin_down_extent(struct btrfs_root *root,
4265                            struct btrfs_block_group_cache *cache,
4266                            u64 bytenr, u64 num_bytes, int reserved)
4267 {
4268         spin_lock(&cache->space_info->lock);
4269         spin_lock(&cache->lock);
4270         cache->pinned += num_bytes;
4271         cache->space_info->bytes_pinned += num_bytes;
4272         if (reserved) {
4273                 cache->reserved -= num_bytes;
4274                 cache->space_info->bytes_reserved -= num_bytes;
4275                 cache->space_info->reservation_progress++;
4276         }
4277         spin_unlock(&cache->lock);
4278         spin_unlock(&cache->space_info->lock);
4279
4280         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4281                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4282         return 0;
4283 }
4284
4285 /*
4286  * this function must be called within transaction
4287  */
4288 int btrfs_pin_extent(struct btrfs_root *root,
4289                      u64 bytenr, u64 num_bytes, int reserved)
4290 {
4291         struct btrfs_block_group_cache *cache;
4292
4293         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4294         BUG_ON(!cache);
4295
4296         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4297
4298         btrfs_put_block_group(cache);
4299         return 0;
4300 }
4301
4302 /*
4303  * update size of reserved extents. this function may return -EAGAIN
4304  * if 'reserve' is true or 'sinfo' is false.
4305  */
4306 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4307                                 u64 num_bytes, int reserve, int sinfo)
4308 {
4309         int ret = 0;
4310         if (sinfo) {
4311                 struct btrfs_space_info *space_info = cache->space_info;
4312                 spin_lock(&space_info->lock);
4313                 spin_lock(&cache->lock);
4314                 if (reserve) {
4315                         if (cache->ro) {
4316                                 ret = -EAGAIN;
4317                         } else {
4318                                 cache->reserved += num_bytes;
4319                                 space_info->bytes_reserved += num_bytes;
4320                         }
4321                 } else {
4322                         if (cache->ro)
4323                                 space_info->bytes_readonly += num_bytes;
4324                         cache->reserved -= num_bytes;
4325                         space_info->bytes_reserved -= num_bytes;
4326                         space_info->reservation_progress++;
4327                 }
4328                 spin_unlock(&cache->lock);
4329                 spin_unlock(&space_info->lock);
4330         } else {
4331                 spin_lock(&cache->lock);
4332                 if (cache->ro) {
4333                         ret = -EAGAIN;
4334                 } else {
4335                         if (reserve)
4336                                 cache->reserved += num_bytes;
4337                         else
4338                                 cache->reserved -= num_bytes;
4339                 }
4340                 spin_unlock(&cache->lock);
4341         }
4342         return ret;
4343 }
4344
4345 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4346                                 struct btrfs_root *root)
4347 {
4348         struct btrfs_fs_info *fs_info = root->fs_info;
4349         struct btrfs_caching_control *next;
4350         struct btrfs_caching_control *caching_ctl;
4351         struct btrfs_block_group_cache *cache;
4352
4353         down_write(&fs_info->extent_commit_sem);
4354
4355         list_for_each_entry_safe(caching_ctl, next,
4356                                  &fs_info->caching_block_groups, list) {
4357                 cache = caching_ctl->block_group;
4358                 if (block_group_cache_done(cache)) {
4359                         cache->last_byte_to_unpin = (u64)-1;
4360                         list_del_init(&caching_ctl->list);
4361                         put_caching_control(caching_ctl);
4362                 } else {
4363                         cache->last_byte_to_unpin = caching_ctl->progress;
4364                 }
4365         }
4366
4367         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4368                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4369         else
4370                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4371
4372         up_write(&fs_info->extent_commit_sem);
4373
4374         update_global_block_rsv(fs_info);
4375         return 0;
4376 }
4377
4378 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4379 {
4380         struct btrfs_fs_info *fs_info = root->fs_info;
4381         struct btrfs_block_group_cache *cache = NULL;
4382         u64 len;
4383
4384         while (start <= end) {
4385                 if (!cache ||
4386                     start >= cache->key.objectid + cache->key.offset) {
4387                         if (cache)
4388                                 btrfs_put_block_group(cache);
4389                         cache = btrfs_lookup_block_group(fs_info, start);
4390                         BUG_ON(!cache);
4391                 }
4392
4393                 len = cache->key.objectid + cache->key.offset - start;
4394                 len = min(len, end + 1 - start);
4395
4396                 if (start < cache->last_byte_to_unpin) {
4397                         len = min(len, cache->last_byte_to_unpin - start);
4398                         btrfs_add_free_space(cache, start, len);
4399                 }
4400
4401                 start += len;
4402
4403                 spin_lock(&cache->space_info->lock);
4404                 spin_lock(&cache->lock);
4405                 cache->pinned -= len;
4406                 cache->space_info->bytes_pinned -= len;
4407                 if (cache->ro) {
4408                         cache->space_info->bytes_readonly += len;
4409                 } else if (cache->reserved_pinned > 0) {
4410                         len = min(len, cache->reserved_pinned);
4411                         cache->reserved_pinned -= len;
4412                         cache->space_info->bytes_reserved += len;
4413                 }
4414                 spin_unlock(&cache->lock);
4415                 spin_unlock(&cache->space_info->lock);
4416         }
4417
4418         if (cache)
4419                 btrfs_put_block_group(cache);
4420         return 0;
4421 }
4422
4423 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4424                                struct btrfs_root *root)
4425 {
4426         struct btrfs_fs_info *fs_info = root->fs_info;
4427         struct extent_io_tree *unpin;
4428         struct btrfs_block_rsv *block_rsv;
4429         struct btrfs_block_rsv *next_rsv;
4430         u64 start;
4431         u64 end;
4432         int idx;
4433         int ret;
4434
4435         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4436                 unpin = &fs_info->freed_extents[1];
4437         else
4438                 unpin = &fs_info->freed_extents[0];
4439
4440         while (1) {
4441                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4442                                             EXTENT_DIRTY);
4443                 if (ret)
4444                         break;
4445
4446                 if (btrfs_test_opt(root, DISCARD))
4447                         ret = btrfs_discard_extent(root, start,
4448                                                    end + 1 - start, NULL);
4449
4450                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4451                 unpin_extent_range(root, start, end);
4452                 cond_resched();
4453         }
4454
4455         mutex_lock(&fs_info->durable_block_rsv_mutex);
4456         list_for_each_entry_safe(block_rsv, next_rsv,
4457                                  &fs_info->durable_block_rsv_list, list) {
4458
4459                 idx = trans->transid & 0x1;
4460                 if (block_rsv->freed[idx] > 0) {
4461                         block_rsv_add_bytes(block_rsv,
4462                                             block_rsv->freed[idx], 0);
4463                         block_rsv->freed[idx] = 0;
4464                 }
4465                 if (atomic_read(&block_rsv->usage) == 0) {
4466                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4467
4468                         if (block_rsv->freed[0] == 0 &&
4469                             block_rsv->freed[1] == 0) {
4470                                 list_del_init(&block_rsv->list);
4471                                 kfree(block_rsv);
4472                         }
4473                 } else {
4474                         btrfs_block_rsv_release(root, block_rsv, 0);
4475                 }
4476         }
4477         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4478
4479         return 0;
4480 }
4481
4482 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4483                                 struct btrfs_root *root,
4484                                 u64 bytenr, u64 num_bytes, u64 parent,
4485                                 u64 root_objectid, u64 owner_objectid,
4486                                 u64 owner_offset, int refs_to_drop,
4487                                 struct btrfs_delayed_extent_op *extent_op)
4488 {
4489         struct btrfs_key key;
4490         struct btrfs_path *path;
4491         struct btrfs_fs_info *info = root->fs_info;
4492         struct btrfs_root *extent_root = info->extent_root;
4493         struct extent_buffer *leaf;
4494         struct btrfs_extent_item *ei;
4495         struct btrfs_extent_inline_ref *iref;
4496         int ret;
4497         int is_data;
4498         int extent_slot = 0;
4499         int found_extent = 0;
4500         int num_to_del = 1;
4501         u32 item_size;
4502         u64 refs;
4503
4504         path = btrfs_alloc_path();
4505         if (!path)
4506                 return -ENOMEM;
4507
4508         path->reada = 1;
4509         path->leave_spinning = 1;
4510
4511         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4512         BUG_ON(!is_data && refs_to_drop != 1);
4513
4514         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4515                                     bytenr, num_bytes, parent,
4516                                     root_objectid, owner_objectid,
4517                                     owner_offset);
4518         if (ret == 0) {
4519                 extent_slot = path->slots[0];
4520                 while (extent_slot >= 0) {
4521                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4522                                               extent_slot);
4523                         if (key.objectid != bytenr)
4524                                 break;
4525                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4526                             key.offset == num_bytes) {
4527                                 found_extent = 1;
4528                                 break;
4529                         }
4530                         if (path->slots[0] - extent_slot > 5)
4531                                 break;
4532                         extent_slot--;
4533                 }
4534 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4535                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4536                 if (found_extent && item_size < sizeof(*ei))
4537                         found_extent = 0;
4538 #endif
4539                 if (!found_extent) {
4540                         BUG_ON(iref);
4541                         ret = remove_extent_backref(trans, extent_root, path,
4542                                                     NULL, refs_to_drop,
4543                                                     is_data);
4544                         BUG_ON(ret);
4545                         btrfs_release_path(extent_root, path);
4546                         path->leave_spinning = 1;
4547
4548                         key.objectid = bytenr;
4549                         key.type = BTRFS_EXTENT_ITEM_KEY;
4550                         key.offset = num_bytes;
4551
4552                         ret = btrfs_search_slot(trans, extent_root,
4553                                                 &key, path, -1, 1);
4554                         if (ret) {
4555                                 printk(KERN_ERR "umm, got %d back from search"
4556                                        ", was looking for %llu\n", ret,
4557                                        (unsigned long long)bytenr);
4558                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4559                         }
4560                         BUG_ON(ret);
4561                         extent_slot = path->slots[0];
4562                 }
4563         } else {
4564                 btrfs_print_leaf(extent_root, path->nodes[0]);
4565                 WARN_ON(1);
4566                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4567                        "parent %llu root %llu  owner %llu offset %llu\n",
4568                        (unsigned long long)bytenr,
4569                        (unsigned long long)parent,
4570                        (unsigned long long)root_objectid,
4571                        (unsigned long long)owner_objectid,
4572                        (unsigned long long)owner_offset);
4573         }
4574
4575         leaf = path->nodes[0];
4576         item_size = btrfs_item_size_nr(leaf, extent_slot);
4577 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4578         if (item_size < sizeof(*ei)) {
4579                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4580                 ret = convert_extent_item_v0(trans, extent_root, path,
4581                                              owner_objectid, 0);
4582                 BUG_ON(ret < 0);
4583
4584                 btrfs_release_path(extent_root, path);
4585                 path->leave_spinning = 1;
4586
4587                 key.objectid = bytenr;
4588                 key.type = BTRFS_EXTENT_ITEM_KEY;
4589                 key.offset = num_bytes;
4590
4591                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4592                                         -1, 1);
4593                 if (ret) {
4594                         printk(KERN_ERR "umm, got %d back from search"
4595                                ", was looking for %llu\n", ret,
4596                                (unsigned long long)bytenr);
4597                         btrfs_print_leaf(extent_root, path->nodes[0]);
4598                 }
4599                 BUG_ON(ret);
4600                 extent_slot = path->slots[0];
4601                 leaf = path->nodes[0];
4602                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4603         }
4604 #endif
4605         BUG_ON(item_size < sizeof(*ei));
4606         ei = btrfs_item_ptr(leaf, extent_slot,
4607                             struct btrfs_extent_item);
4608         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4609                 struct btrfs_tree_block_info *bi;
4610                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4611                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4612                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4613         }
4614
4615         refs = btrfs_extent_refs(leaf, ei);
4616         BUG_ON(refs < refs_to_drop);
4617         refs -= refs_to_drop;
4618
4619         if (refs > 0) {
4620                 if (extent_op)
4621                         __run_delayed_extent_op(extent_op, leaf, ei);
4622                 /*
4623                  * In the case of inline back ref, reference count will
4624                  * be updated by remove_extent_backref
4625                  */
4626                 if (iref) {
4627                         BUG_ON(!found_extent);
4628                 } else {
4629                         btrfs_set_extent_refs(leaf, ei, refs);
4630                         btrfs_mark_buffer_dirty(leaf);
4631                 }
4632                 if (found_extent) {
4633                         ret = remove_extent_backref(trans, extent_root, path,
4634                                                     iref, refs_to_drop,
4635                                                     is_data);
4636                         BUG_ON(ret);
4637                 }
4638         } else {
4639                 if (found_extent) {
4640                         BUG_ON(is_data && refs_to_drop !=
4641                                extent_data_ref_count(root, path, iref));
4642                         if (iref) {
4643                                 BUG_ON(path->slots[0] != extent_slot);
4644                         } else {
4645                                 BUG_ON(path->slots[0] != extent_slot + 1);
4646                                 path->slots[0] = extent_slot;
4647                                 num_to_del = 2;
4648                         }
4649                 }
4650
4651                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4652                                       num_to_del);
4653                 BUG_ON(ret);
4654                 btrfs_release_path(extent_root, path);
4655
4656                 if (is_data) {
4657                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4658                         BUG_ON(ret);
4659                 } else {
4660                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4661                              bytenr >> PAGE_CACHE_SHIFT,
4662                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4663                 }
4664
4665                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4666                 BUG_ON(ret);
4667         }
4668         btrfs_free_path(path);
4669         return ret;
4670 }
4671
4672 /*
4673  * when we free an block, it is possible (and likely) that we free the last
4674  * delayed ref for that extent as well.  This searches the delayed ref tree for
4675  * a given extent, and if there are no other delayed refs to be processed, it
4676  * removes it from the tree.
4677  */
4678 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4679                                       struct btrfs_root *root, u64 bytenr)
4680 {
4681         struct btrfs_delayed_ref_head *head;
4682         struct btrfs_delayed_ref_root *delayed_refs;
4683         struct btrfs_delayed_ref_node *ref;
4684         struct rb_node *node;
4685         int ret = 0;
4686
4687         delayed_refs = &trans->transaction->delayed_refs;
4688         spin_lock(&delayed_refs->lock);
4689         head = btrfs_find_delayed_ref_head(trans, bytenr);
4690         if (!head)
4691                 goto out;
4692
4693         node = rb_prev(&head->node.rb_node);
4694         if (!node)
4695                 goto out;
4696
4697         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4698
4699         /* there are still entries for this ref, we can't drop it */
4700         if (ref->bytenr == bytenr)
4701                 goto out;
4702
4703         if (head->extent_op) {
4704                 if (!head->must_insert_reserved)
4705                         goto out;
4706                 kfree(head->extent_op);
4707                 head->extent_op = NULL;
4708         }
4709
4710         /*
4711          * waiting for the lock here would deadlock.  If someone else has it
4712          * locked they are already in the process of dropping it anyway
4713          */
4714         if (!mutex_trylock(&head->mutex))
4715                 goto out;
4716
4717         /*
4718          * at this point we have a head with no other entries.  Go
4719          * ahead and process it.
4720          */
4721         head->node.in_tree = 0;
4722         rb_erase(&head->node.rb_node, &delayed_refs->root);
4723
4724         delayed_refs->num_entries--;
4725
4726         /*
4727          * we don't take a ref on the node because we're removing it from the
4728          * tree, so we just steal the ref the tree was holding.
4729          */
4730         delayed_refs->num_heads--;
4731         if (list_empty(&head->cluster))
4732                 delayed_refs->num_heads_ready--;
4733
4734         list_del_init(&head->cluster);
4735         spin_unlock(&delayed_refs->lock);
4736
4737         BUG_ON(head->extent_op);
4738         if (head->must_insert_reserved)
4739                 ret = 1;
4740
4741         mutex_unlock(&head->mutex);
4742         btrfs_put_delayed_ref(&head->node);
4743         return ret;
4744 out:
4745         spin_unlock(&delayed_refs->lock);
4746         return 0;
4747 }
4748
4749 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4750                            struct btrfs_root *root,
4751                            struct extent_buffer *buf,
4752                            u64 parent, int last_ref)
4753 {
4754         struct btrfs_block_rsv *block_rsv;
4755         struct btrfs_block_group_cache *cache = NULL;
4756         int ret;
4757
4758         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4759                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4760                                                 parent, root->root_key.objectid,
4761                                                 btrfs_header_level(buf),
4762                                                 BTRFS_DROP_DELAYED_REF, NULL);
4763                 BUG_ON(ret);
4764         }
4765
4766         if (!last_ref)
4767                 return;
4768
4769         block_rsv = get_block_rsv(trans, root);
4770         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4771         if (block_rsv->space_info != cache->space_info)
4772                 goto out;
4773
4774         if (btrfs_header_generation(buf) == trans->transid) {
4775                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4776                         ret = check_ref_cleanup(trans, root, buf->start);
4777                         if (!ret)
4778                                 goto pin;
4779                 }
4780
4781                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4782                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4783                         goto pin;
4784                 }
4785
4786                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4787
4788                 btrfs_add_free_space(cache, buf->start, buf->len);
4789                 ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0);
4790                 if (ret == -EAGAIN) {
4791                         /* block group became read-only */
4792                         btrfs_update_reserved_bytes(cache, buf->len, 0, 1);
4793                         goto out;
4794                 }
4795
4796                 ret = 1;
4797                 spin_lock(&block_rsv->lock);
4798                 if (block_rsv->reserved < block_rsv->size) {
4799                         block_rsv->reserved += buf->len;
4800                         ret = 0;
4801                 }
4802                 spin_unlock(&block_rsv->lock);
4803
4804                 if (ret) {
4805                         spin_lock(&cache->space_info->lock);
4806                         cache->space_info->bytes_reserved -= buf->len;
4807                         cache->space_info->reservation_progress++;
4808                         spin_unlock(&cache->space_info->lock);
4809                 }
4810                 goto out;
4811         }
4812 pin:
4813         if (block_rsv->durable && !cache->ro) {
4814                 ret = 0;
4815                 spin_lock(&cache->lock);
4816                 if (!cache->ro) {
4817                         cache->reserved_pinned += buf->len;
4818                         ret = 1;
4819                 }
4820                 spin_unlock(&cache->lock);
4821
4822                 if (ret) {
4823                         spin_lock(&block_rsv->lock);
4824                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4825                         spin_unlock(&block_rsv->lock);
4826                 }
4827         }
4828 out:
4829         /*
4830          * Deleting the buffer, clear the corrupt flag since it doesn't matter
4831          * anymore.
