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 <linux/ratelimit.h>
  27 #include "compat.h"
  28 #include "hash.h"
  29 #include "ctree.h"
  30 #include "disk-io.h"
  31 #include "print-tree.h"
  32 #include "transaction.h"
  33 #include "volumes.h"
  34 #include "locking.h"
  35 #include "free-space-cache.h"
  36
  37 /* control flags for do_chunk_alloc's force field
  38  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
  39  * if we really need one.
  40  *
  41  * CHUNK_ALLOC_FORCE means it must try to allocate one
  42  *
  43  * CHUNK_ALLOC_LIMITED means to only try and allocate one
  44  * if we have very few chunks already allocated.  This is
  45  * used as part of the clustering code to help make sure
  46  * we have a good pool of storage to cluster in, without
  47  * filling the FS with empty chunks
  48  *
  49  */
  50 enum {
  51         CHUNK_ALLOC_NO_FORCE = 0,
  52         CHUNK_ALLOC_FORCE = 1,
  53         CHUNK_ALLOC_LIMITED = 2,
  54 };
  55
  56 /*
  57  * Control how reservations are dealt with.
  58  *
  59  * RESERVE_FREE - freeing a reservation.
  60  * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
  61  *   ENOSPC accounting
  62  * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
  63  *   bytes_may_use as the ENOSPC accounting is done elsewhere
  64  */
  65 enum {
  66         RESERVE_FREE = 0,
  67         RESERVE_ALLOC = 1,
  68         RESERVE_ALLOC_NO_ACCOUNT = 2,
  69 };
  70
  71 static int update_block_group(struct btrfs_trans_handle *trans,
  72                               struct btrfs_root *root,
  73                               u64 bytenr, u64 num_bytes, int alloc);
  74 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  75                                 struct btrfs_root *root,
  76                                 u64 bytenr, u64 num_bytes, u64 parent,
  77                                 u64 root_objectid, u64 owner_objectid,
  78                                 u64 owner_offset, int refs_to_drop,
  79                                 struct btrfs_delayed_extent_op *extra_op);
  80 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  81                                     struct extent_buffer *leaf,
  82                                     struct btrfs_extent_item *ei);
  83 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  84                                       struct btrfs_root *root,
  85                                       u64 parent, u64 root_objectid,
  86                                       u64 flags, u64 owner, u64 offset,
  87                                       struct btrfs_key *ins, int ref_mod);
  88 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  89                                      struct btrfs_root *root,
  90                                      u64 parent, u64 root_objectid,
  91                                      u64 flags, struct btrfs_disk_key *key,
  92                                      int level, struct btrfs_key *ins);
  93 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  94                           struct btrfs_root *extent_root, u64 alloc_bytes,
  95                           u64 flags, int force);
  96 static int find_next_key(struct btrfs_path *path, int level,
  97                          struct btrfs_key *key);
  98 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  99                             int dump_block_groups);
 100 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
 101                                        u64 num_bytes, int reserve);
 102
 103 static noinline int
 104 block_group_cache_done(struct btrfs_block_group_cache *cache)
 105 {
 106         smp_mb();
 107         return cache->cached == BTRFS_CACHE_FINISHED;
 108 }
 109
 110 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 111 {
 112         return (cache->flags & bits) == bits;
 113 }
 114
 115 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
 116 {
 117         atomic_inc(&cache->count);
 118 }
 119
 120 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 121 {
 122         if (atomic_dec_and_test(&cache->count)) {
 123                 WARN_ON(cache->pinned > 0);
 124                 WARN_ON(cache->reserved > 0);
 125                 kfree(cache->free_space_ctl);
 126                 kfree(cache);
 127         }
 128 }
 129
 130 /*
 131  * this adds the block group to the fs_info rb tree for the block group
 132  * cache
 133  */
 134 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 135                                 struct btrfs_block_group_cache *block_group)
 136 {
 137         struct rb_node **p;
 138         struct rb_node *parent = NULL;
 139         struct btrfs_block_group_cache *cache;
 140
 141         spin_lock(&info->block_group_cache_lock);
 142         p = &info->block_group_cache_tree.rb_node;
 143
 144         while (*p) {
 145                 parent = *p;
 146                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 147                                  cache_node);
 148                 if (block_group->key.objectid < cache->key.objectid) {
 149                         p = &(*p)->rb_left;
 150                 } else if (block_group->key.objectid > cache->key.objectid) {
 151                         p = &(*p)->rb_right;
 152                 } else {
 153                         spin_unlock(&info->block_group_cache_lock);
 154                         return -EEXIST;
 155                 }
 156         }
 157
 158         rb_link_node(&block_group->cache_node, parent, p);
 159         rb_insert_color(&block_group->cache_node,
 160                         &info->block_group_cache_tree);
 161         spin_unlock(&info->block_group_cache_lock);
 162
 163         return 0;
 164 }
 165
 166 /*
 167  * This will return the block group at or after bytenr if contains is 0, else
 168  * it will return the block group that contains the bytenr
 169  */
 170 static struct btrfs_block_group_cache *
 171 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 172                               int contains)
 173 {
 174         struct btrfs_block_group_cache *cache, *ret = NULL;
 175         struct rb_node *n;
 176         u64 end, start;
 177
 178         spin_lock(&info->block_group_cache_lock);
 179         n = info->block_group_cache_tree.rb_node;
 180
 181         while (n) {
 182                 cache = rb_entry(n, struct btrfs_block_group_cache,
 183                                  cache_node);
 184                 end = cache->key.objectid + cache->key.offset - 1;
 185                 start = cache->key.objectid;
 186
 187                 if (bytenr < start) {
 188                         if (!contains && (!ret || start < ret->key.objectid))
 189                                 ret = cache;
 190                         n = n->rb_left;
 191                 } else if (bytenr > start) {
 192                         if (contains && bytenr <= end) {
 193                                 ret = cache;
 194                                 break;
 195                         }
 196                         n = n->rb_right;
 197                 } else {
 198                         ret = cache;
 199                         break;
 200                 }
 201         }
 202         if (ret)
 203                 btrfs_get_block_group(ret);
 204         spin_unlock(&info->block_group_cache_lock);
 205
 206         return ret;
 207 }
 208
 209 static int add_excluded_extent(struct btrfs_root *root,
 210                                u64 start, u64 num_bytes)
 211 {
 212         u64 end = start + num_bytes - 1;
 213         set_extent_bits(&root->fs_info->freed_extents[0],
 214                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 215         set_extent_bits(&root->fs_info->freed_extents[1],
 216                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 217         return 0;
 218 }
 219
 220 static void free_excluded_extents(struct btrfs_root *root,
 221                                   struct btrfs_block_group_cache *cache)
 222 {
 223         u64 start, end;
 224
 225         start = cache->key.objectid;
 226         end = start + cache->key.offset - 1;
 227
 228         clear_extent_bits(&root->fs_info->freed_extents[0],
 229                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 230         clear_extent_bits(&root->fs_info->freed_extents[1],
 231                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 232 }
 233
 234 static int exclude_super_stripes(struct btrfs_root *root,
 235                                  struct btrfs_block_group_cache *cache)
 236 {
 237         u64 bytenr;
 238         u64 *logical;
 239         int stripe_len;
 240         int i, nr, ret;
 241
 242         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 243                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 244                 cache->bytes_super += stripe_len;
 245                 ret = add_excluded_extent(root, cache->key.objectid,
 246                                           stripe_len);
 247                 BUG_ON(ret);
 248         }
 249
 250         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 251                 bytenr = btrfs_sb_offset(i);
 252                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 253                                        cache->key.objectid, bytenr,
 254                                        0, &logical, &nr, &stripe_len);
 255                 BUG_ON(ret);
 256
 257                 while (nr--) {
 258                         cache->bytes_super += stripe_len;
 259                         ret = add_excluded_extent(root, logical[nr],
 260                                                   stripe_len);
 261                         BUG_ON(ret);
 262                 }
 263
 264                 kfree(logical);
 265         }
 266         return 0;
 267 }
 268
 269 static struct btrfs_caching_control *
 270 get_caching_control(struct btrfs_block_group_cache *cache)
 271 {
 272         struct btrfs_caching_control *ctl;
 273
 274         spin_lock(&cache->lock);
 275         if (cache->cached != BTRFS_CACHE_STARTED) {
 276                 spin_unlock(&cache->lock);
 277                 return NULL;
 278         }
 279
 280         /* We're loading it the fast way, so we don't have a caching_ctl. */
 281         if (!cache->caching_ctl) {
 282                 spin_unlock(&cache->lock);
 283                 return NULL;
 284         }
 285
 286         ctl = cache->caching_ctl;
 287         atomic_inc(&ctl->count);
 288         spin_unlock(&cache->lock);
 289         return ctl;
 290 }
 291
 292 static void put_caching_control(struct btrfs_caching_control *ctl)
 293 {
 294         if (atomic_dec_and_test(&ctl->count))
 295                 kfree(ctl);
 296 }
 297
 298 /*
 299  * this is only called by cache_block_group, since we could have freed extents
 300  * we need to check the pinned_extents for any extents that can't be used yet
 301  * since their free space will be released as soon as the transaction commits.