4832          */
4833         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4834         btrfs_put_block_group(cache);
4835 }
4836
4837 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4838                       struct btrfs_root *root,
4839                       u64 bytenr, u64 num_bytes, u64 parent,
4840                       u64 root_objectid, u64 owner, u64 offset)
4841 {
4842         int ret;
4843
4844         /*
4845          * tree log blocks never actually go into the extent allocation
4846          * tree, just update pinning info and exit early.
4847          */
4848         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4849                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4850                 /* unlocks the pinned mutex */
4851                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4852                 ret = 0;
4853         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4854                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4855                                         parent, root_objectid, (int)owner,
4856                                         BTRFS_DROP_DELAYED_REF, NULL);
4857                 BUG_ON(ret);
4858         } else {
4859                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4860                                         parent, root_objectid, owner,
4861                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4862                 BUG_ON(ret);
4863         }
4864         return ret;
4865 }
4866
4867 static u64 stripe_align(struct btrfs_root *root, u64 val)
4868 {
4869         u64 mask = ((u64)root->stripesize - 1);
4870         u64 ret = (val + mask) & ~mask;
4871         return ret;
4872 }
4873
4874 /*
4875  * when we wait for progress in the block group caching, its because
4876  * our allocation attempt failed at least once.  So, we must sleep
4877  * and let some progress happen before we try again.
4878  *
4879  * This function will sleep at least once waiting for new free space to
4880  * show up, and then it will check the block group free space numbers
4881  * for our min num_bytes.  Another option is to have it go ahead
4882  * and look in the rbtree for a free extent of a given size, but this
4883  * is a good start.
4884  */
4885 static noinline int
4886 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4887                                 u64 num_bytes)
4888 {
4889         struct btrfs_caching_control *caching_ctl;
4890         DEFINE_WAIT(wait);
4891
4892         caching_ctl = get_caching_control(cache);
4893         if (!caching_ctl)
4894                 return 0;
4895
4896         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4897                    (cache->free_space >= num_bytes));
4898
4899         put_caching_control(caching_ctl);
4900         return 0;
4901 }
4902
4903 static noinline int
4904 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4905 {
4906         struct btrfs_caching_control *caching_ctl;
4907         DEFINE_WAIT(wait);
4908
4909         caching_ctl = get_caching_control(cache);
4910         if (!caching_ctl)
4911                 return 0;
4912
4913         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4914
4915         put_caching_control(caching_ctl);
4916         return 0;
4917 }
4918
4919 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4920 {
4921         int index;
4922         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4923                 index = 0;
4924         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4925                 index = 1;
4926         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4927                 index = 2;
4928         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4929                 index = 3;
4930         else
4931                 index = 4;
4932         return index;
4933 }
4934
4935 enum btrfs_loop_type {
4936         LOOP_FIND_IDEAL = 0,
4937         LOOP_CACHING_NOWAIT = 1,
4938         LOOP_CACHING_WAIT = 2,
4939         LOOP_ALLOC_CHUNK = 3,
4940         LOOP_NO_EMPTY_SIZE = 4,
4941 };
4942
4943 /*
4944  * walks the btree of allocated extents and find a hole of a given size.
4945  * The key ins is changed to record the hole:
4946  * ins->objectid == block start
4947  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4948  * ins->offset == number of blocks
4949  * Any available blocks before search_start are skipped.
4950  */
4951 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4952                                      struct btrfs_root *orig_root,
4953                                      u64 num_bytes, u64 empty_size,
4954                                      u64 search_start, u64 search_end,
4955                                      u64 hint_byte, struct btrfs_key *ins,
4956                                      int data)
4957 {
4958         int ret = 0;
4959         struct btrfs_root *root = orig_root->fs_info->extent_root;
4960         struct btrfs_free_cluster *last_ptr = NULL;
4961         struct btrfs_block_group_cache *block_group = NULL;
4962         int empty_cluster = 2 * 1024 * 1024;
4963         int allowed_chunk_alloc = 0;
4964         int done_chunk_alloc = 0;
4965         struct btrfs_space_info *space_info;
4966         int last_ptr_loop = 0;
4967         int loop = 0;
4968         int index = 0;
4969         bool found_uncached_bg = false;
4970         bool failed_cluster_refill = false;
4971         bool failed_alloc = false;
4972         bool use_cluster = true;
4973         u64 ideal_cache_percent = 0;
4974         u64 ideal_cache_offset = 0;
4975
4976         WARN_ON(num_bytes < root->sectorsize);
4977         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4978         ins->objectid = 0;
4979         ins->offset = 0;
4980
4981         space_info = __find_space_info(root->fs_info, data);
4982         if (!space_info) {
4983                 printk(KERN_ERR "No space info for %d\n", data);
4984                 return -ENOSPC;
4985         }
4986
4987         /*
4988          * If the space info is for both data and metadata it means we have a
4989          * small filesystem and we can't use the clustering stuff.
4990          */
4991         if (btrfs_mixed_space_info(space_info))
4992                 use_cluster = false;
4993
4994         if (orig_root->ref_cows || empty_size)
4995                 allowed_chunk_alloc = 1;
4996
4997         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4998                 last_ptr = &root->fs_info->meta_alloc_cluster;
4999                 if (!btrfs_test_opt(root, SSD))
5000                         empty_cluster = 64 * 1024;
5001         }
5002
5003         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5004             btrfs_test_opt(root, SSD)) {
5005                 last_ptr = &root->fs_info->data_alloc_cluster;
5006         }
5007
5008         if (last_ptr) {
5009                 spin_lock(&last_ptr->lock);
5010                 if (last_ptr->block_group)
5011                         hint_byte = last_ptr->window_start;
5012                 spin_unlock(&last_ptr->lock);
5013         }
5014
5015         search_start = max(search_start, first_logical_byte(root, 0));
5016         search_start = max(search_start, hint_byte);
5017
5018         if (!last_ptr)
5019                 empty_cluster = 0;
5020
5021         if (search_start == hint_byte) {
5022 ideal_cache:
5023                 block_group = btrfs_lookup_block_group(root->fs_info,
5024                                                        search_start);
5025                 /*
5026                  * we don't want to use the block group if it doesn't match our
5027                  * allocation bits, or if its not cached.
5028                  *
5029                  * However if we are re-searching with an ideal block group
5030                  * picked out then we don't care that the block group is cached.
5031                  */
5032                 if (block_group && block_group_bits(block_group, data) &&
5033                     (block_group->cached != BTRFS_CACHE_NO ||
5034                      search_start == ideal_cache_offset)) {
5035                         down_read(&space_info->groups_sem);
5036                         if (list_empty(&block_group->list) ||
5037                             block_group->ro) {
5038                                 /*
5039                                  * someone is removing this block group,
5040                                  * we can't jump into the have_block_group
5041                                  * target because our list pointers are not
5042                                  * valid
5043                                  */
5044                                 btrfs_put_block_group(block_group);
5045                                 up_read(&space_info->groups_sem);
5046                         } else {
5047                                 index = get_block_group_index(block_group);
5048                                 goto have_block_group;
5049                         }
5050                 } else if (block_group) {
5051                         btrfs_put_block_group(block_group);
5052                 }
5053         }
5054 search:
5055         down_read(&space_info->groups_sem);
5056         list_for_each_entry(block_group, &space_info->block_groups[index],
5057                             list) {
5058                 u64 offset;
5059                 int cached;
5060
5061                 btrfs_get_block_group(block_group);
5062                 search_start = block_group->key.objectid;
5063
5064                 /*
5065                  * this can happen if we end up cycling through all the
5066                  * raid types, but we want to make sure we only allocate
5067                  * for the proper type.
5068                  */
5069                 if (!block_group_bits(block_group, data)) {
5070                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
5071                                 BTRFS_BLOCK_GROUP_RAID1 |
5072                                 BTRFS_BLOCK_GROUP_RAID10;
5073
5074                         /*
5075                          * if they asked for extra copies and this block group
5076                          * doesn't provide them, bail.  This does allow us to
5077                          * fill raid0 from raid1.
5078                          */
5079                         if ((data & extra) && !(block_group->flags & extra))
5080                                 goto loop;
5081                 }
5082
5083 have_block_group:
5084                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5085                         u64 free_percent;
5086
5087                         ret = cache_block_group(block_group, trans,
5088                                                 orig_root, 1);
5089                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5090                                 goto have_block_group;
5091
5092                         free_percent = btrfs_block_group_used(&block_group->item);
5093                         free_percent *= 100;
5094                         free_percent = div64_u64(free_percent,
5095                                                  block_group->key.offset);
5096                         free_percent = 100 - free_percent;
5097                         if (free_percent > ideal_cache_percent &&
5098                             likely(!block_group->ro)) {
5099                                 ideal_cache_offset = block_group->key.objectid;
5100                                 ideal_cache_percent = free_percent;
5101                         }
5102
5103                         /*
5104                          * We only want to start kthread caching if we are at
5105                          * the point where we will wait for caching to make
5106                          * progress, or if our ideal search is over and we've
5107                          * found somebody to start caching.
5108                          */
5109                         if (loop > LOOP_CACHING_NOWAIT ||
5110                             (loop > LOOP_FIND_IDEAL &&
5111                              atomic_read(&space_info->caching_threads) < 2)) {
5112                                 ret = cache_block_group(block_group, trans,
5113                                                         orig_root, 0);
5114                                 BUG_ON(ret);
5115                         }
5116                         found_uncached_bg = true;
5117
5118                         /*
5119                          * If loop is set for cached only, try the next block
5120                          * group.
5121                          */
5122                         if (loop == LOOP_FIND_IDEAL)
5123                                 goto loop;
5124                 }
5125
5126                 cached = block_group_cache_done(block_group);
5127                 if (unlikely(!cached))
5128                         found_uncached_bg = true;
5129
5130                 if (unlikely(block_group->ro))
5131                         goto loop;
5132
5133                 /*
5134                  * Ok we want to try and use the cluster allocator, so lets look
5135                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5136                  * have tried the cluster allocator plenty of times at this
5137                  * point and not have found anything, so we are likely way too
5138                  * fragmented for the clustering stuff to find anything, so lets
5139                  * just skip it and let the allocator find whatever block it can
5140                  * find
5141                  */
5142                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5143                         /*
5144                          * the refill lock keeps out other
5145                          * people trying to start a new cluster
5146                          */
5147                         spin_lock(&last_ptr->refill_lock);
5148                         if (last_ptr->block_group &&
5149                             (last_ptr->block_group->ro ||
5150                             !block_group_bits(last_ptr->block_group, data))) {
5151                                 offset = 0;
5152                                 goto refill_cluster;
5153                         }
5154
5155                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5156                                                  num_bytes, search_start);
5157                         if (offset) {
5158                                 /* we have a block, we're done */
5159                                 spin_unlock(&last_ptr->refill_lock);
5160                                 goto checks;
5161                         }
5162
5163                         spin_lock(&last_ptr->lock);
5164                         /*
5165                          * whoops, this cluster doesn't actually point to
5166                          * this block group.  Get a ref on the block
5167                          * group is does point to and try again
5168                          */
5169                         if (!last_ptr_loop && last_ptr->block_group &&
5170                             last_ptr->block_group != block_group) {
5171
5172                                 btrfs_put_block_group(block_group);
5173                                 block_group = last_ptr->block_group;
5174                                 btrfs_get_block_group(block_group);
5175                                 spin_unlock(&last_ptr->lock);
5176                                 spin_unlock(&last_ptr->refill_lock);
5177
5178                                 last_ptr_loop = 1;
5179                                 search_start = block_group->key.objectid;
5180                                 /*
5181                                  * we know this block group is properly
5182                                  * in the list because
5183                                  * btrfs_remove_block_group, drops the
5184                                  * cluster before it removes the block
5185                                  * group from the list
5186                                  */
5187                                 goto have_block_group;
5188                         }
5189                         spin_unlock(&last_ptr->lock);
5190 refill_cluster:
5191                         /*
5192                          * this cluster didn't work out, free it and
5193                          * start over
5194                          */
5195                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5196
5197                         last_ptr_loop = 0;
5198
5199                         /* allocate a cluster in this block group */
5200                         ret = btrfs_find_space_cluster(trans, root,
5201                                                block_group, last_ptr,
5202                                                offset, num_bytes,
5203                                                empty_cluster + empty_size);
5204                         if (ret == 0) {
5205                                 /*
5206                                  * now pull our allocation out of this
5207                                  * cluster
5208                                  */
5209                                 offset = btrfs_alloc_from_cluster(block_group,
5210                                                   last_ptr, num_bytes,
5211                                                   search_start);
5212                                 if (offset) {
5213                                         /* we found one, proceed */
5214                                         spin_unlock(&last_ptr->refill_lock);
5215                                         goto checks;
5216                                 }
5217                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5218                                    && !failed_cluster_refill) {
5219                                 spin_unlock(&last_ptr->refill_lock);
5220
5221                                 failed_cluster_refill = true;
5222                                 wait_block_group_cache_progress(block_group,
5223                                        num_bytes + empty_cluster + empty_size);
5224                                 goto have_block_group;
5225                         }
5226
5227                         /*
5228                          * at this point we either didn't find a cluster
5229                          * or we weren't able to allocate a block from our
5230                          * cluster.  Free the cluster we've been trying
5231                          * to use, and go to the next block group
5232                          */
5233                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5234                         spin_unlock(&last_ptr->refill_lock);
5235                         goto loop;
5236                 }
5237
5238                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5239                                                     num_bytes, empty_size);
5240                 /*
5241                  * If we didn't find a chunk, and we haven't failed on this
5242                  * block group before, and this block group is in the middle of
5243                  * caching and we are ok with waiting, then go ahead and wait
5244                  * for progress to be made, and set failed_alloc to true.
5245                  *
5246                  * If failed_alloc is true then we've already waited on this
5247                  * block group once and should move on to the next block group.
5248                  */
5249                 if (!offset && !failed_alloc && !cached &&
5250                     loop > LOOP_CACHING_NOWAIT) {
5251                         wait_block_group_cache_progress(block_group,
5252                                                 num_bytes + empty_size);
5253                         failed_alloc = true;
5254                         goto have_block_group;
5255                 } else if (!offset) {
5256                         goto loop;
5257                 }
5258 checks:
5259                 search_start = stripe_align(root, offset);
5260                 /* move on to the next group */
5261                 if (search_start + num_bytes >= search_end) {
5262                         btrfs_add_free_space(block_group, offset, num_bytes);
5263                         goto loop;
5264                 }
5265
5266                 /* move on to the next group */
5267                 if (search_start + num_bytes >
5268                     block_group->key.objectid + block_group->key.offset) {
5269                         btrfs_add_free_space(block_group, offset, num_bytes);
5270                         goto loop;
5271                 }
5272
5273                 ins->objectid = search_start;
5274                 ins->offset = num_bytes;
5275
5276                 if (offset < search_start)
5277                         btrfs_add_free_space(block_group, offset,
5278                                              search_start - offset);
5279                 BUG_ON(offset > search_start);
5280
5281                 ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1,
5282                                             (data & BTRFS_BLOCK_GROUP_DATA));
5283                 if (ret == -EAGAIN) {
5284                         btrfs_add_free_space(block_group, offset, num_bytes);
5285                         goto loop;
5286                 }
5287
5288                 /* we are all good, lets return */
5289                 ins->objectid = search_start;
5290                 ins->offset = num_bytes;
5291
5292                 if (offset < search_start)
5293                         btrfs_add_free_space(block_group, offset,
5294                                              search_start - offset);
5295                 BUG_ON(offset > search_start);
5296                 break;
5297 loop:
5298                 failed_cluster_refill = false;
5299                 failed_alloc = false;
5300                 BUG_ON(index != get_block_group_index(block_group));
5301                 btrfs_put_block_group(block_group);
5302         }
5303         up_read(&space_info->groups_sem);
5304
5305         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5306                 goto search;
5307
5308         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5309          *                      for them to make caching progress.  Also
5310          *                      determine the best possible bg to cache
5311          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5312          *                      caching kthreads as we move along
5313          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5314          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5315          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5316          *                      again
5317          */
5318         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5319             (found_uncached_bg || empty_size || empty_cluster ||
5320              allowed_chunk_alloc)) {
5321                 index = 0;
5322                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5323                         found_uncached_bg = false;
5324                         loop++;
5325                         if (!ideal_cache_percent &&
5326                             atomic_read(&space_info->caching_threads))
5327                                 goto search;
5328
5329                         /*
5330                          * 1 of the following 2 things have happened so far
5331                          *
5332                          * 1) We found an ideal block group for caching that
5333                          * is mostly full and will cache quickly, so we might
5334                          * as well wait for it.
5335                          *
5336                          * 2) We searched for cached only and we didn't find
5337                          * anything, and we didn't start any caching kthreads
5338                          * either, so chances are we will loop through and
5339                          * start a couple caching kthreads, and then come back
5340                          * around and just wait for them.  This will be slower
5341                          * because we will have 2 caching kthreads reading at
5342                          * the same time when we could have just started one
5343                          * and waited for it to get far enough to give us an
5344                          * allocation, so go ahead and go to the wait caching
5345                          * loop.
5346                          */
5347                         loop = LOOP_CACHING_WAIT;
5348                         search_start = ideal_cache_offset;
5349                         ideal_cache_percent = 0;
5350                         goto ideal_cache;
5351                 } else if (loop == LOOP_FIND_IDEAL) {
5352                         /*
5353                          * Didn't find a uncached bg, wait on anything we find
5354                          * next.
5355                          */
5356                         loop = LOOP_CACHING_WAIT;
5357                         goto search;
5358                 }
5359
5360                 if (loop < LOOP_CACHING_WAIT) {
5361                         loop++;
5362                         goto search;
5363                 }
5364
5365                 if (loop == LOOP_ALLOC_CHUNK) {
5366                         empty_size = 0;
5367                         empty_cluster = 0;
5368                 }
5369
5370                 if (allowed_chunk_alloc) {
5371                         ret = do_chunk_alloc(trans, root, num_bytes +
5372                                              2 * 1024 * 1024, data,
5373                                              CHUNK_ALLOC_LIMITED);
5374                         allowed_chunk_alloc = 0;
5375                         done_chunk_alloc = 1;
5376                 } else if (!done_chunk_alloc &&
5377                            space_info->force_alloc == CHUNK_ALLOC_NO_FORCE) {
5378                         space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5379                 }
5380
5381                 if (loop < LOOP_NO_EMPTY_SIZE) {
5382                         loop++;
5383                         goto search;
5384                 }
5385                 ret = -ENOSPC;
5386         } else if (!ins->objectid) {
5387                 ret = -ENOSPC;
5388         }
5389
5390         /* we found what we needed */
5391         if (ins->objectid) {
5392                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5393                         trans->block_group = block_group->key.objectid;
5394
5395                 btrfs_put_block_group(block_group);
5396                 ret = 0;
5397         }
5398
5399         return ret;
5400 }
5401
5402 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5403                             int dump_block_groups)
5404 {
5405         struct btrfs_block_group_cache *cache;
5406         int index = 0;
5407
5408         spin_lock(&info->lock);
5409         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5410                (unsigned long long)(info->total_bytes - info->bytes_used -
5411                                     info->bytes_pinned - info->bytes_reserved -
5412                                     info->bytes_readonly),
5413                (info->full) ? "" : "not ");
5414         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5415                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5416                (unsigned long long)info->total_bytes,
5417                (unsigned long long)info->bytes_used,
5418                (unsigned long long)info->bytes_pinned,
5419                (unsigned long long)info->bytes_reserved,
5420                (unsigned long long)info->bytes_may_use,
5421                (unsigned long long)info->bytes_readonly);
5422         spin_unlock(&info->lock);
5423
5424         if (!dump_block_groups)
5425                 return;
5426
5427         down_read(&info->groups_sem);
5428 again:
5429         list_for_each_entry(cache, &info->block_groups[index], list) {
5430                 spin_lock(&cache->lock);
5431                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5432                        "%llu pinned %llu reserved\n",
5433                        (unsigned long long)cache->key.objectid,
5434                        (unsigned long long)cache->key.offset,
5435                        (unsigned long long)btrfs_block_group_used(&cache->item),
5436                        (unsigned long long)cache->pinned,
5437                        (unsigned long long)cache->reserved);
5438                 btrfs_dump_free_space(cache, bytes);
5439                 spin_unlock(&cache->lock);
5440         }
5441         if (++index < BTRFS_NR_RAID_TYPES)
5442                 goto again;
5443         up_read(&info->groups_sem);
5444 }
5445
5446 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5447                          struct btrfs_root *root,
5448                          u64 num_bytes, u64 min_alloc_size,
5449                          u64 empty_size, u64 hint_byte,
5450                          u64 search_end, struct btrfs_key *ins,
5451                          u64 data)
5452 {
5453         int ret;
5454         u64 search_start = 0;
5455
5456         data = btrfs_get_alloc_profile(root, data);
5457 again:
5458         /*
5459          * the only place that sets empty_size is btrfs_realloc_node, which
5460          * is not called recursively on allocations
5461          */
5462         if (empty_size || root->ref_cows)
5463                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5464                                      num_bytes + 2 * 1024 * 1024, data,
5465                                      CHUNK_ALLOC_NO_FORCE);
5466
5467         WARN_ON(num_bytes < root->sectorsize);
5468         ret = find_free_extent(trans, root, num_bytes, empty_size,
5469                                search_start, search_end, hint_byte,
5470                                ins, data);
5471
5472         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5473                 num_bytes = num_bytes >> 1;
5474                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5475                 num_bytes = max(num_bytes, min_alloc_size);
5476                 do_chunk_alloc(trans, root->fs_info->extent_root,
5477                                num_bytes, data, CHUNK_ALLOC_FORCE);
5478                 goto again;
5479         }
5480         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5481                 struct btrfs_space_info *sinfo;
5482
5483                 sinfo = __find_space_info(root->fs_info, data);
5484                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5485                        "wanted %llu\n", (unsigned long long)data,
5486                        (unsigned long long)num_bytes);
5487                 dump_space_info(sinfo, num_bytes, 1);
5488         }
5489
5490         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5491
5492         return ret;
5493 }
5494
5495 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5496 {
5497         struct btrfs_block_group_cache *cache;
5498         int ret = 0;
5499
5500         cache = btrfs_lookup_block_group(root->fs_info, start);
5501         if (!cache) {
5502                 printk(KERN_ERR "Unable to find block group for %llu\n",
5503                        (unsigned long long)start);
5504                 return -ENOSPC;
5505         }
5506
5507         if (btrfs_test_opt(root, DISCARD))
5508                 ret = btrfs_discard_extent(root, start, len, NULL);
5509
5510         btrfs_add_free_space(cache, start, len);
5511         btrfs_update_reserved_bytes(cache, len, 0, 1);
5512         btrfs_put_block_group(cache);
5513
5514         trace_btrfs_reserved_extent_free(root, start, len);
5515
5516         return ret;
5517 }
5518
5519 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5520                                       struct btrfs_root *root,
5521                                       u64 parent, u64 root_objectid,
5522                                       u64 flags, u64 owner, u64 offset,
5523                                       struct btrfs_key *ins, int ref_mod)
5524 {
5525         int ret;
5526         struct btrfs_fs_info *fs_info = root->fs_info;
5527         struct btrfs_extent_item *extent_item;
5528         struct btrfs_extent_inline_ref *iref;
5529         struct btrfs_path *path;
5530         struct extent_buffer *leaf;
5531         int type;
5532         u32 size;
5533
5534         if (parent > 0)
5535                 type = BTRFS_SHARED_DATA_REF_KEY;
5536         else
5537                 type = BTRFS_EXTENT_DATA_REF_KEY;
5538
5539         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5540
5541         path = btrfs_alloc_path();
5542         if (!path)
5543                 return -ENOMEM;
5544
5545         path->leave_spinning = 1;
5546         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5547                                       ins, size);
5548         BUG_ON(ret);
5549
5550         leaf = path->nodes[0];
5551         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5552                                      struct btrfs_extent_item);
5553         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5554         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5555         btrfs_set_extent_flags(leaf, extent_item,
5556                                flags | BTRFS_EXTENT_FLAG_DATA);
5557
5558         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5559         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5560         if (parent > 0) {
5561                 struct btrfs_shared_data_ref *ref;
5562                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5563                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5564                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5565         } else {
5566                 struct btrfs_extent_data_ref *ref;
5567                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5568                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5569                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5570                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5571                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5572         }
5573
5574         btrfs_mark_buffer_dirty(path->nodes[0]);
5575         btrfs_free_path(path);
5576
5577         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5578         if (ret) {
5579                 printk(KERN_ERR "btrfs update block group failed for %llu "
5580                        "%llu\n", (unsigned long long)ins->objectid,
5581                        (unsigned long long)ins->offset);
5582                 BUG();
5583         }
5584         return ret;
5585 }
5586
5587 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5588                                      struct btrfs_root *root,
5589                                      u64 parent, u64 root_objectid,
5590                                      u64 flags, struct btrfs_disk_key *key,
5591                                      int level, struct btrfs_key *ins)
5592 {
5593         int ret;
5594         struct btrfs_fs_info *fs_info = root->fs_info;
5595         struct btrfs_extent_item *extent_item;
5596         struct btrfs_tree_block_info *block_info;
5597         struct btrfs_extent_inline_ref *iref;
5598         struct btrfs_path *path;
5599         struct extent_buffer *leaf;
5600         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5601
5602         path = btrfs_alloc_path();
5603         BUG_ON(!path);
5604
5605         path->leave_spinning = 1;
5606         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5607                                       ins, size);
5608         BUG_ON(ret);
5609
5610         leaf = path->nodes[0];
5611         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5612                                      struct btrfs_extent_item);
5613         btrfs_set_extent_refs(leaf, extent_item, 1);
5614         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5615         btrfs_set_extent_flags(leaf, extent_item,
5616                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5617         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5618
5619         btrfs_set_tree_block_key(leaf, block_info, key);
5620         btrfs_set_tree_block_level(leaf, block_info, level);
5621
5622         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5623         if (parent > 0) {
5624                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5625                 btrfs_set_extent_inline_ref_type(leaf, iref,
5626                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5627                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5628         } else {
5629                 btrfs_set_extent_inline_ref_type(leaf, iref,
5630                                                  BTRFS_TREE_BLOCK_REF_KEY);
5631                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5632         }
5633
5634         btrfs_mark_buffer_dirty(leaf);
5635         btrfs_free_path(path);
5636
5637         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5638         if (ret) {
5639                 printk(KERN_ERR "btrfs update block group failed for %llu "
5640                        "%llu\n", (unsigned long long)ins->objectid,
5641                        (unsigned long long)ins->offset);
5642                 BUG();
5643         }
5644         return ret;
5645 }
5646
5647 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5648                                      struct btrfs_root *root,
5649                                      u64 root_objectid, u64 owner,
5650                                      u64 offset, struct btrfs_key *ins)
5651 {
5652         int ret;
5653
5654         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5655
5656         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5657                                          0, root_objectid, owner, offset,
5658                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5659         return ret;
5660 }
5661
5662 /*
5663  * this is used by the tree logging recovery code.  It records that
5664  * an extent has been allocated and makes sure to clear the free
5665  * space cache bits as well
5666  */
5667 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5668                                    struct btrfs_root *root,
5669                                    u64 root_objectid, u64 owner, u64 offset,
5670                                    struct btrfs_key *ins)
5671 {
5672         int ret;
5673         struct btrfs_block_group_cache *block_group;
5674         struct btrfs_caching_control *caching_ctl;
5675         u64 start = ins->objectid;
5676         u64 num_bytes = ins->offset;
5677
5678         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5679         cache_block_group(block_group, trans, NULL, 0);
5680         caching_ctl = get_caching_control(block_group);
5681
5682         if (!caching_ctl) {
5683                 BUG_ON(!block_group_cache_done(block_group));
5684                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5685                 BUG_ON(ret);
5686         } else {
5687                 mutex_lock(&caching_ctl->mutex);
5688
5689                 if (start >= caching_ctl->progress) {
5690                         ret = add_excluded_extent(root, start, num_bytes);
5691                         BUG_ON(ret);
5692                 } else if (start + num_bytes <= caching_ctl->progress) {
5693                         ret = btrfs_remove_free_space(block_group,
5694                                                       start, num_bytes);
5695                         BUG_ON(ret);
5696                 } else {
5697                         num_bytes = caching_ctl->progress - start;
5698                         ret = btrfs_remove_free_space(block_group,
5699                                                       start, num_bytes);
5700                         BUG_ON(ret);
5701
5702                         start = caching_ctl->progress;
5703                         num_bytes = ins->objectid + ins->offset -
5704                                     caching_ctl->progress;
5705                         ret = add_excluded_extent(root, start, num_bytes);
5706                         BUG_ON(ret);
5707                 }
5708
5709                 mutex_unlock(&caching_ctl->mutex);
5710                 put_caching_control(caching_ctl);
5711         }
5712
5713         ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1);
5714         BUG_ON(ret);
5715         btrfs_put_block_group(block_group);
5716         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5717                                          0, owner, offset, ins, 1);
5718         return ret;
5719 }
5720
5721 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5722                                             struct btrfs_root *root,
5723                                             u64 bytenr, u32 blocksize,
5724                                             int level)
5725 {
5726         struct extent_buffer *buf;
5727
5728         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5729         if (!buf)
5730                 return ERR_PTR(-ENOMEM);
5731         btrfs_set_header_generation(buf, trans->transid);
5732         btrfs_set_buffer_lockdep_class(buf, level);
5733         btrfs_tree_lock(buf);
5734         clean_tree_block(trans, root, buf);
5735
5736         btrfs_set_lock_blocking(buf);
5737         btrfs_set_buffer_uptodate(buf);
5738
5739         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5740                 /*
5741                  * we allow two log transactions at a time, use different
5742                  * EXENT bit to differentiate dirty pages.
5743                  */
5744                 if (root->log_transid % 2 == 0)
5745                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5746                                         buf->start + buf->len - 1, GFP_NOFS);
5747                 else
5748                         set_extent_new(&root->dirty_log_pages, buf->start,
5749                                         buf->start + buf->len - 1, GFP_NOFS);
5750         } else {
5751                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5752                          buf->start + buf->len - 1, GFP_NOFS);
5753         }
5754         trans->blocks_used++;
5755         /* this returns a buffer locked for blocking */
5756         return buf;
5757 }
5758
5759 static struct btrfs_block_rsv *
5760 use_block_rsv(struct btrfs_trans_handle *trans,
5761               struct btrfs_root *root, u32 blocksize)
5762 {
5763         struct btrfs_block_rsv *block_rsv;
5764         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5765         int ret;
5766
5767         block_rsv = get_block_rsv(trans, root);
5768
5769         if (block_rsv->size == 0) {
5770                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5771                                              blocksize, 0);
5772                 /*
5773                  * If we couldn't reserve metadata bytes try and use some from
5774                  * the global reserve.
5775                  */
5776                 if (ret && block_rsv != global_rsv) {
5777                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5778                         if (!ret)
5779                                 return global_rsv;
5780                         return ERR_PTR(ret);
5781                 } else if (ret) {
5782                         return ERR_PTR(ret);
5783                 }
5784                 return block_rsv;
5785         }
5786
5787         ret = block_rsv_use_bytes(block_rsv, blocksize);
5788         if (!ret)
5789                 return block_rsv;
5790         if (ret) {
5791                 WARN_ON(1);
5792                 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5793                                              0);
5794                 if (!ret) {
5795                         spin_lock(&block_rsv->lock);
5796                         block_rsv->size += blocksize;
5797                         spin_unlock(&block_rsv->lock);
5798                         return block_rsv;
5799                 } else if (ret && block_rsv != global_rsv) {
5800                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5801                         if (!ret)
5802                                 return global_rsv;
5803                 }
5804         }
5805
5806         return ERR_PTR(-ENOSPC);
5807 }
5808
5809 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5810 {
5811         block_rsv_add_bytes(block_rsv, blocksize, 0);
5812         block_rsv_release_bytes(block_rsv, NULL, 0);
5813 }
5814
5815 /*
5816  * finds a free extent and does all the dirty work required for allocation
5817  * returns the key for the extent through ins, and a tree buffer for
5818  * the first block of the extent through buf.
5819  *
5820  * returns the tree buffer or NULL.
5821  */
5822 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5823                                         struct btrfs_root *root, u32 blocksize,
5824                                         u64 parent, u64 root_objectid,
5825                                         struct btrfs_disk_key *key, int level,
5826                                         u64 hint, u64 empty_size)
5827 {
5828         struct btrfs_key ins;
5829         struct btrfs_block_rsv *block_rsv;
5830         struct extent_buffer *buf;
5831         u64 flags = 0;
5832         int ret;
5833
5834
5835         block_rsv = use_block_rsv(trans, root, blocksize);
5836         if (IS_ERR(block_rsv))
5837                 return ERR_CAST(block_rsv);
5838
5839         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5840                                    empty_size, hint, (u64)-1, &ins, 0);
5841         if (ret) {
5842                 unuse_block_rsv(block_rsv, blocksize);
5843                 return ERR_PTR(ret);
5844         }
5845
5846         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5847                                     blocksize, level);
5848         BUG_ON(IS_ERR(buf));
5849
5850         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5851                 if (parent == 0)
5852                         parent = ins.objectid;
5853                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5854         } else
5855                 BUG_ON(parent > 0);
5856
5857         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5858                 struct btrfs_delayed_extent_op *extent_op;
5859                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5860                 BUG_ON(!extent_op);
5861                 if (key)
5862                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5863                 else
5864                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5865                 extent_op->flags_to_set = flags;
5866                 extent_op->update_key = 1;
5867                 extent_op->update_flags = 1;
5868                 extent_op->is_data = 0;
5869
5870                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5871                                         ins.offset, parent, root_objectid,
5872                                         level, BTRFS_ADD_DELAYED_EXTENT,
5873                                         extent_op);
5874                 BUG_ON(ret);
5875         }
5876         return buf;
5877 }
5878
5879 struct walk_control {
5880         u64 refs[BTRFS_MAX_LEVEL];
5881         u64 flags[BTRFS_MAX_LEVEL];
5882         struct btrfs_key update_progress;
5883         int stage;
5884         int level;
5885         int shared_level;
5886         int update_ref;
5887         int keep_locks;
5888         int reada_slot;
5889         int reada_count;
5890 };
5891
5892 #define DROP_REFERENCE  1
5893 #define UPDATE_BACKREF  2
5894
5895 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5896                                      struct btrfs_root *root,
5897                                      struct walk_control *wc,
5898                                      struct btrfs_path *path)
5899 {
5900         u64 bytenr;
5901         u64 generation;
5902         u64 refs;
5903         u64 flags;
5904         u32 nritems;
5905         u32 blocksize;
5906         struct btrfs_key key;
5907         struct extent_buffer *eb;
5908         int ret;
5909         int slot;
5910         int nread = 0;
5911
5912         if (path->slots[wc->level] < wc->reada_slot) {
5913                 wc->reada_count = wc->reada_count * 2 / 3;
5914                 wc->reada_count = max(wc->reada_count, 2);
5915         } else {
5916                 wc->reada_count = wc->reada_count * 3 / 2;
5917                 wc->reada_count = min_t(int, wc->reada_count,
5918                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5919         }
5920
5921         eb = path->nodes[wc->level];
5922         nritems = btrfs_header_nritems(eb);
5923         blocksize = btrfs_level_size(root, wc->level - 1);
5924
5925         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5926                 if (nread >= wc->reada_count)
5927                         break;
5928
5929                 cond_resched();
5930                 bytenr = btrfs_node_blockptr(eb, slot);
5931                 generation = btrfs_node_ptr_generation(eb, slot);
5932
5933                 if (slot == path->slots[wc->level])
5934                         goto reada;
5935
5936                 if (wc->stage == UPDATE_BACKREF &&
5937                     generation <= root->root_key.offset)
5938                         continue;
5939
5940                 /* We don't lock the tree block, it's OK to be racy here */
5941                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5942                                                &refs, &flags);
5943                 BUG_ON(ret);
5944                 BUG_ON(refs == 0);
5945
5946                 if (wc->stage == DROP_REFERENCE) {
5947                         if (refs == 1)
5948                                 goto reada;
5949
5950                         if (wc->level == 1 &&
5951                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5952                                 continue;
5953                         if (!wc->update_ref ||
5954                             generation <= root->root_key.offset)
5955                                 continue;
5956                         btrfs_node_key_to_cpu(eb, &key, slot);
5957                         ret = btrfs_comp_cpu_keys(&key,
5958                                                   &wc->update_progress);
5959                         if (ret < 0)
5960                                 continue;
5961                 } else {
5962                         if (wc->level == 1 &&
5963                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5964                                 continue;
5965                 }
5966 reada:
5967                 ret = readahead_tree_block(root, bytenr, blocksize,
5968                                            generation);
5969                 if (ret)
5970                         break;
5971                 nread++;
5972         }
5973         wc->reada_slot = slot;
5974 }
5975
5976 /*
5977  * hepler to process tree block while walking down the tree.
5978  *
5979  * when wc->stage == UPDATE_BACKREF, this function updates
5980  * back refs for pointers in the block.