 302  */
 303 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 304                               struct btrfs_fs_info *info, u64 start, u64 end)
 305 {
 306         u64 extent_start, extent_end, size, total_added = 0;
 307         int ret;
 308
 309         while (start < end) {
 310                 ret = find_first_extent_bit(info->pinned_extents, start,
 311                                             &extent_start, &extent_end,
 312                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 313                 if (ret)
 314                         break;
 315
 316                 if (extent_start <= start) {
 317                         start = extent_end + 1;
 318                 } else if (extent_start > start && extent_start < end) {
 319                         size = extent_start - start;
 320                         total_added += size;
 321                         ret = btrfs_add_free_space(block_group, start,
 322                                                    size);
 323                         BUG_ON(ret);
 324                         start = extent_end + 1;
 325                 } else {
 326                         break;
 327                 }
 328         }
 329
 330         if (start < end) {
 331                 size = end - start;
 332                 total_added += size;
 333                 ret = btrfs_add_free_space(block_group, start, size);
 334                 BUG_ON(ret);
 335         }
 336
 337         return total_added;
 338 }
 339
 340 static noinline void caching_thread(struct btrfs_work *work)
 341 {
 342         struct btrfs_block_group_cache *block_group;
 343         struct btrfs_fs_info *fs_info;
 344         struct btrfs_caching_control *caching_ctl;
 345         struct btrfs_root *extent_root;
 346         struct btrfs_path *path;
 347         struct extent_buffer *leaf;
 348         struct btrfs_key key;
 349         u64 total_found = 0;
 350         u64 last = 0;
 351         u32 nritems;
 352         int ret = 0;
 353
 354         caching_ctl = container_of(work, struct btrfs_caching_control, work);
 355         block_group = caching_ctl->block_group;
 356         fs_info = block_group->fs_info;
 357         extent_root = fs_info->extent_root;
 358
 359         path = btrfs_alloc_path();
 360         if (!path)
 361                 goto out;
 362
 363         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 364
 365         /*
 366          * We don't want to deadlock with somebody trying to allocate a new
 367          * extent for the extent root while also trying to search the extent
 368          * root to add free space.  So we skip locking and search the commit
 369          * root, since its read-only
 370          */
 371         path->skip_locking = 1;
 372         path->search_commit_root = 1;
 373         path->reada = 1;
 374
 375         key.objectid = last;
 376         key.offset = 0;
 377         key.type = BTRFS_EXTENT_ITEM_KEY;
 378 again:
 379         mutex_lock(&caching_ctl->mutex);
 380         /* need to make sure the commit_root doesn't disappear */
 381         down_read(&fs_info->extent_commit_sem);
 382
 383         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 384         if (ret < 0)
 385                 goto err;
 386
 387         leaf = path->nodes[0];
 388         nritems = btrfs_header_nritems(leaf);
 389
 390         while (1) {
 391                 if (btrfs_fs_closing(fs_info) > 1) {
 392                         last = (u64)-1;
 393                         break;
 394                 }
 395
 396                 if (path->slots[0] < nritems) {
 397                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 398                 } else {
 399                         ret = find_next_key(path, 0, &key);
 400                         if (ret)
 401                                 break;
 402
 403                         if (need_resched() ||
 404                             btrfs_next_leaf(extent_root, path)) {
 405                                 caching_ctl->progress = last;
 406                                 btrfs_release_path(path);
 407                                 up_read(&fs_info->extent_commit_sem);
 408                                 mutex_unlock(&caching_ctl->mutex);
 409                                 cond_resched();
 410                                 goto again;
 411                         }
 412                         leaf = path->nodes[0];
 413                         nritems = btrfs_header_nritems(leaf);
 414                         continue;
 415                 }
 416
 417                 if (key.objectid < block_group->key.objectid) {
 418                         path->slots[0]++;
 419                         continue;
 420                 }
 421
 422                 if (key.objectid >= block_group->key.objectid +
 423                     block_group->key.offset)
 424                         break;
 425
 426                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 427                         total_found += add_new_free_space(block_group,
 428                                                           fs_info, last,
 429                                                           key.objectid);
 430                         last = key.objectid + key.offset;
 431
 432                         if (total_found > (1024 * 1024 * 2)) {
 433                                 total_found = 0;
 434                                 wake_up(&caching_ctl->wait);
 435                         }
 436                 }
 437                 path->slots[0]++;
 438         }
 439         ret = 0;
 440
 441         total_found += add_new_free_space(block_group, fs_info, last,
 442                                           block_group->key.objectid +
 443                                           block_group->key.offset);
 444         caching_ctl->progress = (u64)-1;
 445
 446         spin_lock(&block_group->lock);
 447         block_group->caching_ctl = NULL;
 448         block_group->cached = BTRFS_CACHE_FINISHED;
 449         spin_unlock(&block_group->lock);
 450
 451 err:
 452         btrfs_free_path(path);
 453         up_read(&fs_info->extent_commit_sem);
 454
 455         free_excluded_extents(extent_root, block_group);
 456
 457         mutex_unlock(&caching_ctl->mutex);
 458 out:
 459         wake_up(&caching_ctl->wait);
 460
 461         put_caching_control(caching_ctl);
 462         btrfs_put_block_group(block_group);
 463 }
 464
 465 static int cache_block_group(struct btrfs_block_group_cache *cache,
 466                              struct btrfs_trans_handle *trans,
 467                              struct btrfs_root *root,
 468                              int load_cache_only)
 469 {
 470         DEFINE_WAIT(wait);
 471         struct btrfs_fs_info *fs_info = cache->fs_info;
 472         struct btrfs_caching_control *caching_ctl;
 473         int ret = 0;
 474
 475         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 476         BUG_ON(!caching_ctl);
 477
 478         INIT_LIST_HEAD(&caching_ctl->list);
 479         mutex_init(&caching_ctl->mutex);
 480         init_waitqueue_head(&caching_ctl->wait);
 481         caching_ctl->block_group = cache;
 482         caching_ctl->progress = cache->key.objectid;
 483         atomic_set(&caching_ctl->count, 1);
 484         caching_ctl->work.func = caching_thread;
 485
 486         spin_lock(&cache->lock);
 487         /*
 488          * This should be a rare occasion, but this could happen I think in the
 489          * case where one thread starts to load the space cache info, and then
 490          * some other thread starts a transaction commit which tries to do an
 491          * allocation while the other thread is still loading the space cache
 492          * info.  The previous loop should have kept us from choosing this block
 493          * group, but if we've moved to the state where we will wait on caching
 494          * block groups we need to first check if we're doing a fast load here,
 495          * so we can wait for it to finish, otherwise we could end up allocating
 496          * from a block group who's cache gets evicted for one reason or
 497          * another.
 498          */
 499         while (cache->cached == BTRFS_CACHE_FAST) {
 500                 struct btrfs_caching_control *ctl;
 501
 502                 ctl = cache->caching_ctl;
 503                 atomic_inc(&ctl->count);
 504                 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
 505                 spin_unlock(&cache->lock);
 506
 507                 schedule();
 508
 509                 finish_wait(&ctl->wait, &wait);
 510                 put_caching_control(ctl);
 511                 spin_lock(&cache->lock);
 512         }
 513
 514         if (cache->cached != BTRFS_CACHE_NO) {
 515                 spin_unlock(&cache->lock);
 516                 kfree(caching_ctl);
 517                 return 0;
 518         }
 519         WARN_ON(cache->caching_ctl);
 520         cache->caching_ctl = caching_ctl;
 521         cache->cached = BTRFS_CACHE_FAST;
 522         spin_unlock(&cache->lock);
 523
 524         /*
 525          * We can't do the read from on-disk cache during a commit since we need
 526          * to have the normal tree locking.  Also if we are currently trying to
 527          * allocate blocks for the tree root we can't do the fast caching since
 528          * we likely hold important locks.