5981  *
5982  * NOTE: return value 1 means we should stop walking down.
5983  */
5984 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5985                                    struct btrfs_root *root,
5986                                    struct btrfs_path *path,
5987                                    struct walk_control *wc, int lookup_info)
5988 {
5989         int level = wc->level;
5990         struct extent_buffer *eb = path->nodes[level];
5991         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5992         int ret;
5993
5994         if (wc->stage == UPDATE_BACKREF &&
5995             btrfs_header_owner(eb) != root->root_key.objectid)
5996                 return 1;
5997
5998         /*
5999          * when reference count of tree block is 1, it won't increase
6000          * again. once full backref flag is set, we never clear it.
6001          */
6002         if (lookup_info &&
6003             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6004              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6005                 BUG_ON(!path->locks[level]);
6006                 ret = btrfs_lookup_extent_info(trans, root,
6007                                                eb->start, eb->len,
6008                                                &wc->refs[level],
6009                                                &wc->flags[level]);
6010                 BUG_ON(ret);
6011                 BUG_ON(wc->refs[level] == 0);
6012         }
6013
6014         if (wc->stage == DROP_REFERENCE) {
6015                 if (wc->refs[level] > 1)
6016                         return 1;
6017
6018                 if (path->locks[level] && !wc->keep_locks) {
6019                         btrfs_tree_unlock(eb);
6020                         path->locks[level] = 0;
6021                 }
6022                 return 0;
6023         }
6024
6025         /* wc->stage == UPDATE_BACKREF */
6026         if (!(wc->flags[level] & flag)) {
6027                 BUG_ON(!path->locks[level]);
6028                 ret = btrfs_inc_ref(trans, root, eb, 1);
6029                 BUG_ON(ret);
6030                 ret = btrfs_dec_ref(trans, root, eb, 0);
6031                 BUG_ON(ret);
6032                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6033                                                   eb->len, flag, 0);
6034                 BUG_ON(ret);
6035                 wc->flags[level] |= flag;
6036         }
6037
6038         /*
6039          * the block is shared by multiple trees, so it's not good to
6040          * keep the tree lock
6041          */
6042         if (path->locks[level] && level > 0) {
6043                 btrfs_tree_unlock(eb);
6044                 path->locks[level] = 0;
6045         }
6046         return 0;
6047 }
6048
6049 /*
6050  * hepler to process tree block pointer.
6051  *
6052  * when wc->stage == DROP_REFERENCE, this function checks
6053  * reference count of the block pointed to. if the block
6054  * is shared and we need update back refs for the subtree
6055  * rooted at the block, this function changes wc->stage to
6056  * UPDATE_BACKREF. if the block is shared and there is no
6057  * need to update back, this function drops the reference
6058  * to the block.
6059  *
6060  * NOTE: return value 1 means we should stop walking down.
6061  */
6062 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6063                                  struct btrfs_root *root,
6064                                  struct btrfs_path *path,
6065                                  struct walk_control *wc, int *lookup_info)
6066 {
6067         u64 bytenr;
6068         u64 generation;
6069         u64 parent;
6070         u32 blocksize;
6071         struct btrfs_key key;
6072         struct extent_buffer *next;
6073         int level = wc->level;
6074         int reada = 0;
6075         int ret = 0;
6076
6077         generation = btrfs_node_ptr_generation(path->nodes[level],
6078                                                path->slots[level]);
6079         /*
6080          * if the lower level block was created before the snapshot
6081          * was created, we know there is no need to update back refs
6082          * for the subtree
6083          */
6084         if (wc->stage == UPDATE_BACKREF &&
6085             generation <= root->root_key.offset) {
6086                 *lookup_info = 1;
6087                 return 1;
6088         }
6089
6090         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6091         blocksize = btrfs_level_size(root, level - 1);
6092
6093         next = btrfs_find_tree_block(root, bytenr, blocksize);
6094         if (!next) {
6095                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6096                 if (!next)
6097                         return -ENOMEM;
6098                 reada = 1;
6099         }
6100         btrfs_tree_lock(next);
6101         btrfs_set_lock_blocking(next);
6102
6103         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6104                                        &wc->refs[level - 1],
6105                                        &wc->flags[level - 1]);
6106         BUG_ON(ret);
6107         BUG_ON(wc->refs[level - 1] == 0);
6108         *lookup_info = 0;
6109
6110         if (wc->stage == DROP_REFERENCE) {
6111                 if (wc->refs[level - 1] > 1) {
6112                         if (level == 1 &&
6113                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6114                                 goto skip;
6115
6116                         if (!wc->update_ref ||
6117                             generation <= root->root_key.offset)
6118                                 goto skip;
6119
6120                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6121                                               path->slots[level]);
6122                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6123                         if (ret < 0)
6124                                 goto skip;
6125
6126                         wc->stage = UPDATE_BACKREF;
6127                         wc->shared_level = level - 1;
6128                 }
6129         } else {
6130                 if (level == 1 &&
6131                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6132                         goto skip;
6133         }
6134
6135         if (!btrfs_buffer_uptodate(next, generation)) {
6136                 btrfs_tree_unlock(next);
6137                 free_extent_buffer(next);
6138                 next = NULL;
6139                 *lookup_info = 1;
6140         }
6141
6142         if (!next) {
6143                 if (reada && level == 1)
6144                         reada_walk_down(trans, root, wc, path);
6145                 next = read_tree_block(root, bytenr, blocksize, generation);
6146                 if (!next)
6147                         return -EIO;
6148                 btrfs_tree_lock(next);
6149                 btrfs_set_lock_blocking(next);
6150         }
6151
6152         level--;
6153         BUG_ON(level != btrfs_header_level(next));
6154         path->nodes[level] = next;
6155         path->slots[level] = 0;
6156         path->locks[level] = 1;
6157         wc->level = level;
6158         if (wc->level == 1)
6159                 wc->reada_slot = 0;
6160         return 0;
6161 skip:
6162         wc->refs[level - 1] = 0;
6163         wc->flags[level - 1] = 0;
6164         if (wc->stage == DROP_REFERENCE) {
6165                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6166                         parent = path->nodes[level]->start;
6167                 } else {
6168                         BUG_ON(root->root_key.objectid !=
6169                                btrfs_header_owner(path->nodes[level]));
6170                         parent = 0;
6171                 }
6172
6173                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6174                                         root->root_key.objectid, level - 1, 0);
6175                 BUG_ON(ret);
6176         }
6177         btrfs_tree_unlock(next);
6178         free_extent_buffer(next);
6179         *lookup_info = 1;
6180         return 1;
6181 }
6182
6183 /*
6184  * hepler to process tree block while walking up the tree.
6185  *
6186  * when wc->stage == DROP_REFERENCE, this function drops
6187  * reference count on the block.
6188  *
6189  * when wc->stage == UPDATE_BACKREF, this function changes
6190  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6191  * to UPDATE_BACKREF previously while processing the block.
6192  *
6193  * NOTE: return value 1 means we should stop walking up.
6194  */
6195 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6196                                  struct btrfs_root *root,
6197                                  struct btrfs_path *path,
6198                                  struct walk_control *wc)
6199 {
6200         int ret;
6201         int level = wc->level;
6202         struct extent_buffer *eb = path->nodes[level];
6203         u64 parent = 0;
6204
6205         if (wc->stage == UPDATE_BACKREF) {
6206                 BUG_ON(wc->shared_level < level);
6207                 if (level < wc->shared_level)
6208                         goto out;
6209
6210                 ret = find_next_key(path, level + 1, &wc->update_progress);
6211                 if (ret > 0)
6212                         wc->update_ref = 0;
6213
6214                 wc->stage = DROP_REFERENCE;
6215                 wc->shared_level = -1;
6216                 path->slots[level] = 0;
6217
6218                 /*
6219                  * check reference count again if the block isn't locked.
6220                  * we should start walking down the tree again if reference
6221                  * count is one.
6222                  */
6223                 if (!path->locks[level]) {
6224                         BUG_ON(level == 0);
6225                         btrfs_tree_lock(eb);
6226                         btrfs_set_lock_blocking(eb);
6227                         path->locks[level] = 1;
6228
6229                         ret = btrfs_lookup_extent_info(trans, root,
6230                                                        eb->start, eb->len,
6231                                                        &wc->refs[level],
6232                                                        &wc->flags[level]);
6233                         BUG_ON(ret);
6234                         BUG_ON(wc->refs[level] == 0);
6235                         if (wc->refs[level] == 1) {
6236                                 btrfs_tree_unlock(eb);
6237                                 path->locks[level] = 0;
6238                                 return 1;
6239                         }
6240                 }
6241         }
6242
6243         /* wc->stage == DROP_REFERENCE */
6244         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6245
6246         if (wc->refs[level] == 1) {
6247                 if (level == 0) {
6248                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6249                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6250                         else
6251                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6252                         BUG_ON(ret);
6253                 }
6254                 /* make block locked assertion in clean_tree_block happy */
6255                 if (!path->locks[level] &&
6256                     btrfs_header_generation(eb) == trans->transid) {
6257                         btrfs_tree_lock(eb);
6258                         btrfs_set_lock_blocking(eb);
6259                         path->locks[level] = 1;
6260                 }
6261                 clean_tree_block(trans, root, eb);
6262         }
6263
6264         if (eb == root->node) {
6265                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6266                         parent = eb->start;
6267                 else
6268                         BUG_ON(root->root_key.objectid !=
6269                                btrfs_header_owner(eb));
6270         } else {
6271                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6272                         parent = path->nodes[level + 1]->start;
6273                 else
6274                         BUG_ON(root->root_key.objectid !=
6275                                btrfs_header_owner(path->nodes[level + 1]));
6276         }
6277
6278         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6279 out:
6280         wc->refs[level] = 0;
6281         wc->flags[level] = 0;
6282         return 0;
6283 }
6284
6285 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6286                                    struct btrfs_root *root,
6287                                    struct btrfs_path *path,
6288                                    struct walk_control *wc)
6289 {
6290         int level = wc->level;
6291         int lookup_info = 1;
6292         int ret;
6293
6294         while (level >= 0) {
6295                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6296                 if (ret > 0)
6297                         break;
6298
6299                 if (level == 0)
6300                         break;
6301
6302                 if (path->slots[level] >=
6303                     btrfs_header_nritems(path->nodes[level]))
6304                         break;
6305
6306                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6307                 if (ret > 0) {
6308                         path->slots[level]++;
6309                         continue;
6310                 } else if (ret < 0)
6311                         return ret;
6312                 level = wc->level;
6313         }
6314         return 0;
6315 }
6316
6317 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6318                                  struct btrfs_root *root,
6319                                  struct btrfs_path *path,
6320                                  struct walk_control *wc, int max_level)
6321 {
6322         int level = wc->level;
6323         int ret;
6324
6325         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6326         while (level < max_level && path->nodes[level]) {
6327                 wc->level = level;
6328                 if (path->slots[level] + 1 <
6329                     btrfs_header_nritems(path->nodes[level])) {
6330                         path->slots[level]++;
6331                         return 0;
6332                 } else {
6333                         ret = walk_up_proc(trans, root, path, wc);
6334                         if (ret > 0)
6335                                 return 0;
6336
6337                         if (path->locks[level]) {
6338                                 btrfs_tree_unlock(path->nodes[level]);
6339                                 path->locks[level] = 0;
6340                         }
6341                         free_extent_buffer(path->nodes[level]);
6342                         path->nodes[level] = NULL;
6343                         level++;
6344                 }
6345         }
6346         return 1;
6347 }
6348
6349 /*
6350  * drop a subvolume tree.
6351  *
6352  * this function traverses the tree freeing any blocks that only
6353  * referenced by the tree.
6354  *
6355  * when a shared tree block is found. this function decreases its
6356  * reference count by one. if update_ref is true, this function
6357  * also make sure backrefs for the shared block and all lower level
6358  * blocks are properly updated.
6359  */
6360 int btrfs_drop_snapshot(struct btrfs_root *root,
6361                         struct btrfs_block_rsv *block_rsv, int update_ref)
6362 {
6363         struct btrfs_path *path;
6364         struct btrfs_trans_handle *trans;
6365         struct btrfs_root *tree_root = root->fs_info->tree_root;
6366         struct btrfs_root_item *root_item = &root->root_item;
6367         struct walk_control *wc;
6368         struct btrfs_key key;
6369         int err = 0;
6370         int ret;
6371         int level;
6372
6373         path = btrfs_alloc_path();
6374         BUG_ON(!path);
6375
6376         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6377         BUG_ON(!wc);
6378
6379         trans = btrfs_start_transaction(tree_root, 0);
6380         BUG_ON(IS_ERR(trans));
6381
6382         if (block_rsv)
6383                 trans->block_rsv = block_rsv;
6384
6385         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6386                 level = btrfs_header_level(root->node);
6387                 path->nodes[level] = btrfs_lock_root_node(root);
6388                 btrfs_set_lock_blocking(path->nodes[level]);
6389                 path->slots[level] = 0;
6390                 path->locks[level] = 1;
6391                 memset(&wc->update_progress, 0,
6392                        sizeof(wc->update_progress));
6393         } else {
6394                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6395                 memcpy(&wc->update_progress, &key,
6396                        sizeof(wc->update_progress));
6397
6398                 level = root_item->drop_level;
6399                 BUG_ON(level == 0);
6400                 path->lowest_level = level;
6401                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6402                 path->lowest_level = 0;
6403                 if (ret < 0) {
6404                         err = ret;
6405                         goto out;
6406                 }
6407                 WARN_ON(ret > 0);
6408
6409                 /*
6410                  * unlock our path, this is safe because only this
6411                  * function is allowed to delete this snapshot
6412                  */
6413                 btrfs_unlock_up_safe(path, 0);
6414
6415                 level = btrfs_header_level(root->node);
6416                 while (1) {
6417                         btrfs_tree_lock(path->nodes[level]);
6418                         btrfs_set_lock_blocking(path->nodes[level]);
6419
6420                         ret = btrfs_lookup_extent_info(trans, root,
6421                                                 path->nodes[level]->start,
6422                                                 path->nodes[level]->len,
6423                                                 &wc->refs[level],
6424                                                 &wc->flags[level]);
6425                         BUG_ON(ret);
6426                         BUG_ON(wc->refs[level] == 0);
6427
6428                         if (level == root_item->drop_level)
6429                                 break;
6430
6431                         btrfs_tree_unlock(path->nodes[level]);
6432                         WARN_ON(wc->refs[level] != 1);
6433                         level--;
6434                 }
6435         }
6436
6437         wc->level = level;
6438         wc->shared_level = -1;
6439         wc->stage = DROP_REFERENCE;
6440         wc->update_ref = update_ref;
6441         wc->keep_locks = 0;
6442         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6443
6444         while (1) {
6445                 ret = walk_down_tree(trans, root, path, wc);
6446                 if (ret < 0) {
6447                         err = ret;
6448                         break;
6449                 }
6450
6451                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6452                 if (ret < 0) {
6453                         err = ret;
6454                         break;
6455                 }
6456
6457                 if (ret > 0) {
6458                         BUG_ON(wc->stage != DROP_REFERENCE);
6459                         break;
6460                 }
6461
6462                 if (wc->stage == DROP_REFERENCE) {
6463                         level = wc->level;
6464                         btrfs_node_key(path->nodes[level],
6465                                        &root_item->drop_progress,
6466                                        path->slots[level]);
6467                         root_item->drop_level = level;
6468                 }
6469
6470                 BUG_ON(wc->level == 0);
6471                 if (btrfs_should_end_transaction(trans, tree_root)) {
6472                         ret = btrfs_update_root(trans, tree_root,
6473                                                 &root->root_key,
6474                                                 root_item);
6475                         BUG_ON(ret);
6476
6477                         btrfs_end_transaction_throttle(trans, tree_root);
6478                         trans = btrfs_start_transaction(tree_root, 0);
6479                         BUG_ON(IS_ERR(trans));
6480                         if (block_rsv)
6481                                 trans->block_rsv = block_rsv;
6482                 }
6483         }
6484         btrfs_release_path(root, path);
6485         BUG_ON(err);
6486
6487         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6488         BUG_ON(ret);
6489
6490         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6491                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6492                                            NULL, NULL);
6493                 BUG_ON(ret < 0);
6494                 if (ret > 0) {
6495                         /* if we fail to delete the orphan item this time
6496                          * around, it'll get picked up the next time.
6497                          *
6498                          * The most common failure here is just -ENOENT.
6499                          */
6500                         btrfs_del_orphan_item(trans, tree_root,
6501                                               root->root_key.objectid);
6502                 }
6503         }
6504
6505         if (root->in_radix) {
6506                 btrfs_free_fs_root(tree_root->fs_info, root);
6507         } else {
6508                 free_extent_buffer(root->node);
6509                 free_extent_buffer(root->commit_root);
6510                 kfree(root);
6511         }
6512 out:
6513         btrfs_end_transaction_throttle(trans, tree_root);
6514         kfree(wc);
6515         btrfs_free_path(path);
6516         return err;
6517 }
6518
6519 /*
6520  * drop subtree rooted at tree block 'node'.
6521  *
6522  * NOTE: this function will unlock and release tree block 'node'
6523  */
6524 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6525                         struct btrfs_root *root,
6526                         struct extent_buffer *node,
6527                         struct extent_buffer *parent)
6528 {
6529         struct btrfs_path *path;
6530         struct walk_control *wc;
6531         int level;
6532         int parent_level;
6533         int ret = 0;
6534         int wret;
6535
6536         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6537
6538         path = btrfs_alloc_path();
6539         if (!path)
6540                 return -ENOMEM;
6541
6542         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6543         if (!wc) {
6544                 btrfs_free_path(path);
6545                 return -ENOMEM;
6546         }
6547
6548         btrfs_assert_tree_locked(parent);
6549         parent_level = btrfs_header_level(parent);
6550         extent_buffer_get(parent);
6551         path->nodes[parent_level] = parent;
6552         path->slots[parent_level] = btrfs_header_nritems(parent);
6553
6554         btrfs_assert_tree_locked(node);
6555         level = btrfs_header_level(node);
6556         path->nodes[level] = node;
6557         path->slots[level] = 0;
6558         path->locks[level] = 1;
6559
6560         wc->refs[parent_level] = 1;
6561         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6562         wc->level = level;
6563         wc->shared_level = -1;
6564         wc->stage = DROP_REFERENCE;
6565         wc->update_ref = 0;
6566         wc->keep_locks = 1;
6567         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6568
6569         while (1) {
6570                 wret = walk_down_tree(trans, root, path, wc);
6571                 if (wret < 0) {
6572                         ret = wret;
6573                         break;
6574                 }
6575
6576                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6577                 if (wret < 0)
6578                         ret = wret;
6579                 if (wret != 0)
6580                         break;
6581         }
6582
6583         kfree(wc);
6584         btrfs_free_path(path);
6585         return ret;
6586 }
6587
6588 #if 0
6589 static unsigned long calc_ra(unsigned long start, unsigned long last,
6590                              unsigned long nr)
6591 {
6592         return min(last, start + nr - 1);
6593 }
6594
6595 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6596                                          u64 len)
6597 {
6598         u64 page_start;
6599         u64 page_end;
6600         unsigned long first_index;
6601         unsigned long last_index;
6602         unsigned long i;
6603         struct page *page;
6604         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6605         struct file_ra_state *ra;
6606         struct btrfs_ordered_extent *ordered;
6607         unsigned int total_read = 0;
6608         unsigned int total_dirty = 0;
6609         int ret = 0;
6610
6611         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6612         if (!ra)
6613                 return -ENOMEM;
6614
6615         mutex_lock(&inode->i_mutex);
6616         first_index = start >> PAGE_CACHE_SHIFT;
6617         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6618
6619         /* make sure the dirty trick played by the caller work */
6620         ret = invalidate_inode_pages2_range(inode->i_mapping,
6621                                             first_index, last_index);
6622         if (ret)
6623                 goto out_unlock;
6624
6625         file_ra_state_init(ra, inode->i_mapping);
6626
6627         for (i = first_index ; i <= last_index; i++) {
6628                 if (total_read % ra->ra_pages == 0) {
6629                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6630                                        calc_ra(i, last_index, ra->ra_pages));
6631                 }
6632                 total_read++;
6633 again:
6634                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6635                         BUG_ON(1);
6636                 page = grab_cache_page(inode->i_mapping, i);
6637                 if (!page) {
6638                         ret = -ENOMEM;
6639                         goto out_unlock;
6640                 }
6641                 if (!PageUptodate(page)) {
6642                         btrfs_readpage(NULL, page);
6643                         lock_page(page);
6644                         if (!PageUptodate(page)) {
6645                                 unlock_page(page);
6646                                 page_cache_release(page);
6647                                 ret = -EIO;
6648                                 goto out_unlock;
6649                         }
6650                 }
6651                 wait_on_page_writeback(page);
6652
6653                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6654                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6655                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6656
6657                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6658                 if (ordered) {
6659                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6660                         unlock_page(page);
6661                         page_cache_release(page);
6662                         btrfs_start_ordered_extent(inode, ordered, 1);
6663                         btrfs_put_ordered_extent(ordered);
6664                         goto again;
6665                 }
6666                 set_page_extent_mapped(page);
6667
6668                 if (i == first_index)
6669                         set_extent_bits(io_tree, page_start, page_end,
6670                                         EXTENT_BOUNDARY, GFP_NOFS);
6671                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6672
6673                 set_page_dirty(page);
6674                 total_dirty++;
6675
6676                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6677                 unlock_page(page);
6678                 page_cache_release(page);
6679         }
6680
6681 out_unlock:
6682         kfree(ra);
6683         mutex_unlock(&inode->i_mutex);
6684         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6685         return ret;
6686 }
6687
6688 static noinline int relocate_data_extent(struct inode *reloc_inode,
6689                                          struct btrfs_key *extent_key,
6690                                          u64 offset)
6691 {
6692         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6693         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6694         struct extent_map *em;
6695         u64 start = extent_key->objectid - offset;
6696         u64 end = start + extent_key->offset - 1;
6697
6698         em = alloc_extent_map(GFP_NOFS);
6699         BUG_ON(!em);
6700
6701         em->start = start;
6702         em->len = extent_key->offset;
6703         em->block_len = extent_key->offset;
6704         em->block_start = extent_key->objectid;
6705         em->bdev = root->fs_info->fs_devices->latest_bdev;
6706         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6707
6708         /* setup extent map to cheat btrfs_readpage */
6709         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6710         while (1) {
6711                 int ret;
6712                 write_lock(&em_tree->lock);
6713                 ret = add_extent_mapping(em_tree, em);
6714                 write_unlock(&em_tree->lock);
6715                 if (ret != -EEXIST) {
6716                         free_extent_map(em);
6717                         break;
6718                 }
6719                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6720         }
6721         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6722
6723         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6724 }
6725
6726 struct btrfs_ref_path {
6727         u64 extent_start;
6728         u64 nodes[BTRFS_MAX_LEVEL];
6729         u64 root_objectid;
6730         u64 root_generation;
6731         u64 owner_objectid;
6732         u32 num_refs;
6733         int lowest_level;
6734         int current_level;
6735         int shared_level;
6736
6737         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6738         u64 new_nodes[BTRFS_MAX_LEVEL];
6739 };
6740
6741 struct disk_extent {
6742         u64 ram_bytes;
6743         u64 disk_bytenr;
6744         u64 disk_num_bytes;
6745         u64 offset;
6746         u64 num_bytes;
6747         u8 compression;
6748         u8 encryption;
6749         u16 other_encoding;
6750 };
6751
6752 static int is_cowonly_root(u64 root_objectid)
6753 {
6754         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6755             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6756             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6757             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6758             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6759             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6760                 return 1;
6761         return 0;
6762 }
6763
6764 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6765                                     struct btrfs_root *extent_root,
6766                                     struct btrfs_ref_path *ref_path,
6767                                     int first_time)
6768 {
6769         struct extent_buffer *leaf;
6770         struct btrfs_path *path;
6771         struct btrfs_extent_ref *ref;
6772         struct btrfs_key key;
6773         struct btrfs_key found_key;
6774         u64 bytenr;
6775         u32 nritems;
6776         int level;
6777         int ret = 1;
6778
6779         path = btrfs_alloc_path();
6780         if (!path)
6781                 return -ENOMEM;
6782
6783         if (first_time) {
6784                 ref_path->lowest_level = -1;
6785                 ref_path->current_level = -1;
6786                 ref_path->shared_level = -1;
6787                 goto walk_up;
6788         }
6789 walk_down:
6790         level = ref_path->current_level - 1;
6791         while (level >= -1) {
6792                 u64 parent;
6793                 if (level < ref_path->lowest_level)
6794                         break;
6795
6796                 if (level >= 0)
6797                         bytenr = ref_path->nodes[level];
6798                 else
6799                         bytenr = ref_path->extent_start;
6800                 BUG_ON(bytenr == 0);
6801
6802                 parent = ref_path->nodes[level + 1];
6803                 ref_path->nodes[level + 1] = 0;
6804                 ref_path->current_level = level;
6805                 BUG_ON(parent == 0);
6806
6807                 key.objectid = bytenr;
6808                 key.offset = parent + 1;
6809                 key.type = BTRFS_EXTENT_REF_KEY;
6810
6811                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6812                 if (ret < 0)
6813                         goto out;
6814                 BUG_ON(ret == 0);
6815
6816                 leaf = path->nodes[0];
6817                 nritems = btrfs_header_nritems(leaf);
6818                 if (path->slots[0] >= nritems) {
6819                         ret = btrfs_next_leaf(extent_root, path);
6820                         if (ret < 0)
6821                                 goto out;
6822                         if (ret > 0)
6823                                 goto next;
6824                         leaf = path->nodes[0];
6825                 }
6826
6827                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6828                 if (found_key.objectid == bytenr &&
6829                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6830                         if (level < ref_path->shared_level)
6831                                 ref_path->shared_level = level;
6832                         goto found;
6833                 }
6834 next:
6835                 level--;
6836                 btrfs_release_path(extent_root, path);
6837                 cond_resched();
6838         }
6839         /* reached lowest level */
6840         ret = 1;
6841         goto out;
6842 walk_up:
6843         level = ref_path->current_level;
6844         while (level < BTRFS_MAX_LEVEL - 1) {
6845                 u64 ref_objectid;
6846
6847                 if (level >= 0)
6848                         bytenr = ref_path->nodes[level];
6849                 else
6850                         bytenr = ref_path->extent_start;
6851
6852                 BUG_ON(bytenr == 0);
6853
6854                 key.objectid = bytenr;
6855                 key.offset = 0;
6856                 key.type = BTRFS_EXTENT_REF_KEY;
6857
6858                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6859                 if (ret < 0)
6860                         goto out;
6861
6862                 leaf = path->nodes[0];
6863                 nritems = btrfs_header_nritems(leaf);
6864                 if (path->slots[0] >= nritems) {
6865                         ret = btrfs_next_leaf(extent_root, path);
6866                         if (ret < 0)
6867                                 goto out;
6868                         if (ret > 0) {
6869                                 /* the extent was freed by someone */
6870                                 if (ref_path->lowest_level == level)
6871                                         goto out;
6872                                 btrfs_release_path(extent_root, path);
6873                                 goto walk_down;
6874                         }
6875                         leaf = path->nodes[0];
6876                 }
6877
6878                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6879                 if (found_key.objectid != bytenr ||
6880                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6881                         /* the extent was freed by someone */
6882                         if (ref_path->lowest_level == level) {
6883                                 ret = 1;
6884                                 goto out;
6885                         }
6886                         btrfs_release_path(extent_root, path);
6887                         goto walk_down;
6888                 }
6889 found:
6890                 ref = btrfs_item_ptr(leaf, path->slots[0],
6891                                 struct btrfs_extent_ref);
6892                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6893                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6894                         if (first_time) {
6895                                 level = (int)ref_objectid;
6896                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6897                                 ref_path->lowest_level = level;
6898                                 ref_path->current_level = level;
6899                                 ref_path->nodes[level] = bytenr;
6900                         } else {
6901                                 WARN_ON(ref_objectid != level);
6902                         }
6903                 } else {
6904                         WARN_ON(level != -1);
6905                 }
6906                 first_time = 0;
6907
6908                 if (ref_path->lowest_level == level) {
6909                         ref_path->owner_objectid = ref_objectid;
6910                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6911                 }
6912
6913                 /*
6914                  * the block is tree root or the block isn't in reference
6915                  * counted tree.
6916                  */
6917                 if (found_key.objectid == found_key.offset ||
6918                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6919                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6920                         ref_path->root_generation =
6921                                 btrfs_ref_generation(leaf, ref);
6922                         if (level < 0) {
6923                                 /* special reference from the tree log */
6924                                 ref_path->nodes[0] = found_key.offset;
6925                                 ref_path->current_level = 0;
6926                         }
6927                         ret = 0;
6928                         goto out;
6929                 }
6930
6931                 level++;
6932                 BUG_ON(ref_path->nodes[level] != 0);
6933                 ref_path->nodes[level] = found_key.offset;
6934                 ref_path->current_level = level;
6935
6936                 /*
6937                  * the reference was created in the running transaction,
6938                  * no need to continue walking up.
6939                  */
6940                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6941                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6942                         ref_path->root_generation =
6943                                 btrfs_ref_generation(leaf, ref);
6944                         ret = 0;
6945                         goto out;
6946                 }
6947
6948                 btrfs_release_path(extent_root, path);
6949                 cond_resched();
6950         }
6951         /* reached max tree level, but no tree root found. */
6952         BUG();
6953 out:
6954         btrfs_free_path(path);
6955         return ret;
6956 }
6957
6958 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6959                                 struct btrfs_root *extent_root,
6960                                 struct btrfs_ref_path *ref_path,
6961                                 u64 extent_start)
6962 {
6963         memset(ref_path, 0, sizeof(*ref_path));
6964         ref_path->extent_start = extent_start;
6965
6966         return __next_ref_path(trans, extent_root, ref_path, 1);
6967 }
6968
6969 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6970                                struct btrfs_root *extent_root,
6971                                struct btrfs_ref_path *ref_path)
6972 {
6973         return __next_ref_path(trans, extent_root, ref_path, 0);
6974 }
6975
6976 static noinline int get_new_locations(struct inode *reloc_inode,
6977                                       struct btrfs_key *extent_key,
6978                                       u64 offset, int no_fragment,
6979                                       struct disk_extent **extents,
6980                                       int *nr_extents)
6981 {
6982         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6983         struct btrfs_path *path;
6984         struct btrfs_file_extent_item *fi;
6985         struct extent_buffer *leaf;
6986         struct disk_extent *exts = *extents;
6987         struct btrfs_key found_key;
6988         u64 cur_pos;
6989         u64 last_byte;
6990         u32 nritems;
6991         int nr = 0;
6992         int max = *nr_extents;
6993         int ret;
6994
6995         WARN_ON(!no_fragment && *extents);
6996         if (!exts) {
6997                 max = 1;
6998                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6999                 if (!exts)
7000                         return -ENOMEM;
7001         }
7002
7003         path = btrfs_alloc_path();
7004         if (!path) {
7005                 if (exts != *extents)
7006                         kfree(exts);
7007                 return -ENOMEM;
7008         }
7009
7010         cur_pos = extent_key->objectid - offset;
7011         last_byte = extent_key->objectid + extent_key->offset;
7012         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
7013                                        cur_pos, 0);
7014         if (ret < 0)
7015                 goto out;
7016         if (ret > 0) {
7017                 ret = -ENOENT;
7018                 goto out;
7019         }
7020
7021         while (1) {
7022                 leaf = path->nodes[0];
7023                 nritems = btrfs_header_nritems(leaf);
7024                 if (path->slots[0] >= nritems) {
7025                         ret = btrfs_next_leaf(root, path);
7026                         if (ret < 0)
7027                                 goto out;
7028                         if (ret > 0)
7029                                 break;
7030                         leaf = path->nodes[0];
7031                 }
7032
7033                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7034                 if (found_key.offset != cur_pos ||
7035                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
7036                     found_key.objectid != reloc_inode->i_ino)
7037                         break;
7038
7039                 fi = btrfs_item_ptr(leaf, path->slots[0],
7040                                     struct btrfs_file_extent_item);
7041                 if (btrfs_file_extent_type(leaf, fi) !=
7042                     BTRFS_FILE_EXTENT_REG ||
7043                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7044                         break;
7045
7046                 if (nr == max) {
7047                         struct disk_extent *old = exts;
7048                         max *= 2;
7049                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
7050                         if (!exts) {
7051                                 ret = -ENOMEM;
7052                                 goto out;
7053                         }
7054                         memcpy(exts, old, sizeof(*exts) * nr);
7055                         if (old != *extents)
7056                                 kfree(old);
7057                 }
7058
7059                 exts[nr].disk_bytenr =
7060                         btrfs_file_extent_disk_bytenr(leaf, fi);
7061                 exts[nr].disk_num_bytes =
7062                         btrfs_file_extent_disk_num_bytes(leaf, fi);
7063                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
7064                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7065                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7066                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
7067                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
7068                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
7069                                                                            fi);
7070                 BUG_ON(exts[nr].offset > 0);
7071                 BUG_ON(exts[nr].compression || exts[nr].encryption);
7072                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
7073
7074                 cur_pos += exts[nr].num_bytes;
7075                 nr++;
7076
7077                 if (cur_pos + offset >= last_byte)
7078                         break;
7079
7080                 if (no_fragment) {
7081                         ret = 1;
7082                         goto out;
7083                 }
7084                 path->slots[0]++;
7085         }
7086
7087         BUG_ON(cur_pos + offset > last_byte);
7088         if (cur_pos + offset < last_byte) {
7089                 ret = -ENOENT;
7090                 goto out;
7091         }
7092         ret = 0;
7093 out:
7094         btrfs_free_path(path);
7095         if (ret) {
7096                 if (exts != *extents)
7097                         kfree(exts);
7098         } else {
7099                 *extents = exts;
7100                 *nr_extents = nr;
7101         }
7102         return ret;
7103 }
7104
7105 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
7106                                         struct btrfs_root *root,
7107                                         struct btrfs_path *path,
7108                                         struct btrfs_key *extent_key,
7109                                         struct btrfs_key *leaf_key,
7110                                         struct btrfs_ref_path *ref_path,
7111                                         struct disk_extent *new_extents,
7112                                         int nr_extents)
7113 {
7114         struct extent_buffer *leaf;
7115         struct btrfs_file_extent_item *fi;
7116         struct inode *inode = NULL;
7117         struct btrfs_key key;
7118         u64 lock_start = 0;
7119         u64 lock_end = 0;
7120         u64 num_bytes;
7121         u64 ext_offset;
7122         u64 search_end = (u64)-1;
7123         u32 nritems;
7124         int nr_scaned = 0;
7125         int extent_locked = 0;
7126         int extent_type;
7127         int ret;
7128
7129         memcpy(&key, leaf_key, sizeof(key));
7130         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7131                 if (key.objectid < ref_path->owner_objectid ||
7132                     (key.objectid == ref_path->owner_objectid &&
7133                      key.type < BTRFS_EXTENT_DATA_KEY)) {
7134                         key.objectid = ref_path->owner_objectid;
7135                         key.type = BTRFS_EXTENT_DATA_KEY;
7136                         key.offset = 0;
7137                 }
7138         }
7139
7140         while (1) {
7141                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7142                 if (ret < 0)
7143                         goto out;
7144
7145                 leaf = path->nodes[0];
7146                 nritems = btrfs_header_nritems(leaf);
7147 next:
7148                 if (extent_locked && ret > 0) {
7149                         /*
7150                          * the file extent item was modified by someone
7151                          * before the extent got locked.