 529          */
 530         if (trans && (!trans->transaction->in_commit) &&
 531             (root && root != root->fs_info->tree_root) &&
 532             btrfs_test_opt(root, SPACE_CACHE)) {
 533                 ret = load_free_space_cache(fs_info, cache);
 534
 535                 spin_lock(&cache->lock);
 536                 if (ret == 1) {
 537                         cache->caching_ctl = NULL;
 538                         cache->cached = BTRFS_CACHE_FINISHED;
 539                         cache->last_byte_to_unpin = (u64)-1;
 540                 } else {
 541                         if (load_cache_only) {
 542                                 cache->caching_ctl = NULL;
 543                                 cache->cached = BTRFS_CACHE_NO;
 544                         } else {
 545                                 cache->cached = BTRFS_CACHE_STARTED;
 546                         }
 547                 }
 548                 spin_unlock(&cache->lock);
 549                 wake_up(&caching_ctl->wait);
 550                 if (ret == 1) {
 551                         put_caching_control(caching_ctl);
 552                         free_excluded_extents(fs_info->extent_root, cache);
 553                         return 0;
 554                 }
 555         } else {
 556                 /*
 557                  * We are not going to do the fast caching, set cached to the
 558                  * appropriate value and wakeup any waiters.
 559                  */
 560                 spin_lock(&cache->lock);
 561                 if (load_cache_only) {
 562                         cache->caching_ctl = NULL;
 563                         cache->cached = BTRFS_CACHE_NO;
 564                 } else {
 565                         cache->cached = BTRFS_CACHE_STARTED;
 566                 }
 567                 spin_unlock(&cache->lock);
 568                 wake_up(&caching_ctl->wait);
 569         }
 570
 571         if (load_cache_only) {
 572                 put_caching_control(caching_ctl);
 573                 return 0;
 574         }
 575
 576         down_write(&fs_info->extent_commit_sem);
 577         atomic_inc(&caching_ctl->count);
 578         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 579         up_write(&fs_info->extent_commit_sem);
 580
 581         btrfs_get_block_group(cache);
 582
 583         btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
 584
 585         return ret;
 586 }
 587
 588 /*
 589  * return the block group that starts at or after bytenr
 590  */
 591 static struct btrfs_block_group_cache *
 592 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 593 {
 594         struct btrfs_block_group_cache *cache;
 595
 596         cache = block_group_cache_tree_search(info, bytenr, 0);
 597
 598         return cache;
 599 }
 600
 601 /*
 602  * return the block group that contains the given bytenr
 603  */
 604 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 605                                                  struct btrfs_fs_info *info,
 606                                                  u64 bytenr)
 607 {
 608         struct btrfs_block_group_cache *cache;
 609
 610         cache = block_group_cache_tree_search(info, bytenr, 1);
 611
 612         return cache;
 613 }
 614
 615 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 616                                                   u64 flags)
 617 {
 618         struct list_head *head = &info->space_info;
 619         struct btrfs_space_info *found;
 620
 621         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 622                  BTRFS_BLOCK_GROUP_METADATA;
 623
 624         rcu_read_lock();
 625         list_for_each_entry_rcu(found, head, list) {
 626                 if (found->flags & flags) {
 627                         rcu_read_unlock();
 628                         return found;
 629                 }
 630         }
 631         rcu_read_unlock();
 632         return NULL;
 633 }
 634
 635 /*
 636  * after adding space to the filesystem, we need to clear the full flags
 637  * on all the space infos.
 638  */
 639 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 640 {
 641         struct list_head *head = &info->space_info;
 642         struct btrfs_space_info *found;
 643
 644         rcu_read_lock();
 645         list_for_each_entry_rcu(found, head, list)
 646                 found->full = 0;
 647         rcu_read_unlock();
 648 }
 649
 650 static u64 div_factor(u64 num, int factor)
 651 {
 652         if (factor == 10)
 653                 return num;
 654         num *= factor;
 655         do_div(num, 10);
 656         return num;
 657 }
 658
 659 static u64 div_factor_fine(u64 num, int factor)
 660 {
 661         if (factor == 100)
 662                 return num;
 663         num *= factor;
 664         do_div(num, 100);
 665         return num;
 666 }
 667
 668 u64 btrfs_find_block_group(struct btrfs_root *root,
 669                            u64 search_start, u64 search_hint, int owner)
 670 {
 671         struct btrfs_block_group_cache *cache;
 672         u64 used;
 673         u64 last = max(search_hint, search_start);
 674         u64 group_start = 0;
 675         int full_search = 0;
 676         int factor = 9;
 677         int wrapped = 0;
 678 again:
 679         while (1) {
 680                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 681                 if (!cache)
 682                         break;
 683
 684                 spin_lock(&cache->lock);
 685                 last = cache->key.objectid + cache->key.offset;
 686                 used = btrfs_block_group_used(&cache->item);
 687
 688                 if ((full_search || !cache->ro) &&
 689                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 690                         if (used + cache->pinned + cache->reserved <
 691                             div_factor(cache->key.offset, factor)) {
 692                                 group_start = cache->key.objectid;
 693                                 spin_unlock(&cache->lock);
 694                                 btrfs_put_block_group(cache);
 695                                 goto found;
 696                         }
 697                 }
 698                 spin_unlock(&cache->lock);
 699                 btrfs_put_block_group(cache);
 700                 cond_resched();
 701         }
 702         if (!wrapped) {
 703                 last = search_start;
 704                 wrapped = 1;
 705                 goto again;
 706         }
 707         if (!full_search && factor < 10) {
 708                 last = search_start;
 709                 full_search = 1;
 710                 factor = 10;
 711                 goto again;
 712         }
 713 found:
 714         return group_start;
 715 }
 716
 717 /* simple helper to search for an existing extent at a given offset */
 718 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 719 {
 720         int ret;
 721         struct btrfs_key key;
 722         struct btrfs_path *path;
 723
 724         path = btrfs_alloc_path();
 725         if (!path)
 726                 return -ENOMEM;
 727
 728         key.objectid = start;
 729         key.offset = len;
 730         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 731         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 732                                 0, 0);
 733         btrfs_free_path(path);
 734         return ret;
 735 }
 736
 737 /*
 738  * helper function to lookup reference count and flags of extent.
 739  *
 740  * the head node for delayed ref is used to store the sum of all the
 741  * reference count modifications queued up in the rbtree. the head
 742  * node may also store the extent flags to set. This way you can check
 743  * to see what the reference count and extent flags would be if all of
 744  * the delayed refs are not processed.
 745  */
 746 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 747                              struct btrfs_root *root, u64 bytenr,
 748                              u64 num_bytes, u64 *refs, u64 *flags)
 749 {
 750         struct btrfs_delayed_ref_head *head;
 751         struct btrfs_delayed_ref_root *delayed_refs;
 752         struct btrfs_path *path;
 753         struct btrfs_extent_item *ei;
 754         struct extent_buffer *leaf;
 755         struct btrfs_key key;
 756         u32 item_size;
 757         u64 num_refs;
 758         u64 extent_flags;
 759         int ret;
 760
 761         path = btrfs_alloc_path();
 762         if (!path)
 763                 return -ENOMEM;
 764
 765         key.objectid = bytenr;
 766         key.type = BTRFS_EXTENT_ITEM_KEY;
 767         key.offset = num_bytes;
 768         if (!trans) {
 769                 path->skip_locking = 1;
 770                 path->search_commit_root = 1;
 771         }
 772 again:
 773         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 774                                 &key, path, 0, 0);
 775         if (ret < 0)
 776                 goto out_free;
 777
 778         if (ret == 0) {
 779                 leaf = path->nodes[0];
 780                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 781                 if (item_size >= sizeof(*ei)) {
 782                         ei = btrfs_item_ptr(leaf, path->slots[0],
 783                                             struct btrfs_extent_item);
 784                         num_refs = btrfs_extent_refs(leaf, ei);
 785                         extent_flags = btrfs_extent_flags(leaf, ei);
 786                 } else {
 787 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 788                         struct btrfs_extent_item_v0 *ei0;
 789                         BUG_ON(item_size != sizeof(*ei0));
 790                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 791                                              struct btrfs_extent_item_v0);
 792                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 793                         /* FIXME: this isn't correct for data */
 794                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 795 #else
 796                         BUG();
 797 #endif
 798                 }
 799                 BUG_ON(num_refs == 0);
 800         } else {
 801                 num_refs = 0;
 802                 extent_flags = 0;
 803                 ret = 0;
 804         }
 805
 806         if (!trans)
 807                 goto out;
 808
 809         delayed_refs = &trans->transaction->delayed_refs;
 810         spin_lock(&delayed_refs->lock);
 811         head = btrfs_find_delayed_ref_head(trans, bytenr);
 812         if (head) {
 813                 if (!mutex_trylock(&head->mutex)) {
 814                         atomic_inc(&head->node.refs);
 815                         spin_unlock(&delayed_refs->lock);
 816
 817                         btrfs_release_path(path);
 818
 819                         /*
 820                          * Mutex was contended, block until it's released and try
 821                          * again
 822                          */
 823                         mutex_lock(&head->mutex);
 824                         mutex_unlock(&head->mutex);
 825                         btrfs_put_delayed_ref(&head->node);
 826                         goto again;
 827                 }
 828                 if (head->extent_op && head->extent_op->update_flags)
 829                         extent_flags |= head->extent_op->flags_to_set;
 830                 else
 831                         BUG_ON(num_refs == 0);
 832
 833                 num_refs += head->node.ref_mod;
 834                 mutex_unlock(&head->mutex);
 835         }
 836         spin_unlock(&delayed_refs->lock);
 837 out:
 838         WARN_ON(num_refs == 0);
 839         if (refs)
 840                 *refs = num_refs;
 841         if (flags)
 842                 *flags = extent_flags;
 843 out_free:
 844         btrfs_free_path(path);
 845         return ret;
 846 }
 847
 848 /*
 849  * Back reference rules.  Back refs have three main goals:
 850  *
 851  * 1) differentiate between all holders of references to an extent so that
 852  *    when a reference is dropped we can make sure it was a valid reference
 853  *    before freeing the extent.
 854  *
 855  * 2) Provide enough information to quickly find the holders of an extent
 856  *    if we notice a given block is corrupted or bad.
 857  *
 858  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 859  *    maintenance.  This is actually the same as #2, but with a slightly
 860  *    different use case.
 861  *
 862  * There are two kinds of back refs. The implicit back refs is optimized
 863  * for pointers in non-shared tree blocks. For a given pointer in a block,
 864  * back refs of this kind provide information about the block's owner tree
 865  * and the pointer's key. These information allow us to find the block by
 866  * b-tree searching. The full back refs is for pointers in tree blocks not
 867  * referenced by their owner trees. The location of tree block is recorded
 868  * in the back refs. Actually the full back refs is generic, and can be
 869  * used in all cases the implicit back refs is used. The major shortcoming
 870  * of the full back refs is its overhead. Every time a tree block gets
 871  * COWed, we have to update back refs entry for all pointers in it.
 872  *
 873  * For a newly allocated tree block, we use implicit back refs for
 874  * pointers in it. This means most tree related operations only involve
 875  * implicit back refs. For a tree block created in old transaction, the
 876  * only way to drop a reference to it is COW it. So we can detect the
 877  * event that tree block loses its owner tree's reference and do the
 878  * back refs conversion.
 879  *
 880  * When a tree block is COW'd through a tree, there are four cases:
 881  *
 882  * The reference count of the block is one and the tree is the block's
 883  * owner tree. Nothing to do in this case.
 884  *
 885  * The reference count of the block is one and the tree is not the
 886  * block's owner tree. In this case, full back refs is used for pointers
 887  * in the block. Remove these full back refs, add implicit back refs for
 888  * every pointers in the new block.
 889  *
 890  * The reference count of the block is greater than one and the tree is
 891  * the block's owner tree. In this case, implicit back refs is used for
 892  * pointers in the block. Add full back refs for every pointers in the
 893  * block, increase lower level extents' reference counts. The original
 894  * implicit back refs are entailed to the new block.
 895  *
 896  * The reference count of the block is greater than one and the tree is
 897  * not the block's owner tree. Add implicit back refs for every pointer in
 898  * the new block, increase lower level extents' reference count.
 899  *
 900  * Back Reference Key composing:
 901  *
 902  * The key objectid corresponds to the first byte in the extent,
 903  * The key type is used to differentiate between types of back refs.
 904  * There are different meanings of the key offset for different types
 905  * of back refs.
 906  *
 907  * File extents can be referenced by:
 908  *
 909  * - multiple snapshots, subvolumes, or different generations in one subvol
 910  * - different files inside a single subvolume
 911  * - different offsets inside a file (bookend extents in file.c)
 912  *
 913  * The extent ref structure for the implicit back refs has fields for:
 914  *
 915  * - Objectid of the subvolume root
 916  * - objectid of the file holding the reference
 917  * - original offset in the file
 918  * - how many bookend extents
 919  *
 920  * The key offset for the implicit back refs is hash of the first
 921  * three fields.
 922  *
 923  * The extent ref structure for the full back refs has field for:
 924  *
 925  * - number of pointers in the tree leaf
 926  *
 927  * The key offset for the implicit back refs is the first byte of
 928  * the tree leaf
 929  *
 930  * When a file extent is allocated, The implicit back refs is used.
 931  * the fields are filled in:
 932  *
 933  *     (root_key.objectid, inode objectid, offset in file, 1)
 934  *
 935  * When a file extent is removed file truncation, we find the
 936  * corresponding implicit back refs and check the following fields:
 937  *
 938  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 939  *
 940  * Btree extents can be referenced by:
 941  *
 942  * - Different subvolumes
 943  *
 944  * Both the implicit back refs and the full back refs for tree blocks
 945  * only consist of key. The key offset for the implicit back refs is
 946  * objectid of block's owner tree. The key offset for the full back refs
 947  * is the first byte of parent block.
 948  *
 949  * When implicit back refs is used, information about the lowest key and
 950  * level of the tree block are required. These information are stored in
 951  * tree block info structure.
 952  */
 953
 954 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 955 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 956                                   struct btrfs_root *root,
 957                                   struct btrfs_path *path,
 958                                   u64 owner, u32 extra_size)
 959 {
 960         struct btrfs_extent_item *item;
 961         struct btrfs_extent_item_v0 *ei0;
 962         struct btrfs_extent_ref_v0 *ref0;
 963         struct btrfs_tree_block_info *bi;
 964         struct extent_buffer *leaf;
 965         struct btrfs_key key;
 966         struct btrfs_key found_key;
 967         u32 new_size = sizeof(*item);
 968         u64 refs;
 969         int ret;
 970
 971         leaf = path->nodes[0];
 972         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 973
 974         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 975         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 976                              struct btrfs_extent_item_v0);
 977         refs = btrfs_extent_refs_v0(leaf, ei0);
 978
 979         if (owner == (u64)-1) {
 980                 while (1) {
 981                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 982                                 ret = btrfs_next_leaf(root, path);
 983                                 if (ret < 0)
 984                                         return ret;
 985                                 BUG_ON(ret > 0);
 986                                 leaf = path->nodes[0];
 987                         }
 988                         btrfs_item_key_to_cpu(leaf, &found_key,
 989                                               path->slots[0]);
 990                         BUG_ON(key.objectid != found_key.objectid);
 991                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 992                                 path->slots[0]++;
 993                                 continue;
 994                         }
 995                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 996                                               struct btrfs_extent_ref_v0);
 997                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 998                         break;
 999                 }
1000         }
1001         btrfs_release_path(path);
1002
1003         if (owner < BTRFS_FIRST_FREE_OBJECTID)
1004                 new_size += sizeof(*bi);
1005
1006         new_size -= sizeof(*ei0);
1007         ret = btrfs_search_slot(trans, root, &key, path,
1008                                 new_size + extra_size, 1);
1009         if (ret < 0)
1010                 return ret;
1011         BUG_ON(ret);
1012
1013         ret = btrfs_extend_item(trans, root, path, new_size);
1014
1015         leaf = path->nodes[0];
1016         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1017         btrfs_set_extent_refs(leaf, item, refs);
1018         /* FIXME: get real generation */
1019         btrfs_set_extent_generation(leaf, item, 0);
1020         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1021                 btrfs_set_extent_flags(leaf, item,
1022                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
1023                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
1024                 bi = (struct btrfs_tree_block_info *)(item + 1);
1025                 /* FIXME: get first key of the block */
1026                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1027                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1028         } else {
1029                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1030         }
1031         btrfs_mark_buffer_dirty(leaf);
1032         return 0;
1033 }
1034 #endif
1035
1036 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1037 {
1038         u32 high_crc = ~(u32)0;
1039         u32 low_crc = ~(u32)0;
1040         __le64 lenum;
1041
1042         lenum = cpu_to_le64(root_objectid);
1043         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1044         lenum = cpu_to_le64(owner);
1045         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1046         lenum = cpu_to_le64(offset);
1047         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1048
1049         return ((u64)high_crc << 31) ^ (u64)low_crc;
1050 }
1051
1052 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1053                                      struct btrfs_extent_data_ref *ref)
1054 {
1055         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1056                                     btrfs_extent_data_ref_objectid(leaf, ref),
1057                                     btrfs_extent_data_ref_offset(leaf, ref));
1058 }
1059
1060 static int match_extent_data_ref(struct extent_buffer *leaf,
1061                                  struct btrfs_extent_data_ref *ref,
1062                                  u64 root_objectid, u64 owner, u64 offset)
1063 {
1064         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1065             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1066             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1067                 return 0;
1068         return 1;
1069 }
1070
1071 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1072                                            struct btrfs_root *root,
1073                                            struct btrfs_path *path,
1074                                            u64 bytenr, u64 parent,
1075                                            u64 root_objectid,
1076                                            u64 owner, u64 offset)
1077 {
1078         struct btrfs_key key;
1079         struct btrfs_extent_data_ref *ref;
1080         struct extent_buffer *leaf;
1081         u32 nritems;
1082         int ret;
1083         int recow;
1084         int err = -ENOENT;
1085
1086         key.objectid = bytenr;
1087         if (parent) {
1088                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1089                 key.offset = parent;
1090         } else {
1091                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1092                 key.offset = hash_extent_data_ref(root_objectid,
1093                                                   owner, offset);
1094         }
1095 again:
1096         recow = 0;
1097         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1098         if (ret < 0) {
1099                 err = ret;
1100                 goto fail;
1101         }
1102
1103         if (parent) {
1104                 if (!ret)
1105                         return 0;
1106 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1107                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1108                 btrfs_release_path(path);
1109                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1110                 if (ret < 0) {
1111                         err = ret;
1112                         goto fail;
1113                 }
1114                 if (!ret)
1115                         return 0;
1116 #endif
1117                 goto fail;
1118         }
1119
1120         leaf = path->nodes[0];
1121         nritems = btrfs_header_nritems(leaf);
1122         while (1) {
1123                 if (path->slots[0] >= nritems) {
1124                         ret = btrfs_next_leaf(root, path);
1125                         if (ret < 0)
1126                                 err = ret;
1127                         if (ret)
1128                                 goto fail;
1129
1130                         leaf = path->nodes[0];
1131                         nritems = btrfs_header_nritems(leaf);
1132                         recow = 1;