7152                          */
7153                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7154                                       lock_end, GFP_NOFS);
7155                         extent_locked = 0;
7156                 }
7157
7158                 if (path->slots[0] >= nritems) {
7159                         if (++nr_scaned > 2)
7160                                 break;
7161
7162                         BUG_ON(extent_locked);
7163                         ret = btrfs_next_leaf(root, path);
7164                         if (ret < 0)
7165                                 goto out;
7166                         if (ret > 0)
7167                                 break;
7168                         leaf = path->nodes[0];
7169                         nritems = btrfs_header_nritems(leaf);
7170                 }
7171
7172                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7173
7174                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7175                         if ((key.objectid > ref_path->owner_objectid) ||
7176                             (key.objectid == ref_path->owner_objectid &&
7177                              key.type > BTRFS_EXTENT_DATA_KEY) ||
7178                             key.offset >= search_end)
7179                                 break;
7180                 }
7181
7182                 if (inode && key.objectid != inode->i_ino) {
7183                         BUG_ON(extent_locked);
7184                         btrfs_release_path(root, path);
7185                         mutex_unlock(&inode->i_mutex);
7186                         iput(inode);
7187                         inode = NULL;
7188                         continue;
7189                 }
7190
7191                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7192                         path->slots[0]++;
7193                         ret = 1;
7194                         goto next;
7195                 }
7196                 fi = btrfs_item_ptr(leaf, path->slots[0],
7197                                     struct btrfs_file_extent_item);
7198                 extent_type = btrfs_file_extent_type(leaf, fi);
7199                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7200                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7201                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7202                      extent_key->objectid)) {
7203                         path->slots[0]++;
7204                         ret = 1;
7205                         goto next;
7206                 }
7207
7208                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7209                 ext_offset = btrfs_file_extent_offset(leaf, fi);
7210
7211                 if (search_end == (u64)-1) {
7212                         search_end = key.offset - ext_offset +
7213                                 btrfs_file_extent_ram_bytes(leaf, fi);
7214                 }
7215
7216                 if (!extent_locked) {
7217                         lock_start = key.offset;
7218                         lock_end = lock_start + num_bytes - 1;
7219                 } else {
7220                         if (lock_start > key.offset ||
7221                             lock_end + 1 < key.offset + num_bytes) {
7222                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7223                                               lock_start, lock_end, GFP_NOFS);
7224                                 extent_locked = 0;
7225                         }
7226                 }
7227
7228                 if (!inode) {
7229                         btrfs_release_path(root, path);
7230
7231                         inode = btrfs_iget_locked(root->fs_info->sb,
7232                                                   key.objectid, root);
7233                         if (inode->i_state & I_NEW) {
7234                                 BTRFS_I(inode)->root = root;
7235                                 BTRFS_I(inode)->location.objectid =
7236                                         key.objectid;
7237                                 BTRFS_I(inode)->location.type =
7238                                         BTRFS_INODE_ITEM_KEY;
7239                                 BTRFS_I(inode)->location.offset = 0;
7240                                 btrfs_read_locked_inode(inode);
7241                                 unlock_new_inode(inode);
7242                         }
7243                         /*
7244                          * some code call btrfs_commit_transaction while
7245                          * holding the i_mutex, so we can't use mutex_lock
7246                          * here.
7247                          */
7248                         if (is_bad_inode(inode) ||
7249                             !mutex_trylock(&inode->i_mutex)) {
7250                                 iput(inode);
7251                                 inode = NULL;
7252                                 key.offset = (u64)-1;
7253                                 goto skip;
7254                         }
7255                 }
7256
7257                 if (!extent_locked) {
7258                         struct btrfs_ordered_extent *ordered;
7259
7260                         btrfs_release_path(root, path);
7261
7262                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7263                                     lock_end, GFP_NOFS);
7264                         ordered = btrfs_lookup_first_ordered_extent(inode,
7265                                                                     lock_end);
7266                         if (ordered &&
7267                             ordered->file_offset <= lock_end &&
7268                             ordered->file_offset + ordered->len > lock_start) {
7269                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7270                                               lock_start, lock_end, GFP_NOFS);
7271                                 btrfs_start_ordered_extent(inode, ordered, 1);
7272                                 btrfs_put_ordered_extent(ordered);
7273                                 key.offset += num_bytes;
7274                                 goto skip;
7275                         }
7276                         if (ordered)
7277                                 btrfs_put_ordered_extent(ordered);
7278
7279                         extent_locked = 1;
7280                         continue;
7281                 }
7282
7283                 if (nr_extents == 1) {
7284                         /* update extent pointer in place */
7285                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7286                                                 new_extents[0].disk_bytenr);
7287                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7288                                                 new_extents[0].disk_num_bytes);
7289                         btrfs_mark_buffer_dirty(leaf);
7290
7291                         btrfs_drop_extent_cache(inode, key.offset,
7292                                                 key.offset + num_bytes - 1, 0);
7293
7294                         ret = btrfs_inc_extent_ref(trans, root,
7295                                                 new_extents[0].disk_bytenr,
7296                                                 new_extents[0].disk_num_bytes,
7297                                                 leaf->start,
7298                                                 root->root_key.objectid,
7299                                                 trans->transid,
7300                                                 key.objectid);
7301                         BUG_ON(ret);
7302
7303                         ret = btrfs_free_extent(trans, root,
7304                                                 extent_key->objectid,
7305                                                 extent_key->offset,
7306                                                 leaf->start,
7307                                                 btrfs_header_owner(leaf),
7308                                                 btrfs_header_generation(leaf),
7309                                                 key.objectid, 0);
7310                         BUG_ON(ret);
7311
7312                         btrfs_release_path(root, path);
7313                         key.offset += num_bytes;
7314                 } else {
7315                         BUG_ON(1);
7316 #if 0
7317                         u64 alloc_hint;
7318                         u64 extent_len;
7319                         int i;
7320                         /*
7321                          * drop old extent pointer at first, then insert the
7322                          * new pointers one bye one
7323                          */
7324                         btrfs_release_path(root, path);
7325                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7326                                                  key.offset + num_bytes,
7327                                                  key.offset, &alloc_hint);
7328                         BUG_ON(ret);
7329
7330                         for (i = 0; i < nr_extents; i++) {
7331                                 if (ext_offset >= new_extents[i].num_bytes) {
7332                                         ext_offset -= new_extents[i].num_bytes;
7333                                         continue;
7334                                 }
7335                                 extent_len = min(new_extents[i].num_bytes -
7336                                                  ext_offset, num_bytes);
7337
7338                                 ret = btrfs_insert_empty_item(trans, root,
7339                                                               path, &key,
7340                                                               sizeof(*fi));
7341                                 BUG_ON(ret);
7342
7343                                 leaf = path->nodes[0];
7344                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7345                                                 struct btrfs_file_extent_item);
7346                                 btrfs_set_file_extent_generation(leaf, fi,
7347                                                         trans->transid);
7348                                 btrfs_set_file_extent_type(leaf, fi,
7349                                                         BTRFS_FILE_EXTENT_REG);
7350                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7351                                                 new_extents[i].disk_bytenr);
7352                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7353                                                 new_extents[i].disk_num_bytes);
7354                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7355                                                 new_extents[i].ram_bytes);
7356
7357                                 btrfs_set_file_extent_compression(leaf, fi,
7358                                                 new_extents[i].compression);
7359                                 btrfs_set_file_extent_encryption(leaf, fi,
7360                                                 new_extents[i].encryption);
7361                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7362                                                 new_extents[i].other_encoding);
7363
7364                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7365                                                         extent_len);
7366                                 ext_offset += new_extents[i].offset;
7367                                 btrfs_set_file_extent_offset(leaf, fi,
7368                                                         ext_offset);
7369                                 btrfs_mark_buffer_dirty(leaf);
7370
7371                                 btrfs_drop_extent_cache(inode, key.offset,
7372                                                 key.offset + extent_len - 1, 0);
7373
7374                                 ret = btrfs_inc_extent_ref(trans, root,
7375                                                 new_extents[i].disk_bytenr,
7376                                                 new_extents[i].disk_num_bytes,
7377                                                 leaf->start,
7378                                                 root->root_key.objectid,
7379                                                 trans->transid, key.objectid);
7380                                 BUG_ON(ret);
7381                                 btrfs_release_path(root, path);
7382
7383                                 inode_add_bytes(inode, extent_len);
7384
7385                                 ext_offset = 0;
7386                                 num_bytes -= extent_len;
7387                                 key.offset += extent_len;
7388
7389                                 if (num_bytes == 0)
7390                                         break;
7391                         }
7392                         BUG_ON(i >= nr_extents);
7393 #endif
7394                 }
7395
7396                 if (extent_locked) {
7397                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7398                                       lock_end, GFP_NOFS);
7399                         extent_locked = 0;
7400                 }
7401 skip:
7402                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7403                     key.offset >= search_end)
7404                         break;
7405
7406                 cond_resched();
7407         }
7408         ret = 0;
7409 out:
7410         btrfs_release_path(root, path);
7411         if (inode) {
7412                 mutex_unlock(&inode->i_mutex);
7413                 if (extent_locked) {
7414                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7415                                       lock_end, GFP_NOFS);
7416                 }
7417                 iput(inode);
7418         }
7419         return ret;
7420 }
7421
7422 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7423                                struct btrfs_root *root,
7424                                struct extent_buffer *buf, u64 orig_start)
7425 {
7426         int level;
7427         int ret;
7428
7429         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7430         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7431
7432         level = btrfs_header_level(buf);
7433         if (level == 0) {
7434                 struct btrfs_leaf_ref *ref;
7435                 struct btrfs_leaf_ref *orig_ref;
7436
7437                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7438                 if (!orig_ref)
7439                         return -ENOENT;
7440
7441                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7442                 if (!ref) {
7443                         btrfs_free_leaf_ref(root, orig_ref);
7444                         return -ENOMEM;
7445                 }
7446
7447                 ref->nritems = orig_ref->nritems;
7448                 memcpy(ref->extents, orig_ref->extents,
7449                         sizeof(ref->extents[0]) * ref->nritems);
7450
7451                 btrfs_free_leaf_ref(root, orig_ref);
7452
7453                 ref->root_gen = trans->transid;
7454                 ref->bytenr = buf->start;
7455                 ref->owner = btrfs_header_owner(buf);
7456                 ref->generation = btrfs_header_generation(buf);
7457
7458                 ret = btrfs_add_leaf_ref(root, ref, 0);
7459                 WARN_ON(ret);
7460                 btrfs_free_leaf_ref(root, ref);
7461         }
7462         return 0;
7463 }
7464
7465 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7466                                         struct extent_buffer *leaf,
7467                                         struct btrfs_block_group_cache *group,
7468                                         struct btrfs_root *target_root)
7469 {
7470         struct btrfs_key key;
7471         struct inode *inode = NULL;
7472         struct btrfs_file_extent_item *fi;
7473         struct extent_state *cached_state = NULL;
7474         u64 num_bytes;
7475         u64 skip_objectid = 0;
7476         u32 nritems;
7477         u32 i;
7478
7479         nritems = btrfs_header_nritems(leaf);
7480         for (i = 0; i < nritems; i++) {
7481                 btrfs_item_key_to_cpu(leaf, &key, i);
7482                 if (key.objectid == skip_objectid ||
7483                     key.type != BTRFS_EXTENT_DATA_KEY)
7484                         continue;
7485                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7486                 if (btrfs_file_extent_type(leaf, fi) ==
7487                     BTRFS_FILE_EXTENT_INLINE)
7488                         continue;
7489                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7490                         continue;
7491                 if (!inode || inode->i_ino != key.objectid) {
7492                         iput(inode);
7493                         inode = btrfs_ilookup(target_root->fs_info->sb,
7494                                               key.objectid, target_root, 1);
7495                 }
7496                 if (!inode) {
7497                         skip_objectid = key.objectid;
7498                         continue;
7499                 }
7500                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7501
7502                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7503                                  key.offset + num_bytes - 1, 0, &cached_state,
7504                                  GFP_NOFS);
7505                 btrfs_drop_extent_cache(inode, key.offset,
7506                                         key.offset + num_bytes - 1, 1);
7507                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7508                                      key.offset + num_bytes - 1, &cached_state,
7509                                      GFP_NOFS);
7510                 cond_resched();
7511         }
7512         iput(inode);
7513         return 0;
7514 }
7515
7516 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7517                                         struct btrfs_root *root,
7518                                         struct extent_buffer *leaf,
7519                                         struct btrfs_block_group_cache *group,
7520                                         struct inode *reloc_inode)
7521 {
7522         struct btrfs_key key;
7523         struct btrfs_key extent_key;
7524         struct btrfs_file_extent_item *fi;
7525         struct btrfs_leaf_ref *ref;
7526         struct disk_extent *new_extent;
7527         u64 bytenr;
7528         u64 num_bytes;
7529         u32 nritems;
7530         u32 i;
7531         int ext_index;
7532         int nr_extent;
7533         int ret;
7534
7535         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7536         if (!new_extent)
7537                 return -ENOMEM;
7538
7539         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7540         BUG_ON(!ref);
7541
7542         ext_index = -1;
7543         nritems = btrfs_header_nritems(leaf);
7544         for (i = 0; i < nritems; i++) {
7545                 btrfs_item_key_to_cpu(leaf, &key, i);
7546                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7547                         continue;
7548                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7549                 if (btrfs_file_extent_type(leaf, fi) ==
7550                     BTRFS_FILE_EXTENT_INLINE)
7551                         continue;
7552                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7553                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7554                 if (bytenr == 0)
7555                         continue;
7556
7557                 ext_index++;
7558                 if (bytenr >= group->key.objectid + group->key.offset ||
7559                     bytenr + num_bytes <= group->key.objectid)
7560                         continue;
7561
7562                 extent_key.objectid = bytenr;
7563                 extent_key.offset = num_bytes;
7564                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7565                 nr_extent = 1;
7566                 ret = get_new_locations(reloc_inode, &extent_key,
7567                                         group->key.objectid, 1,
7568                                         &new_extent, &nr_extent);
7569                 if (ret > 0)
7570                         continue;
7571                 BUG_ON(ret < 0);
7572
7573                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7574                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7575                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7576                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7577
7578                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7579                                                 new_extent->disk_bytenr);
7580                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7581                                                 new_extent->disk_num_bytes);
7582                 btrfs_mark_buffer_dirty(leaf);
7583
7584                 ret = btrfs_inc_extent_ref(trans, root,
7585                                         new_extent->disk_bytenr,
7586                                         new_extent->disk_num_bytes,
7587                                         leaf->start,
7588                                         root->root_key.objectid,
7589                                         trans->transid, key.objectid);
7590                 BUG_ON(ret);
7591
7592                 ret = btrfs_free_extent(trans, root,
7593                                         bytenr, num_bytes, leaf->start,
7594                                         btrfs_header_owner(leaf),
7595                                         btrfs_header_generation(leaf),
7596                                         key.objectid, 0);
7597                 BUG_ON(ret);
7598                 cond_resched();
7599         }
7600         kfree(new_extent);
7601         BUG_ON(ext_index + 1 != ref->nritems);
7602         btrfs_free_leaf_ref(root, ref);
7603         return 0;
7604 }
7605
7606 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7607                           struct btrfs_root *root)
7608 {
7609         struct btrfs_root *reloc_root;
7610         int ret;
7611
7612         if (root->reloc_root) {
7613                 reloc_root = root->reloc_root;
7614                 root->reloc_root = NULL;
7615                 list_add(&reloc_root->dead_list,
7616                          &root->fs_info->dead_reloc_roots);
7617
7618                 btrfs_set_root_bytenr(&reloc_root->root_item,
7619                                       reloc_root->node->start);
7620                 btrfs_set_root_level(&root->root_item,
7621                                      btrfs_header_level(reloc_root->node));
7622                 memset(&reloc_root->root_item.drop_progress, 0,
7623                         sizeof(struct btrfs_disk_key));
7624                 reloc_root->root_item.drop_level = 0;
7625
7626                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7627                                         &reloc_root->root_key,
7628                                         &reloc_root->root_item);
7629                 BUG_ON(ret);
7630         }
7631         return 0;
7632 }
7633
7634 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7635 {
7636         struct btrfs_trans_handle *trans;
7637         struct btrfs_root *reloc_root;
7638         struct btrfs_root *prev_root = NULL;
7639         struct list_head dead_roots;
7640         int ret;
7641         unsigned long nr;
7642
7643         INIT_LIST_HEAD(&dead_roots);
7644         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7645
7646         while (!list_empty(&dead_roots)) {
7647                 reloc_root = list_entry(dead_roots.prev,
7648                                         struct btrfs_root, dead_list);
7649                 list_del_init(&reloc_root->dead_list);
7650
7651                 BUG_ON(reloc_root->commit_root != NULL);
7652                 while (1) {
7653                         trans = btrfs_join_transaction(root);
7654                         BUG_ON(IS_ERR(trans));
7655
7656                         mutex_lock(&root->fs_info->drop_mutex);
7657                         ret = btrfs_drop_snapshot(trans, reloc_root);
7658                         if (ret != -EAGAIN)
7659                                 break;
7660                         mutex_unlock(&root->fs_info->drop_mutex);
7661
7662                         nr = trans->blocks_used;
7663                         ret = btrfs_end_transaction(trans, root);
7664                         BUG_ON(ret);
7665                         btrfs_btree_balance_dirty(root, nr);
7666                 }
7667
7668                 free_extent_buffer(reloc_root->node);
7669
7670                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7671                                      &reloc_root->root_key);
7672                 BUG_ON(ret);
7673                 mutex_unlock(&root->fs_info->drop_mutex);
7674
7675                 nr = trans->blocks_used;
7676                 ret = btrfs_end_transaction(trans, root);
7677                 BUG_ON(ret);
7678                 btrfs_btree_balance_dirty(root, nr);
7679
7680                 kfree(prev_root);
7681                 prev_root = reloc_root;
7682         }
7683         if (prev_root) {
7684                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7685                 kfree(prev_root);
7686         }
7687         return 0;
7688 }
7689
7690 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7691 {
7692         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7693         return 0;
7694 }
7695
7696 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7697 {
7698         struct btrfs_root *reloc_root;
7699         struct btrfs_trans_handle *trans;
7700         struct btrfs_key location;
7701         int found;
7702         int ret;
7703
7704         mutex_lock(&root->fs_info->tree_reloc_mutex);
7705         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7706         BUG_ON(ret);
7707         found = !list_empty(&root->fs_info->dead_reloc_roots);
7708         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7709
7710         if (found) {
7711                 trans = btrfs_start_transaction(root, 1);
7712                 BUG_ON(IS_ERR(trans));
7713                 ret = btrfs_commit_transaction(trans, root);
7714                 BUG_ON(ret);
7715         }
7716
7717         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7718         location.offset = (u64)-1;
7719         location.type = BTRFS_ROOT_ITEM_KEY;
7720
7721         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7722         BUG_ON(!reloc_root);
7723         ret = btrfs_orphan_cleanup(reloc_root);
7724         BUG_ON(ret);
7725         return 0;
7726 }
7727
7728 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7729                                     struct btrfs_root *root)
7730 {
7731         struct btrfs_root *reloc_root;
7732         struct extent_buffer *eb;
7733         struct btrfs_root_item *root_item;
7734         struct btrfs_key root_key;
7735         int ret;
7736
7737         BUG_ON(!root->ref_cows);
7738         if (root->reloc_root)
7739                 return 0;
7740
7741         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7742         if (!root_item)
7743                 return -ENOMEM;
7744
7745         ret = btrfs_copy_root(trans, root, root->commit_root,
7746                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7747         BUG_ON(ret);
7748
7749         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7750         root_key.offset = root->root_key.objectid;
7751         root_key.type = BTRFS_ROOT_ITEM_KEY;
7752
7753         memcpy(root_item, &root->root_item, sizeof(root_item));
7754         btrfs_set_root_refs(root_item, 0);
7755         btrfs_set_root_bytenr(root_item, eb->start);
7756         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7757         btrfs_set_root_generation(root_item, trans->transid);
7758
7759         btrfs_tree_unlock(eb);
7760         free_extent_buffer(eb);
7761
7762         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7763                                 &root_key, root_item);
7764         BUG_ON(ret);
7765         kfree(root_item);
7766
7767         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7768                                                  &root_key);
7769         BUG_ON(IS_ERR(reloc_root));
7770         reloc_root->last_trans = trans->transid;
7771         reloc_root->commit_root = NULL;
7772         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7773
7774         root->reloc_root = reloc_root;
7775         return 0;
7776 }
7777
7778 /*
7779  * Core function of space balance.