1133                 }
1134
1135                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1136                 if (key.objectid != bytenr ||
1137                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1138                         goto fail;
1139
1140                 ref = btrfs_item_ptr(leaf, path->slots[0],
1141                                      struct btrfs_extent_data_ref);
1142
1143                 if (match_extent_data_ref(leaf, ref, root_objectid,
1144                                           owner, offset)) {
1145                         if (recow) {
1146                                 btrfs_release_path(path);
1147                                 goto again;
1148                         }
1149                         err = 0;
1150                         break;
1151                 }
1152                 path->slots[0]++;
1153         }
1154 fail:
1155         return err;
1156 }
1157
1158 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1159                                            struct btrfs_root *root,
1160                                            struct btrfs_path *path,
1161                                            u64 bytenr, u64 parent,
1162                                            u64 root_objectid, u64 owner,
1163                                            u64 offset, int refs_to_add)
1164 {
1165         struct btrfs_key key;
1166         struct extent_buffer *leaf;
1167         u32 size;
1168         u32 num_refs;
1169         int ret;
1170
1171         key.objectid = bytenr;
1172         if (parent) {
1173                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1174                 key.offset = parent;
1175                 size = sizeof(struct btrfs_shared_data_ref);
1176         } else {
1177                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1178                 key.offset = hash_extent_data_ref(root_objectid,
1179                                                   owner, offset);
1180                 size = sizeof(struct btrfs_extent_data_ref);
1181         }
1182
1183         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1184         if (ret && ret != -EEXIST)
1185                 goto fail;
1186
1187         leaf = path->nodes[0];
1188         if (parent) {
1189                 struct btrfs_shared_data_ref *ref;
1190                 ref = btrfs_item_ptr(leaf, path->slots[0],
1191                                      struct btrfs_shared_data_ref);
1192                 if (ret == 0) {
1193                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1194                 } else {
1195                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1196                         num_refs += refs_to_add;
1197                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1198                 }
1199         } else {
1200                 struct btrfs_extent_data_ref *ref;
1201                 while (ret == -EEXIST) {
1202                         ref = btrfs_item_ptr(leaf, path->slots[0],
1203                                              struct btrfs_extent_data_ref);
1204                         if (match_extent_data_ref(leaf, ref, root_objectid,
1205                                                   owner, offset))
1206                                 break;
1207                         btrfs_release_path(path);
1208                         key.offset++;
1209                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1210                                                       size);
1211                         if (ret && ret != -EEXIST)
1212                                 goto fail;
1213
1214                         leaf = path->nodes[0];
1215                 }
1216                 ref = btrfs_item_ptr(leaf, path->slots[0],
1217                                      struct btrfs_extent_data_ref);
1218                 if (ret == 0) {
1219                         btrfs_set_extent_data_ref_root(leaf, ref,
1220                                                        root_objectid);
1221                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1222                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1223                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1224                 } else {
1225                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1226                         num_refs += refs_to_add;
1227                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1228                 }
1229         }
1230         btrfs_mark_buffer_dirty(leaf);
1231         ret = 0;
1232 fail:
1233         btrfs_release_path(path);
1234         return ret;
1235 }
1236
1237 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1238                                            struct btrfs_root *root,
1239                                            struct btrfs_path *path,
1240                                            int refs_to_drop)
1241 {
1242         struct btrfs_key key;
1243         struct btrfs_extent_data_ref *ref1 = NULL;
1244         struct btrfs_shared_data_ref *ref2 = NULL;
1245         struct extent_buffer *leaf;
1246         u32 num_refs = 0;
1247         int ret = 0;
1248
1249         leaf = path->nodes[0];
1250         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1251
1252         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                 BUG();
1269         }
1270
1271         BUG_ON(num_refs < refs_to_drop);
1272         num_refs -= refs_to_drop;
1273
1274         if (num_refs == 0) {
1275                 ret = btrfs_del_item(trans, root, path);
1276         } else {
1277                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1278                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1279                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1280                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1281 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1282                 else {
1283                         struct btrfs_extent_ref_v0 *ref0;
1284                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1285                                         struct btrfs_extent_ref_v0);
1286                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1287                 }
1288 #endif
1289                 btrfs_mark_buffer_dirty(leaf);
1290         }
1291         return ret;
1292 }
1293
1294 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1295                                           struct btrfs_path *path,
1296                                           struct btrfs_extent_inline_ref *iref)
1297 {
1298         struct btrfs_key key;
1299         struct extent_buffer *leaf;
1300         struct btrfs_extent_data_ref *ref1;
1301         struct btrfs_shared_data_ref *ref2;
1302         u32 num_refs = 0;
1303
1304         leaf = path->nodes[0];
1305         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1306         if (iref) {
1307                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1308                     BTRFS_EXTENT_DATA_REF_KEY) {
1309                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1310                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1311                 } else {
1312                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1313                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1314                 }
1315         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1316                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1317                                       struct btrfs_extent_data_ref);
1318                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1319         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1320                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1321                                       struct btrfs_shared_data_ref);
1322                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1323 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1324         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1325                 struct btrfs_extent_ref_v0 *ref0;
1326                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1327                                       struct btrfs_extent_ref_v0);
1328                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1329 #endif
1330         } else {
1331                 WARN_ON(1);
1332         }
1333         return num_refs;
1334 }
1335
1336 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1337                                           struct btrfs_root *root,
1338                                           struct btrfs_path *path,
1339                                           u64 bytenr, u64 parent,
1340                                           u64 root_objectid)
1341 {
1342         struct btrfs_key key;
1343         int ret;
1344
1345         key.objectid = bytenr;
1346         if (parent) {
1347                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1348                 key.offset = parent;
1349         } else {
1350                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1351                 key.offset = root_objectid;
1352         }
1353
1354         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1355         if (ret > 0)
1356                 ret = -ENOENT;
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358         if (ret == -ENOENT && parent) {
1359                 btrfs_release_path(path);
1360                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1361                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1362                 if (ret > 0)
1363                         ret = -ENOENT;
1364         }
1365 #endif
1366         return ret;
1367 }
1368
1369 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1370                                           struct btrfs_root *root,
1371                                           struct btrfs_path *path,
1372                                           u64 bytenr, u64 parent,
1373                                           u64 root_objectid)
1374 {
1375         struct btrfs_key key;
1376         int ret;
1377
1378         key.objectid = bytenr;
1379         if (parent) {
1380                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1381                 key.offset = parent;
1382         } else {
1383                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1384                 key.offset = root_objectid;
1385         }
1386
1387         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1388         btrfs_release_path(path);
1389         return ret;
1390 }
1391
1392 static inline int extent_ref_type(u64 parent, u64 owner)
1393 {
1394         int type;
1395         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1396                 if (parent > 0)
1397                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1398                 else
1399                         type = BTRFS_TREE_BLOCK_REF_KEY;
1400         } else {
1401                 if (parent > 0)
1402                         type = BTRFS_SHARED_DATA_REF_KEY;
1403                 else
1404                         type = BTRFS_EXTENT_DATA_REF_KEY;
1405         }
1406         return type;
1407 }
1408
1409 static int find_next_key(struct btrfs_path *path, int level,
1410                          struct btrfs_key *key)
1411
1412 {
1413         for (; level < BTRFS_MAX_LEVEL; level++) {
1414                 if (!path->nodes[level])
1415                         break;
1416                 if (path->slots[level] + 1 >=
1417                     btrfs_header_nritems(path->nodes[level]))
1418                         continue;
1419                 if (level == 0)
1420                         btrfs_item_key_to_cpu(path->nodes[level], key,
1421                                               path->slots[level] + 1);
1422                 else
1423                         btrfs_node_key_to_cpu(path->nodes[level], key,
1424                                               path->slots[level] + 1);
1425                 return 0;
1426         }
1427         return 1;
1428 }
1429
1430 /*
1431  * look for inline back ref. if back ref is found, *ref_ret is set
1432  * to the address of inline back ref, and 0 is returned.
1433  *
1434  * if back ref isn't found, *ref_ret is set to the address where it
1435  * should be inserted, and -ENOENT is returned.
1436  *
1437  * if insert is true and there are too many inline back refs, the path
1438  * points to the extent item, and -EAGAIN is returned.
1439  *
1440  * NOTE: inline back refs are ordered in the same way that back ref
1441  *       items in the tree are ordered.
1442  */
1443 static noinline_for_stack
1444 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1445                                  struct btrfs_root *root,
1446                                  struct btrfs_path *path,
1447                                  struct btrfs_extent_inline_ref **ref_ret,
1448                                  u64 bytenr, u64 num_bytes,
1449                                  u64 parent, u64 root_objectid,
1450                                  u64 owner, u64 offset, int insert)
1451 {
1452         struct btrfs_key key;
1453         struct extent_buffer *leaf;
1454         struct btrfs_extent_item *ei;
1455         struct btrfs_extent_inline_ref *iref;
1456         u64 flags;
1457         u64 item_size;
1458         unsigned long ptr;
1459         unsigned long end;
1460         int extra_size;
1461         int type;
1462         int want;
1463         int ret;
1464         int err = 0;
1465
1466         key.objectid = bytenr;
1467         key.type = BTRFS_EXTENT_ITEM_KEY;
1468         key.offset = num_bytes;
1469
1470         want = extent_ref_type(parent, owner);
1471         if (insert) {
1472                 extra_size = btrfs_extent_inline_ref_size(want);
1473                 path->keep_locks = 1;
1474         } else
1475                 extra_size = -1;
1476         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1477         if (ret < 0) {
1478                 err = ret;
1479                 goto out;
1480         }
1481         BUG_ON(ret);
1482
1483         leaf = path->nodes[0];
1484         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1485 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1486         if (item_size < sizeof(*ei)) {
1487                 if (!insert) {
1488                         err = -ENOENT;
1489                         goto out;
1490                 }
1491                 ret = convert_extent_item_v0(trans, root, path, owner,
1492                                              extra_size);
1493                 if (ret < 0) {
1494                         err = ret;
1495                         goto out;
1496                 }
1497                 leaf = path->nodes[0];
1498                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1499         }
1500 #endif
1501         BUG_ON(item_size < sizeof(*ei));
1502
1503         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1504         flags = btrfs_extent_flags(leaf, ei);
1505
1506         ptr = (unsigned long)(ei + 1);
1507         end = (unsigned long)ei + item_size;
1508
1509         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1510                 ptr += sizeof(struct btrfs_tree_block_info);
1511                 BUG_ON(ptr > end);
1512         } else {
1513                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1514         }
1515
1516         err = -ENOENT;
1517         while (1) {
1518                 if (ptr >= end) {
1519                         WARN_ON(ptr > end);
1520                         break;
1521                 }
1522                 iref = (struct btrfs_extent_inline_ref *)ptr;
1523                 type = btrfs_extent_inline_ref_type(leaf, iref);
1524                 if (want < type)
1525                         break;
1526                 if (want > type) {
1527                         ptr += btrfs_extent_inline_ref_size(type);
1528                         continue;
1529                 }
1530
1531                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1532                         struct btrfs_extent_data_ref *dref;
1533                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1534                         if (match_extent_data_ref(leaf, dref, root_objectid,
1535                                                   owner, offset)) {
1536                                 err = 0;
1537                                 break;
1538                         }
1539                         if (hash_extent_data_ref_item(leaf, dref) <
1540                             hash_extent_data_ref(root_objectid, owner, offset))
1541                                 break;
1542                 } else {
1543                         u64 ref_offset;
1544                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1545                         if (parent > 0) {
1546                                 if (parent == ref_offset) {
1547                                         err = 0;
1548                                         break;
1549                                 }
1550                                 if (ref_offset < parent)
1551                                         break;
1552                         } else {
1553                                 if (root_objectid == ref_offset) {
1554                                         err = 0;
1555                                         break;
1556                                 }
1557                                 if (ref_offset < root_objectid)
1558                                         break;
1559                         }
1560                 }
1561                 ptr += btrfs_extent_inline_ref_size(type);
1562         }
1563         if (err == -ENOENT && insert) {
1564                 if (item_size + extra_size >=
1565                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1566                         err = -EAGAIN;
1567                         goto out;
1568                 }
1569                 /*
1570                  * To add new inline back ref, we have to make sure
1571                  * there is no corresponding back ref item.