7780  *
7781  * The idea is using reloc trees to relocate tree blocks in reference
7782  * counted roots. There is one reloc tree for each subvol, and all
7783  * reloc trees share same root key objectid. Reloc trees are snapshots
7784  * of the latest committed roots of subvols (root->commit_root).
7785  *
7786  * To relocate a tree block referenced by a subvol, there are two steps.
7787  * COW the block through subvol's reloc tree, then update block pointer
7788  * in the subvol to point to the new block. Since all reloc trees share
7789  * same root key objectid, doing special handing for tree blocks owned
7790  * by them is easy. Once a tree block has been COWed in one reloc tree,
7791  * we can use the resulting new block directly when the same block is
7792  * required to COW again through other reloc trees. By this way, relocated
7793  * tree blocks are shared between reloc trees, so they are also shared
7794  * between subvols.
7795  */
7796 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7797                                       struct btrfs_root *root,
7798                                       struct btrfs_path *path,
7799                                       struct btrfs_key *first_key,
7800                                       struct btrfs_ref_path *ref_path,
7801                                       struct btrfs_block_group_cache *group,
7802                                       struct inode *reloc_inode)
7803 {
7804         struct btrfs_root *reloc_root;
7805         struct extent_buffer *eb = NULL;
7806         struct btrfs_key *keys;
7807         u64 *nodes;
7808         int level;
7809         int shared_level;
7810         int lowest_level = 0;
7811         int ret;
7812
7813         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7814                 lowest_level = ref_path->owner_objectid;
7815
7816         if (!root->ref_cows) {
7817                 path->lowest_level = lowest_level;
7818                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7819                 BUG_ON(ret < 0);
7820                 path->lowest_level = 0;
7821                 btrfs_release_path(root, path);
7822                 return 0;
7823         }
7824
7825         mutex_lock(&root->fs_info->tree_reloc_mutex);
7826         ret = init_reloc_tree(trans, root);
7827         BUG_ON(ret);
7828         reloc_root = root->reloc_root;
7829
7830         shared_level = ref_path->shared_level;
7831         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7832
7833         keys = ref_path->node_keys;
7834         nodes = ref_path->new_nodes;
7835         memset(&keys[shared_level + 1], 0,
7836                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7837         memset(&nodes[shared_level + 1], 0,
7838                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7839
7840         if (nodes[lowest_level] == 0) {
7841                 path->lowest_level = lowest_level;
7842                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7843                                         0, 1);
7844                 BUG_ON(ret);
7845                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7846                         eb = path->nodes[level];
7847                         if (!eb || eb == reloc_root->node)
7848                                 break;
7849                         nodes[level] = eb->start;
7850                         if (level == 0)
7851                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7852                         else
7853                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7854                 }
7855                 if (nodes[0] &&
7856                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7857                         eb = path->nodes[0];
7858                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7859                                                       group, reloc_inode);
7860                         BUG_ON(ret);
7861                 }
7862                 btrfs_release_path(reloc_root, path);
7863         } else {
7864                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7865                                        lowest_level);
7866                 BUG_ON(ret);
7867         }
7868
7869         /*
7870          * replace tree blocks in the fs tree with tree blocks in
7871          * the reloc tree.
7872          */
7873         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7874         BUG_ON(ret < 0);
7875
7876         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7877                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7878                                         0, 0);
7879                 BUG_ON(ret);
7880                 extent_buffer_get(path->nodes[0]);
7881                 eb = path->nodes[0];
7882                 btrfs_release_path(reloc_root, path);
7883                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7884                 BUG_ON(ret);
7885                 free_extent_buffer(eb);
7886         }
7887
7888         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7889         path->lowest_level = 0;
7890         return 0;
7891 }
7892
7893 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7894                                         struct btrfs_root *root,
7895                                         struct btrfs_path *path,
7896                                         struct btrfs_key *first_key,
7897                                         struct btrfs_ref_path *ref_path)
7898 {
7899         int ret;
7900
7901         ret = relocate_one_path(trans, root, path, first_key,
7902                                 ref_path, NULL, NULL);
7903         BUG_ON(ret);
7904
7905         return 0;
7906 }
7907
7908 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7909                                     struct btrfs_root *extent_root,
7910                                     struct btrfs_path *path,
7911                                     struct btrfs_key *extent_key)
7912 {
7913         int ret;
7914
7915         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7916         if (ret)
7917                 goto out;
7918         ret = btrfs_del_item(trans, extent_root, path);
7919 out:
7920         btrfs_release_path(extent_root, path);
7921         return ret;
7922 }
7923
7924 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7925                                                 struct btrfs_ref_path *ref_path)
7926 {
7927         struct btrfs_key root_key;
7928
7929         root_key.objectid = ref_path->root_objectid;
7930         root_key.type = BTRFS_ROOT_ITEM_KEY;
7931         if (is_cowonly_root(ref_path->root_objectid))
7932                 root_key.offset = 0;
7933         else
7934                 root_key.offset = (u64)-1;
7935
7936         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7937 }
7938
7939 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7940                                         struct btrfs_path *path,
7941                                         struct btrfs_key *extent_key,
7942                                         struct btrfs_block_group_cache *group,
7943                                         struct inode *reloc_inode, int pass)
7944 {
7945         struct btrfs_trans_handle *trans;
7946         struct btrfs_root *found_root;
7947         struct btrfs_ref_path *ref_path = NULL;
7948         struct disk_extent *new_extents = NULL;
7949         int nr_extents = 0;
7950         int loops;
7951         int ret;
7952         int level;
7953         struct btrfs_key first_key;
7954         u64 prev_block = 0;
7955
7956
7957         trans = btrfs_start_transaction(extent_root, 1);
7958         BUG_ON(IS_ERR(trans));
7959
7960         if (extent_key->objectid == 0) {
7961                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7962                 goto out;
7963         }
7964
7965         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7966         if (!ref_path) {
7967                 ret = -ENOMEM;
7968                 goto out;
7969         }
7970
7971         for (loops = 0; ; loops++) {
7972                 if (loops == 0) {
7973                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7974                                                    extent_key->objectid);
7975                 } else {
7976                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7977                 }
7978                 if (ret < 0)
7979                         goto out;
7980                 if (ret > 0)
7981                         break;
7982
7983                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7984                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7985                         continue;
7986
7987                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7988                 BUG_ON(!found_root);
7989                 /*
7990                  * for reference counted tree, only process reference paths
7991                  * rooted at the latest committed root.
7992                  */
7993                 if (found_root->ref_cows &&
7994                     ref_path->root_generation != found_root->root_key.offset)
7995                         continue;
7996
7997                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7998                         if (pass == 0) {
7999                                 /*
8000                                  * copy data extents to new locations
8001                                  */
8002                                 u64 group_start = group->key.objectid;
8003                                 ret = relocate_data_extent(reloc_inode,
8004                                                            extent_key,
8005                                                            group_start);
8006                                 if (ret < 0)
8007                                         goto out;
8008                                 break;
8009                         }
8010                         level = 0;
8011                 } else {
8012                         level = ref_path->owner_objectid;
8013                 }
8014
8015                 if (prev_block != ref_path->nodes[level]) {
8016                         struct extent_buffer *eb;
8017                         u64 block_start = ref_path->nodes[level];
8018                         u64 block_size = btrfs_level_size(found_root, level);
8019
8020                         eb = read_tree_block(found_root, block_start,
8021                                              block_size, 0);
8022                         if (!eb) {
8023                                 ret = -EIO;
8024                                 goto out;
8025                         }
8026                         btrfs_tree_lock(eb);
8027                         BUG_ON(level != btrfs_header_level(eb));
8028
8029                         if (level == 0)
8030                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
8031                         else
8032                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
8033
8034                         btrfs_tree_unlock(eb);
8035                         free_extent_buffer(eb);
8036                         prev_block = block_start;
8037                 }
8038
8039                 mutex_lock(&extent_root->fs_info->trans_mutex);
8040                 btrfs_record_root_in_trans(found_root);
8041                 mutex_unlock(&extent_root->fs_info->trans_mutex);
8042                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
8043                         /*
8044                          * try to update data extent references while
8045                          * keeping metadata shared between snapshots.
8046                          */
8047                         if (pass == 1) {
8048                                 ret = relocate_one_path(trans, found_root,
8049                                                 path, &first_key, ref_path,
8050                                                 group, reloc_inode);
8051                                 if (ret < 0)
8052                                         goto out;
8053                                 continue;
8054                         }
8055                         /*
8056                          * use fallback method to process the remaining
8057                          * references.
8058                          */
8059                         if (!new_extents) {
8060                                 u64 group_start = group->key.objectid;
8061                                 new_extents = kmalloc(sizeof(*new_extents),
8062                                                       GFP_NOFS);
8063                                 if (!new_extents) {
8064                                         ret = -ENOMEM;
8065                                         goto out;
8066                                 }
8067                                 nr_extents = 1;
8068                                 ret = get_new_locations(reloc_inode,
8069                                                         extent_key,
8070                                                         group_start, 1,
8071                                                         &new_extents,
8072                                                         &nr_extents);
8073                                 if (ret)
8074                                         goto out;
8075                         }
8076                         ret = replace_one_extent(trans, found_root,
8077                                                 path, extent_key,
8078                                                 &first_key, ref_path,
8079                                                 new_extents, nr_extents);
8080                 } else {
8081                         ret = relocate_tree_block(trans, found_root, path,
8082                                                   &first_key, ref_path);
8083                 }
8084                 if (ret < 0)
8085                         goto out;
8086         }
8087         ret = 0;
8088 out:
8089         btrfs_end_transaction(trans, extent_root);
8090         kfree(new_extents);
8091         kfree(ref_path);
8092         return ret;
8093 }
8094 #endif
8095
8096 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8097 {
8098         u64 num_devices;
8099         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8100                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8101
8102         /*
8103          * we add in the count of missing devices because we want
8104          * to make sure that any RAID levels on a degraded FS
8105          * continue to be honored.
8106          */
8107         num_devices = root->fs_info->fs_devices->rw_devices +
8108                 root->fs_info->fs_devices->missing_devices;
8109
8110         if (num_devices == 1) {
8111                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8112                 stripped = flags & ~stripped;
8113
8114                 /* turn raid0 into single device chunks */
8115                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8116                         return stripped;
8117
8118                 /* turn mirroring into duplication */
8119                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8120                              BTRFS_BLOCK_GROUP_RAID10))
8121                         return stripped | BTRFS_BLOCK_GROUP_DUP;
8122                 return flags;
8123         } else {
8124                 /* they already had raid on here, just return */
8125                 if (flags & stripped)
8126                         return flags;
8127
8128                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8129                 stripped = flags & ~stripped;
8130
8131                 /* switch duplicated blocks with raid1 */
8132                 if (flags & BTRFS_BLOCK_GROUP_DUP)
8133                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
8134
8135                 /* turn single device chunks into raid0 */
8136                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8137         }
8138         return flags;
8139 }
8140
8141 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
8142 {
8143         struct btrfs_space_info *sinfo = cache->space_info;
8144         u64 num_bytes;
8145         int ret = -ENOSPC;
8146
8147         if (cache->ro)
8148                 return 0;
8149
8150         spin_lock(&sinfo->lock);
8151         spin_lock(&cache->lock);
8152         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8153                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8154
8155         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8156             sinfo->bytes_may_use + sinfo->bytes_readonly +
8157             cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
8158                 sinfo->bytes_readonly += num_bytes;
8159                 sinfo->bytes_reserved += cache->reserved_pinned;
8160                 cache->reserved_pinned = 0;
8161                 cache->ro = 1;
8162                 ret = 0;
8163         }
8164
8165         spin_unlock(&cache->lock);
8166         spin_unlock(&sinfo->lock);
8167         return ret;
8168 }
8169
8170 int btrfs_set_block_group_ro(struct btrfs_root *root,
8171                              struct btrfs_block_group_cache *cache)
8172
8173 {
8174         struct btrfs_trans_handle *trans;
8175         u64 alloc_flags;
8176         int ret;
8177
8178         BUG_ON(cache->ro);
8179
8180         trans = btrfs_join_transaction(root);
8181         BUG_ON(IS_ERR(trans));
8182
8183         alloc_flags = update_block_group_flags(root, cache->flags);
8184         if (alloc_flags != cache->flags)
8185                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8186                                CHUNK_ALLOC_FORCE);
8187
8188         ret = set_block_group_ro(cache);
8189         if (!ret)
8190                 goto out;
8191         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8192         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8193                              CHUNK_ALLOC_FORCE);
8194         if (ret < 0)
8195                 goto out;
8196         ret = set_block_group_ro(cache);
8197 out:
8198         btrfs_end_transaction(trans, root);
8199         return ret;
8200 }
8201
8202 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8203                             struct btrfs_root *root, u64 type)
8204 {
8205         u64 alloc_flags = get_alloc_profile(root, type);
8206         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8207                               CHUNK_ALLOC_FORCE);
8208 }
8209
8210 /*
8211  * helper to account the unused space of all the readonly block group in the
8212  * list. takes mirrors into account.
8213  */
8214 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8215 {
8216         struct btrfs_block_group_cache *block_group;
8217         u64 free_bytes = 0;
8218         int factor;
8219
8220         list_for_each_entry(block_group, groups_list, list) {
8221                 spin_lock(&block_group->lock);
8222
8223                 if (!block_group->ro) {
8224                         spin_unlock(&block_group->lock);
8225                         continue;
8226                 }
8227
8228                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8229                                           BTRFS_BLOCK_GROUP_RAID10 |
8230                                           BTRFS_BLOCK_GROUP_DUP))
8231                         factor = 2;
8232                 else
8233                         factor = 1;
8234
8235                 free_bytes += (block_group->key.offset -
8236                                btrfs_block_group_used(&block_group->item)) *
8237                                factor;
8238
8239                 spin_unlock(&block_group->lock);
8240         }
8241
8242         return free_bytes;
8243 }
8244
8245 /*
8246  * helper to account the unused space of all the readonly block group in the
8247  * space_info. takes mirrors into account.
8248  */
8249 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8250 {
8251         int i;
8252         u64 free_bytes = 0;
8253
8254         spin_lock(&sinfo->lock);
8255
8256         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8257                 if (!list_empty(&sinfo->block_groups[i]))
8258                         free_bytes += __btrfs_get_ro_block_group_free_space(
8259                                                 &sinfo->block_groups[i]);
8260
8261         spin_unlock(&sinfo->lock);
8262
8263         return free_bytes;
8264 }
8265
8266 int btrfs_set_block_group_rw(struct btrfs_root *root,
8267                               struct btrfs_block_group_cache *cache)
8268 {
8269         struct btrfs_space_info *sinfo = cache->space_info;
8270         u64 num_bytes;
8271
8272         BUG_ON(!cache->ro);
8273
8274         spin_lock(&sinfo->lock);
8275         spin_lock(&cache->lock);
8276         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8277                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8278         sinfo->bytes_readonly -= num_bytes;
8279         cache->ro = 0;
8280         spin_unlock(&cache->lock);
8281         spin_unlock(&sinfo->lock);
8282         return 0;
8283 }
8284
8285 /*
8286  * checks to see if its even possible to relocate this block group.
8287  *
8288  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8289  * ok to go ahead and try.
8290  */
8291 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8292 {
8293         struct btrfs_block_group_cache *block_group;
8294         struct btrfs_space_info *space_info;
8295         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8296         struct btrfs_device *device;
8297         int full = 0;
8298         int ret = 0;
8299
8300         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8301
8302         /* odd, couldn't find the block group, leave it alone */
8303         if (!block_group)
8304                 return -1;
8305
8306         /* no bytes used, we're good */
8307         if (!btrfs_block_group_used(&block_group->item))
8308                 goto out;
8309
8310         space_info = block_group->space_info;
8311         spin_lock(&space_info->lock);
8312
8313         full = space_info->full;
8314
8315         /*
8316          * if this is the last block group we have in this space, we can't
8317          * relocate it unless we're able to allocate a new chunk below.
8318          *
8319          * Otherwise, we need to make sure we have room in the space to handle
8320          * all of the extents from this block group.  If we can, we're good
8321          */
8322         if ((space_info->total_bytes != block_group->key.offset) &&
8323            (space_info->bytes_used + space_info->bytes_reserved +
8324             space_info->bytes_pinned + space_info->bytes_readonly +
8325             btrfs_block_group_used(&block_group->item) <
8326             space_info->total_bytes)) {
8327                 spin_unlock(&space_info->lock);
8328                 goto out;
8329         }
8330         spin_unlock(&space_info->lock);
8331
8332         /*
8333          * ok we don't have enough space, but maybe we have free space on our
8334          * devices to allocate new chunks for relocation, so loop through our
8335          * alloc devices and guess if we have enough space.  However, if we
8336          * were marked as full, then we know there aren't enough chunks, and we
8337          * can just return.
8338          */
8339         ret = -1;
8340         if (full)
8341                 goto out;
8342
8343         mutex_lock(&root->fs_info->chunk_mutex);
8344         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8345                 u64 min_free = btrfs_block_group_used(&block_group->item);
8346                 u64 dev_offset;
8347
8348                 /*
8349                  * check to make sure we can actually find a chunk with enough
8350                  * space to fit our block group in.