1572                  * For simplicity, we just do not add new inline back
1573                  * ref if there is any kind of item for this block
1574                  */
1575                 if (find_next_key(path, 0, &key) == 0 &&
1576                     key.objectid == bytenr &&
1577                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1578                         err = -EAGAIN;
1579                         goto out;
1580                 }
1581         }
1582         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1583 out:
1584         if (insert) {
1585                 path->keep_locks = 0;
1586                 btrfs_unlock_up_safe(path, 1);
1587         }
1588         return err;
1589 }
1590
1591 /*
1592  * helper to add new inline back ref
1593  */
1594 static noinline_for_stack
1595 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1596                                 struct btrfs_root *root,
1597                                 struct btrfs_path *path,
1598                                 struct btrfs_extent_inline_ref *iref,
1599                                 u64 parent, u64 root_objectid,
1600                                 u64 owner, u64 offset, int refs_to_add,
1601                                 struct btrfs_delayed_extent_op *extent_op)
1602 {
1603         struct extent_buffer *leaf;
1604         struct btrfs_extent_item *ei;
1605         unsigned long ptr;
1606         unsigned long end;
1607         unsigned long item_offset;
1608         u64 refs;
1609         int size;
1610         int type;
1611         int ret;
1612
1613         leaf = path->nodes[0];
1614         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1615         item_offset = (unsigned long)iref - (unsigned long)ei;
1616
1617         type = extent_ref_type(parent, owner);
1618         size = btrfs_extent_inline_ref_size(type);
1619
1620         ret = btrfs_extend_item(trans, root, path, size);
1621
1622         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1623         refs = btrfs_extent_refs(leaf, ei);
1624         refs += refs_to_add;
1625         btrfs_set_extent_refs(leaf, ei, refs);
1626         if (extent_op)
1627                 __run_delayed_extent_op(extent_op, leaf, ei);
1628
1629         ptr = (unsigned long)ei + item_offset;
1630         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1631         if (ptr < end - size)
1632                 memmove_extent_buffer(leaf, ptr + size, ptr,
1633                                       end - size - ptr);
1634
1635         iref = (struct btrfs_extent_inline_ref *)ptr;
1636         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1637         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1638                 struct btrfs_extent_data_ref *dref;
1639                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1640                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1641                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1642                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1643                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1644         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1645                 struct btrfs_shared_data_ref *sref;
1646                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1647                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1648                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1649         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1650                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1651         } else {
1652                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1653         }
1654         btrfs_mark_buffer_dirty(leaf);
1655         return 0;
1656 }
1657
1658 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1659                                  struct btrfs_root *root,
1660                                  struct btrfs_path *path,
1661                                  struct btrfs_extent_inline_ref **ref_ret,
1662                                  u64 bytenr, u64 num_bytes, u64 parent,
1663                                  u64 root_objectid, u64 owner, u64 offset)
1664 {
1665         int ret;
1666
1667         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1668                                            bytenr, num_bytes, parent,
1669                                            root_objectid, owner, offset, 0);
1670         if (ret != -ENOENT)
1671                 return ret;
1672
1673         btrfs_release_path(path);
1674         *ref_ret = NULL;
1675
1676         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1677                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1678                                             root_objectid);
1679         } else {
1680                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1681                                              root_objectid, owner, offset);
1682         }
1683         return ret;
1684 }
1685
1686 /*
1687  * helper to update/remove inline back ref
1688  */
1689 static noinline_for_stack
1690 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1691                                  struct btrfs_root *root,
1692                                  struct btrfs_path *path,
1693                                  struct btrfs_extent_inline_ref *iref,
1694                                  int refs_to_mod,
1695                                  struct btrfs_delayed_extent_op *extent_op)
1696 {
1697         struct extent_buffer *leaf;
1698         struct btrfs_extent_item *ei;
1699         struct btrfs_extent_data_ref *dref = NULL;
1700         struct btrfs_shared_data_ref *sref = NULL;
1701         unsigned long ptr;
1702         unsigned long end;
1703         u32 item_size;
1704         int size;
1705         int type;
1706         int ret;
1707         u64 refs;
1708
1709         leaf = path->nodes[0];
1710         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1711         refs = btrfs_extent_refs(leaf, ei);
1712         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1713         refs += refs_to_mod;
1714         btrfs_set_extent_refs(leaf, ei, refs);
1715         if (extent_op)
1716                 __run_delayed_extent_op(extent_op, leaf, ei);
1717
1718         type = btrfs_extent_inline_ref_type(leaf, iref);
1719
1720         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1721                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1722                 refs = btrfs_extent_data_ref_count(leaf, dref);
1723         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1724                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1725                 refs = btrfs_shared_data_ref_count(leaf, sref);
1726         } else {
1727                 refs = 1;
1728                 BUG_ON(refs_to_mod != -1);
1729         }
1730
1731         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1732         refs += refs_to_mod;
1733
1734         if (refs > 0) {
1735                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1736                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1737                 else
1738                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1739         } else {
1740                 size =  btrfs_extent_inline_ref_size(type);
1741                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1742                 ptr = (unsigned long)iref;
1743                 end = (unsigned long)ei + item_size;
1744                 if (ptr + size < end)
1745                         memmove_extent_buffer(leaf, ptr, ptr + size,
1746                                               end - ptr - size);
1747                 item_size -= size;
1748                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1749         }
1750         btrfs_mark_buffer_dirty(leaf);
1751         return 0;
1752 }
1753
1754 static noinline_for_stack
1755 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1756                                  struct btrfs_root *root,
1757                                  struct btrfs_path *path,
1758                                  u64 bytenr, u64 num_bytes, u64 parent,
1759                                  u64 root_objectid, u64 owner,
1760                                  u64 offset, int refs_to_add,
1761                                  struct btrfs_delayed_extent_op *extent_op)
1762 {
1763         struct btrfs_extent_inline_ref *iref;
1764         int ret;
1765
1766         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1767                                            bytenr, num_bytes, parent,
1768                                            root_objectid, owner, offset, 1);
1769         if (ret == 0) {
1770                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1771                 ret = update_inline_extent_backref(trans, root, path, iref,
1772                                                    refs_to_add, extent_op);
1773         } else if (ret == -ENOENT) {
1774                 ret = setup_inline_extent_backref(trans, root, path, iref,
1775                                                   parent, root_objectid,
1776                                                   owner, offset, refs_to_add,
1777                                                   extent_op);
1778         }
1779         return ret;
1780 }
1781
1782 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1783                                  struct btrfs_root *root,
1784                                  struct btrfs_path *path,
1785                                  u64 bytenr, u64 parent, u64 root_objectid,
1786                                  u64 owner, u64 offset, int refs_to_add)
1787 {
1788         int ret;
1789         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1790                 BUG_ON(refs_to_add != 1);
1791                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1792                                             parent, root_objectid);
1793         } else {
1794                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1795                                              parent, root_objectid,
1796                                              owner, offset, refs_to_add);
1797         }
1798         return ret;
1799 }
1800
1801 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1802                                  struct btrfs_root *root,
1803                                  struct btrfs_path *path,
1804                                  struct btrfs_extent_inline_ref *iref,
1805                                  int refs_to_drop, int is_data)
1806 {
1807         int ret;
1808
1809         BUG_ON(!is_data && refs_to_drop != 1);
1810         if (iref) {
1811                 ret = update_inline_extent_backref(trans, root, path, iref,
1812                                                    -refs_to_drop, NULL);
1813         } else if (is_data) {
1814                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1815         } else {
1816                 ret = btrfs_del_item(trans, root, path);
1817         }
1818         return ret;
1819 }
1820
1821 static int btrfs_issue_discard(struct block_device *bdev,
1822                                 u64 start, u64 len)
1823 {
1824         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1825 }
1826
1827 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1828                                 u64 num_bytes, u64 *actual_bytes)
1829 {
1830         int ret;
1831         u64 discarded_bytes = 0;
1832         struct btrfs_bio *bbio = NULL;
1833
1834
1835         /* Tell the block device(s) that the sectors can be discarded */
1836         ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1837                               bytenr, &num_bytes, &bbio, 0);
1838         if (!ret) {
1839                 struct btrfs_bio_stripe *stripe = bbio->stripes;
1840                 int i;
1841
1842
1843                 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1844                         if (!stripe->dev->can_discard)
1845                                 continue;
1846
1847                         ret = btrfs_issue_discard(stripe->dev->bdev,
1848                                                   stripe->physical,
1849                                                   stripe->length);
1850                         if (!ret)
1851                                 discarded_bytes += stripe->length;
1852                         else if (ret != -EOPNOTSUPP)
1853                                 break;
1854
1855                         /*
1856                          * Just in case we get back EOPNOTSUPP for some reason,
1857                          * just ignore the return value so we don't screw up
1858                          * people calling discard_extent.