8351                  */
8352                 if (device->total_bytes > device->bytes_used + min_free) {
8353                         ret = find_free_dev_extent(NULL, device, min_free,
8354                                                    &dev_offset, NULL);
8355                         if (!ret)
8356                                 break;
8357                         ret = -1;
8358                 }
8359         }
8360         mutex_unlock(&root->fs_info->chunk_mutex);
8361 out:
8362         btrfs_put_block_group(block_group);
8363         return ret;
8364 }
8365
8366 static int find_first_block_group(struct btrfs_root *root,
8367                 struct btrfs_path *path, struct btrfs_key *key)
8368 {
8369         int ret = 0;
8370         struct btrfs_key found_key;
8371         struct extent_buffer *leaf;
8372         int slot;
8373
8374         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8375         if (ret < 0)
8376                 goto out;
8377
8378         while (1) {
8379                 slot = path->slots[0];
8380                 leaf = path->nodes[0];
8381                 if (slot >= btrfs_header_nritems(leaf)) {
8382                         ret = btrfs_next_leaf(root, path);
8383                         if (ret == 0)
8384                                 continue;
8385                         if (ret < 0)
8386                                 goto out;
8387                         break;
8388                 }
8389                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8390
8391                 if (found_key.objectid >= key->objectid &&
8392                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8393                         ret = 0;
8394                         goto out;
8395                 }
8396                 path->slots[0]++;
8397         }
8398 out:
8399         return ret;
8400 }
8401
8402 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8403 {
8404         struct btrfs_block_group_cache *block_group;
8405         u64 last = 0;
8406
8407         while (1) {
8408                 struct inode *inode;
8409
8410                 block_group = btrfs_lookup_first_block_group(info, last);
8411                 while (block_group) {
8412                         spin_lock(&block_group->lock);
8413                         if (block_group->iref)
8414                                 break;
8415                         spin_unlock(&block_group->lock);
8416                         block_group = next_block_group(info->tree_root,
8417                                                        block_group);
8418                 }
8419                 if (!block_group) {
8420                         if (last == 0)
8421                                 break;
8422                         last = 0;
8423                         continue;
8424                 }
8425
8426                 inode = block_group->inode;
8427                 block_group->iref = 0;
8428                 block_group->inode = NULL;
8429                 spin_unlock(&block_group->lock);
8430                 iput(inode);
8431                 last = block_group->key.objectid + block_group->key.offset;
8432                 btrfs_put_block_group(block_group);
8433         }
8434 }
8435
8436 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8437 {
8438         struct btrfs_block_group_cache *block_group;
8439         struct btrfs_space_info *space_info;
8440         struct btrfs_caching_control *caching_ctl;
8441         struct rb_node *n;
8442
8443         down_write(&info->extent_commit_sem);
8444         while (!list_empty(&info->caching_block_groups)) {
8445                 caching_ctl = list_entry(info->caching_block_groups.next,
8446                                          struct btrfs_caching_control, list);
8447                 list_del(&caching_ctl->list);
8448                 put_caching_control(caching_ctl);
8449         }
8450         up_write(&info->extent_commit_sem);
8451
8452         spin_lock(&info->block_group_cache_lock);
8453         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8454                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8455                                        cache_node);
8456                 rb_erase(&block_group->cache_node,
8457                          &info->block_group_cache_tree);
8458                 spin_unlock(&info->block_group_cache_lock);
8459
8460                 down_write(&block_group->space_info->groups_sem);
8461                 list_del(&block_group->list);
8462                 up_write(&block_group->space_info->groups_sem);
8463
8464                 if (block_group->cached == BTRFS_CACHE_STARTED)
8465                         wait_block_group_cache_done(block_group);
8466
8467                 /*
8468                  * We haven't cached this block group, which means we could
8469                  * possibly have excluded extents on this block group.
8470                  */
8471                 if (block_group->cached == BTRFS_CACHE_NO)
8472                         free_excluded_extents(info->extent_root, block_group);
8473
8474                 btrfs_remove_free_space_cache(block_group);
8475                 btrfs_put_block_group(block_group);
8476
8477                 spin_lock(&info->block_group_cache_lock);
8478         }
8479         spin_unlock(&info->block_group_cache_lock);
8480
8481         /* now that all the block groups are freed, go through and
8482          * free all the space_info structs.  This is only called during
8483          * the final stages of unmount, and so we know nobody is
8484          * using them.  We call synchronize_rcu() once before we start,
8485          * just to be on the safe side.
8486          */
8487         synchronize_rcu();
8488
8489         release_global_block_rsv(info);
8490
8491         while(!list_empty(&info->space_info)) {
8492                 space_info = list_entry(info->space_info.next,
8493                                         struct btrfs_space_info,
8494                                         list);
8495                 if (space_info->bytes_pinned > 0 ||
8496                     space_info->bytes_reserved > 0) {
8497                         WARN_ON(1);
8498                         dump_space_info(space_info, 0, 0);
8499                 }
8500                 list_del(&space_info->list);
8501                 kfree(space_info);
8502         }
8503         return 0;
8504 }
8505
8506 static void __link_block_group(struct btrfs_space_info *space_info,
8507                                struct btrfs_block_group_cache *cache)
8508 {
8509         int index = get_block_group_index(cache);
8510
8511         down_write(&space_info->groups_sem);
8512         list_add_tail(&cache->list, &space_info->block_groups[index]);
8513         up_write(&space_info->groups_sem);
8514 }
8515
8516 int btrfs_read_block_groups(struct btrfs_root *root)
8517 {
8518         struct btrfs_path *path;
8519         int ret;
8520         struct btrfs_block_group_cache *cache;
8521         struct btrfs_fs_info *info = root->fs_info;
8522         struct btrfs_space_info *space_info;
8523         struct btrfs_key key;
8524         struct btrfs_key found_key;
8525         struct extent_buffer *leaf;
8526         int need_clear = 0;
8527         u64 cache_gen;
8528
8529         root = info->extent_root;
8530         key.objectid = 0;
8531         key.offset = 0;
8532         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8533         path = btrfs_alloc_path();
8534         if (!path)
8535                 return -ENOMEM;
8536
8537         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8538         if (cache_gen != 0 &&
8539             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8540                 need_clear = 1;
8541         if (btrfs_test_opt(root, CLEAR_CACHE))
8542                 need_clear = 1;
8543         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8544                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8545
8546         while (1) {
8547                 ret = find_first_block_group(root, path, &key);
8548                 if (ret > 0)
8549                         break;
8550                 if (ret != 0)
8551                         goto error;
8552                 leaf = path->nodes[0];
8553                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8554                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8555                 if (!cache) {
8556                         ret = -ENOMEM;
8557                         goto error;
8558                 }
8559
8560                 atomic_set(&cache->count, 1);
8561                 spin_lock_init(&cache->lock);
8562                 spin_lock_init(&cache->tree_lock);
8563                 cache->fs_info = info;
8564                 INIT_LIST_HEAD(&cache->list);
8565                 INIT_LIST_HEAD(&cache->cluster_list);
8566
8567                 if (need_clear)
8568                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8569
8570                 /*
8571                  * we only want to have 32k of ram per block group for keeping
8572                  * track of free space, and if we pass 1/2 of that we want to
8573                  * start converting things over to using bitmaps
8574                  */
8575                 cache->extents_thresh = ((1024 * 32) / 2) /
8576                         sizeof(struct btrfs_free_space);
8577
8578                 read_extent_buffer(leaf, &cache->item,
8579                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8580                                    sizeof(cache->item));
8581                 memcpy(&cache->key, &found_key, sizeof(found_key));
8582
8583                 key.objectid = found_key.objectid + found_key.offset;
8584                 btrfs_release_path(root, path);
8585                 cache->flags = btrfs_block_group_flags(&cache->item);
8586                 cache->sectorsize = root->sectorsize;
8587
8588                 /*
8589                  * We need to exclude the super stripes now so that the space
8590                  * info has super bytes accounted for, otherwise we'll think
8591                  * we have more space than we actually do.
8592                  */
8593                 exclude_super_stripes(root, cache);
8594
8595                 /*
8596                  * check for two cases, either we are full, and therefore
8597                  * don't need to bother with the caching work since we won't
8598                  * find any space, or we are empty, and we can just add all
8599                  * the space in and be done with it.  This saves us _alot_ of
8600                  * time, particularly in the full case.
8601                  */
8602                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8603                         cache->last_byte_to_unpin = (u64)-1;
8604                         cache->cached = BTRFS_CACHE_FINISHED;
8605                         free_excluded_extents(root, cache);
8606                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8607                         cache->last_byte_to_unpin = (u64)-1;
8608                         cache->cached = BTRFS_CACHE_FINISHED;
8609                         add_new_free_space(cache, root->fs_info,
8610                                            found_key.objectid,
8611                                            found_key.objectid +
8612                                            found_key.offset);
8613                         free_excluded_extents(root, cache);
8614                 }
8615
8616                 ret = update_space_info(info, cache->flags, found_key.offset,
8617                                         btrfs_block_group_used(&cache->item),
8618                                         &space_info);
8619                 BUG_ON(ret);
8620                 cache->space_info = space_info;
8621                 spin_lock(&cache->space_info->lock);
8622                 cache->space_info->bytes_readonly += cache->bytes_super;
8623                 spin_unlock(&cache->space_info->lock);
8624
8625                 __link_block_group(space_info, cache);
8626
8627                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8628                 BUG_ON(ret);
8629
8630                 set_avail_alloc_bits(root->fs_info, cache->flags);
8631                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8632                         set_block_group_ro(cache);
8633         }
8634
8635         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8636                 if (!(get_alloc_profile(root, space_info->flags) &
8637                       (BTRFS_BLOCK_GROUP_RAID10 |
8638                        BTRFS_BLOCK_GROUP_RAID1 |
8639                        BTRFS_BLOCK_GROUP_DUP)))
8640                         continue;
8641                 /*
8642                  * avoid allocating from un-mirrored block group if there are
8643                  * mirrored block groups.
8644                  */
8645                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8646                         set_block_group_ro(cache);
8647                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8648                         set_block_group_ro(cache);
8649         }
8650
8651         init_global_block_rsv(info);
8652         ret = 0;
8653 error:
8654         btrfs_free_path(path);
8655         return ret;
8656 }
8657
8658 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8659                            struct btrfs_root *root, u64 bytes_used,
8660                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8661                            u64 size)
8662 {
8663         int ret;
8664         struct btrfs_root *extent_root;
8665         struct btrfs_block_group_cache *cache;
8666
8667         extent_root = root->fs_info->extent_root;
8668
8669         root->fs_info->last_trans_log_full_commit = trans->transid;
8670
8671         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8672         if (!cache)
8673                 return -ENOMEM;
8674
8675         cache->key.objectid = chunk_offset;
8676         cache->key.offset = size;
8677         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8678         cache->sectorsize = root->sectorsize;
8679         cache->fs_info = root->fs_info;
8680
8681         /*
8682          * we only want to have 32k of ram per block group for keeping track
8683          * of free space, and if we pass 1/2 of that we want to start
8684          * converting things over to using bitmaps
8685          */
8686         cache->extents_thresh = ((1024 * 32) / 2) /
8687                 sizeof(struct btrfs_free_space);
8688         atomic_set(&cache->count, 1);
8689         spin_lock_init(&cache->lock);
8690         spin_lock_init(&cache->tree_lock);
8691         INIT_LIST_HEAD(&cache->list);
8692         INIT_LIST_HEAD(&cache->cluster_list);
8693
8694         btrfs_set_block_group_used(&cache->item, bytes_used);
8695         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8696         cache->flags = type;
8697         btrfs_set_block_group_flags(&cache->item, type);
8698
8699         cache->last_byte_to_unpin = (u64)-1;
8700         cache->cached = BTRFS_CACHE_FINISHED;
8701         exclude_super_stripes(root, cache);
8702
8703         add_new_free_space(cache, root->fs_info, chunk_offset,
8704                            chunk_offset + size);
8705
8706         free_excluded_extents(root, cache);
8707
8708         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8709                                 &cache->space_info);
8710         BUG_ON(ret);
8711
8712         spin_lock(&cache->space_info->lock);
8713         cache->space_info->bytes_readonly += cache->bytes_super;
8714         spin_unlock(&cache->space_info->lock);
8715
8716         __link_block_group(cache->space_info, cache);
8717
8718         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8719         BUG_ON(ret);
8720
8721         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8722                                 sizeof(cache->item));
8723         BUG_ON(ret);
8724
8725         set_avail_alloc_bits(extent_root->fs_info, type);
8726
8727         return 0;
8728 }
8729
8730 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8731                              struct btrfs_root *root, u64 group_start)
8732 {
8733         struct btrfs_path *path;
8734         struct btrfs_block_group_cache *block_group;
8735         struct btrfs_free_cluster *cluster;
8736         struct btrfs_root *tree_root = root->fs_info->tree_root;
8737         struct btrfs_key key;
8738         struct inode *inode;
8739         int ret;
8740         int factor;
8741
8742         root = root->fs_info->extent_root;
8743
8744         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8745         BUG_ON(!block_group);
8746         BUG_ON(!block_group->ro);
8747
8748         /*
8749          * Free the reserved super bytes from this block group before
8750          * remove it.
8751          */
8752         free_excluded_extents(root, block_group);
8753
8754         memcpy(&key, &block_group->key, sizeof(key));
8755         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8756                                   BTRFS_BLOCK_GROUP_RAID1 |
8757                                   BTRFS_BLOCK_GROUP_RAID10))
8758                 factor = 2;
8759         else
8760                 factor = 1;
8761
8762         /* make sure this block group isn't part of an allocation cluster */
8763         cluster = &root->fs_info->data_alloc_cluster;
8764         spin_lock(&cluster->refill_lock);
8765         btrfs_return_cluster_to_free_space(block_group, cluster);
8766         spin_unlock(&cluster->refill_lock);
8767
8768         /*
8769          * make sure this block group isn't part of a metadata
8770          * allocation cluster
8771          */
8772         cluster = &root->fs_info->meta_alloc_cluster;
8773         spin_lock(&cluster->refill_lock);
8774         btrfs_return_cluster_to_free_space(block_group, cluster);
8775         spin_unlock(&cluster->refill_lock);
8776
8777         path = btrfs_alloc_path();
8778         BUG_ON(!path);
8779
8780         inode = lookup_free_space_inode(root, block_group, path);
8781         if (!IS_ERR(inode)) {
8782                 btrfs_orphan_add(trans, inode);
8783                 clear_nlink(inode);
8784                 /* One for the block groups ref */
8785                 spin_lock(&block_group->lock);
8786                 if (block_group->iref) {
8787                         block_group->iref = 0;
8788                         block_group->inode = NULL;
8789                         spin_unlock(&block_group->lock);
8790                         iput(inode);
8791                 } else {
8792                         spin_unlock(&block_group->lock);
8793                 }
8794                 /* One for our lookup ref */
8795                 iput(inode);
8796         }
8797
8798         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8799         key.offset = block_group->key.objectid;
8800         key.type = 0;
8801
8802         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8803         if (ret < 0)
8804                 goto out;
8805         if (ret > 0)
8806                 btrfs_release_path(tree_root, path);
8807         if (ret == 0) {
8808                 ret = btrfs_del_item(trans, tree_root, path);
8809                 if (ret)
8810                         goto out;
8811                 btrfs_release_path(tree_root, path);
8812         }
8813
8814         spin_lock(&root->fs_info->block_group_cache_lock);
8815         rb_erase(&block_group->cache_node,
8816                  &root->fs_info->block_group_cache_tree);
8817         spin_unlock(&root->fs_info->block_group_cache_lock);
8818
8819         down_write(&block_group->space_info->groups_sem);
8820         /*
8821          * we must use list_del_init so people can check to see if they
8822          * are still on the list after taking the semaphore
8823          */
8824         list_del_init(&block_group->list);
8825         up_write(&block_group->space_info->groups_sem);
8826
8827         if (block_group->cached == BTRFS_CACHE_STARTED)
8828                 wait_block_group_cache_done(block_group);
8829
8830         btrfs_remove_free_space_cache(block_group);
8831
8832         spin_lock(&block_group->space_info->lock);
8833         block_group->space_info->total_bytes -= block_group->key.offset;
8834         block_group->space_info->bytes_readonly -= block_group->key.offset;
8835         block_group->space_info->disk_total -= block_group->key.offset * factor;
8836         spin_unlock(&block_group->space_info->lock);
8837
8838         memcpy(&key, &block_group->key, sizeof(key));
8839
8840         btrfs_clear_space_info_full(root->fs_info);
8841
8842         btrfs_put_block_group(block_group);
8843         btrfs_put_block_group(block_group);
8844
8845         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8846         if (ret > 0)
8847                 ret = -EIO;
8848         if (ret < 0)
8849                 goto out;
8850
8851         ret = btrfs_del_item(trans, root, path);
8852 out:
8853         btrfs_free_path(path);
8854         return ret;
8855 }
8856
8857 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8858 {
8859         struct btrfs_space_info *space_info;
8860         struct btrfs_super_block *disk_super;
8861         u64 features;
8862         u64 flags;
8863         int mixed = 0;
8864         int ret;
8865
8866         disk_super = &fs_info->super_copy;
8867         if (!btrfs_super_root(disk_super))
8868                 return 1;
8869
8870         features = btrfs_super_incompat_flags(disk_super);
8871         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8872                 mixed = 1;
8873
8874         flags = BTRFS_BLOCK_GROUP_SYSTEM;
8875         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8876         if (ret)
8877                 goto out;
8878
8879         if (mixed) {
8880                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8881                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8882         } else {
8883                 flags = BTRFS_BLOCK_GROUP_METADATA;
8884                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8885                 if (ret)
8886                         goto out;
8887
8888                 flags = BTRFS_BLOCK_GROUP_DATA;
8889                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8890         }
8891 out:
8892         return ret;
8893 }
8894
8895 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8896 {
8897         return unpin_extent_range(root, start, end);
8898 }
8899
8900 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8901                                u64 num_bytes, u64 *actual_bytes)
8902 {
8903         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8904 }
8905
8906 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8907 {
8908         struct btrfs_fs_info *fs_info = root->fs_info;
8909         struct btrfs_block_group_cache *cache = NULL;
8910         u64 group_trimmed;
8911         u64 start;
8912         u64 end;
8913         u64 trimmed = 0;
8914         int ret = 0;
8915
8916         cache = btrfs_lookup_block_group(fs_info, range->start);
8917
8918         while (cache) {
8919                 if (cache->key.objectid >= (range->start + range->len)) {
8920                         btrfs_put_block_group(cache);
8921                         break;
8922                 }
8923
8924                 start = max(range->start, cache->key.objectid);
8925                 end = min(range->start + range->len,
8926                                 cache->key.objectid + cache->key.offset);
8927
8928                 if (end - start >= range->minlen) {
8929                         if (!block_group_cache_done(cache)) {
8930                                 ret = cache_block_group(cache, NULL, root, 0);
8931                                 if (!ret)
8932                                         wait_block_group_cache_done(cache);
8933                         }
8934                         ret = btrfs_trim_block_group(cache,
8935                                                      &group_trimmed,
8936                                                      start,
8937                                                      end,
8938                                                      range->minlen);
8939
8940                         trimmed += group_trimmed;
8941                         if (ret) {
8942                                 btrfs_put_block_group(cache);
8943                                 break;
8944                         }
8945                 }
8946
8947                 cache = next_block_group(fs_info->tree_root, cache);
8948         }
8949
8950         range->len = trimmed;
8951         return ret;
8952 }