1859                          */
1860                         ret = 0;
1861                 }
1862                 kfree(bbio);
1863         }
1864
1865         if (actual_bytes)
1866                 *actual_bytes = discarded_bytes;
1867
1868
1869         return ret;
1870 }
1871
1872 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1873                          struct btrfs_root *root,
1874                          u64 bytenr, u64 num_bytes, u64 parent,
1875                          u64 root_objectid, u64 owner, u64 offset)
1876 {
1877         int ret;
1878         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1879                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1880
1881         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1882                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1883                                         parent, root_objectid, (int)owner,
1884                                         BTRFS_ADD_DELAYED_REF, NULL);
1885         } else {
1886                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1887                                         parent, root_objectid, owner, offset,
1888                                         BTRFS_ADD_DELAYED_REF, NULL);
1889         }
1890         return ret;
1891 }
1892
1893 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1894                                   struct btrfs_root *root,
1895                                   u64 bytenr, u64 num_bytes,
1896                                   u64 parent, u64 root_objectid,
1897                                   u64 owner, u64 offset, int refs_to_add,
1898                                   struct btrfs_delayed_extent_op *extent_op)
1899 {
1900         struct btrfs_path *path;
1901         struct extent_buffer *leaf;
1902         struct btrfs_extent_item *item;
1903         u64 refs;
1904         int ret;
1905         int err = 0;
1906
1907         path = btrfs_alloc_path();
1908         if (!path)
1909                 return -ENOMEM;
1910
1911         path->reada = 1;
1912         path->leave_spinning = 1;
1913         /* this will setup the path even if it fails to insert the back ref */
1914         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1915                                            path, bytenr, num_bytes, parent,
1916                                            root_objectid, owner, offset,
1917                                            refs_to_add, extent_op);
1918         if (ret == 0)
1919                 goto out;
1920
1921         if (ret != -EAGAIN) {
1922                 err = ret;
1923                 goto out;
1924         }
1925
1926         leaf = path->nodes[0];
1927         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1928         refs = btrfs_extent_refs(leaf, item);
1929         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1930         if (extent_op)
1931                 __run_delayed_extent_op(extent_op, leaf, item);
1932
1933         btrfs_mark_buffer_dirty(leaf);
1934         btrfs_release_path(path);
1935
1936         path->reada = 1;
1937         path->leave_spinning = 1;
1938
1939         /* now insert the actual backref */
1940         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1941                                     path, bytenr, parent, root_objectid,
1942                                     owner, offset, refs_to_add);
1943         BUG_ON(ret);
1944 out:
1945         btrfs_free_path(path);
1946         return err;
1947 }
1948
1949 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1950                                 struct btrfs_root *root,
1951                                 struct btrfs_delayed_ref_node *node,
1952                                 struct btrfs_delayed_extent_op *extent_op,
1953                                 int insert_reserved)
1954 {
1955         int ret = 0;
1956         struct btrfs_delayed_data_ref *ref;
1957         struct btrfs_key ins;
1958         u64 parent = 0;
1959         u64 ref_root = 0;
1960         u64 flags = 0;
1961
1962         ins.objectid = node->bytenr;
1963         ins.offset = node->num_bytes;
1964         ins.type = BTRFS_EXTENT_ITEM_KEY;
1965
1966         ref = btrfs_delayed_node_to_data_ref(node);
1967         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1968                 parent = ref->parent;
1969         else
1970                 ref_root = ref->root;
1971
1972         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1973                 if (extent_op) {
1974                         BUG_ON(extent_op->update_key);
1975                         flags |= extent_op->flags_to_set;
1976                 }
1977                 ret = alloc_reserved_file_extent(trans, root,
1978                                                  parent, ref_root, flags,
1979                                                  ref->objectid, ref->offset,
1980                                                  &ins, node->ref_mod);
1981         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1982                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1983                                              node->num_bytes, parent,
1984                                              ref_root, ref->objectid,
1985                                              ref->offset, node->ref_mod,
1986                                              extent_op);
1987         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1988                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1989                                           node->num_bytes, parent,
1990                                           ref_root, ref->objectid,
1991                                           ref->offset, node->ref_mod,
1992                                           extent_op);
1993         } else {
1994                 BUG();
1995         }
1996         return ret;
1997 }
1998
1999 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2000                                     struct extent_buffer *leaf,
2001                                     struct btrfs_extent_item *ei)
2002 {
2003         u64 flags = btrfs_extent_flags(leaf, ei);
2004         if (extent_op->update_flags) {
2005                 flags |= extent_op->flags_to_set;
2006                 btrfs_set_extent_flags(leaf, ei, flags);
2007         }
2008
2009         if (extent_op->update_key) {
2010                 struct btrfs_tree_block_info *bi;
2011                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2012                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2013                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2014         }
2015 }
2016
2017 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2018                                  struct btrfs_root *root,
2019                                  struct btrfs_delayed_ref_node *node,
2020                                  struct btrfs_delayed_extent_op *extent_op)
2021 {
2022         struct btrfs_key key;
2023         struct btrfs_path *path;
2024         struct btrfs_extent_item *ei;
2025         struct extent_buffer *leaf;
2026         u32 item_size;
2027         int ret;
2028         int err = 0;
2029
2030         path = btrfs_alloc_path();
2031         if (!path)
2032                 return -ENOMEM;
2033
2034         key.objectid = node->bytenr;
2035         key.type = BTRFS_EXTENT_ITEM_KEY;
2036         key.offset = node->num_bytes;
2037
2038         path->reada = 1;
2039         path->leave_spinning = 1;
2040         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2041                                 path, 0, 1);
2042         if (ret < 0) {
2043                 err = ret;
2044                 goto out;
2045         }
2046         if (ret > 0) {
2047                 err = -EIO;
2048                 goto out;
2049         }
2050
2051         leaf = path->nodes[0];
2052         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2053 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2054         if (item_size < sizeof(*ei)) {
2055                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2056                                              path, (u64)-1, 0);
2057                 if (ret < 0) {
2058                         err = ret;
2059                         goto out;
2060                 }
2061                 leaf = path->nodes[0];
2062                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2063         }
2064 #endif
2065         BUG_ON(item_size < sizeof(*ei));
2066         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2067         __run_delayed_extent_op(extent_op, leaf, ei);
2068
2069         btrfs_mark_buffer_dirty(leaf);
2070 out:
2071         btrfs_free_path(path);
2072         return err;
2073 }
2074
2075 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2076                                 struct btrfs_root *root,
2077                                 struct btrfs_delayed_ref_node *node,
2078                                 struct btrfs_delayed_extent_op *extent_op,
2079                                 int insert_reserved)
2080 {
2081         int ret = 0;
2082         struct btrfs_delayed_tree_ref *ref;
2083         struct btrfs_key ins;
2084         u64 parent = 0;
2085         u64 ref_root = 0;
2086
2087         ins.objectid = node->bytenr;
2088         ins.offset = node->num_bytes;
2089         ins.type = BTRFS_EXTENT_ITEM_KEY;
2090
2091         ref = btrfs_delayed_node_to_tree_ref(node);
2092         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2093                 parent = ref->parent;
2094         else
2095                 ref_root = ref->root;
2096
2097         BUG_ON(node->ref_mod != 1);
2098         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2099                 BUG_ON(!extent_op || !extent_op->update_flags ||
2100                        !extent_op->update_key);
2101                 ret = alloc_reserved_tree_block(trans, root,
2102                                                 parent, ref_root,
2103                                                 extent_op->flags_to_set,
2104                                                 &extent_op->key,
2105                                                 ref->level, &ins);
2106         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2107                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2108                                              node->num_bytes, parent, ref_root,
2109                                              ref->level, 0, 1, extent_op);
2110         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2111                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2112                                           node->num_bytes, parent, ref_root,
2113                                           ref->level, 0, 1, extent_op);
2114         } else {
2115                 BUG();
2116         }
2117         return ret;
2118 }
2119
2120 /* helper function to actually process a single delayed ref entry */
2121 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2122                                struct btrfs_root *root,
2123                                struct btrfs_delayed_ref_node *node,
2124                                struct btrfs_delayed_extent_op *extent_op,
2125                                int insert_reserved)
2126 {
2127         int ret;
2128         if (btrfs_delayed_ref_is_head(node)) {
2129                 struct btrfs_delayed_ref_head *head;
2130                 /*
2131                  * we've hit the end of the chain and we were supposed
2132                  * to insert this extent into the tree.  But, it got
2133                  * deleted before we ever needed to insert it, so all
2134                  * we have to do is clean up the accounting
2135                  */
2136                 BUG_ON(extent_op);
2137                 head = btrfs_delayed_node_to_head(node);
2138                 if (insert_reserved) {
2139                         btrfs_pin_extent(root, node->bytenr,
2140                                          node->num_bytes, 1);
2141                         if (head->is_data) {
2142                                 ret = btrfs_del_csums(trans, root,
2143                                                       node->bytenr,
2144                                                       node->num_bytes);
2145                                 BUG_ON(ret);
2146                         }
2147                 }
2148                 mutex_unlock(&head->mutex);
2149                 return 0;
2150         }
2151
2152         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2153             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2154                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2155                                            insert_reserved);
2156         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2157                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2158                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2159                                            insert_reserved);
2160         else
2161                 BUG();
2162         return ret;
2163 }
2164
2165 static noinline struct btrfs_delayed_ref_node *
2166 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2167 {
2168         struct rb_node *node;
2169         struct btrfs_delayed_ref_node *ref;
2170         int action = BTRFS_ADD_DELAYED_REF;
2171 again:
2172         /*
2173          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2174          * this prevents ref count from going down to zero when
2175          * there still are pending delayed ref.
2176          */
2177         node = rb_prev(&head->node.rb_node);
2178         while (1) {
2179                 if (!node)
2180                         break;
2181                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2182                                 rb_node);
2183                 if (ref->bytenr != head->node.bytenr)
2184                         break;
2185                 if (ref->action == action)
2186                         return ref;
2187                 node = rb_prev(node);
2188         }
2189         if (action == BTRFS_ADD_DELAYED_REF) {
2190                 action = BTRFS_DROP_DELAYED_REF;
2191                 goto again;
2192         }
2193         return NULL;
2194 }
2195
2196 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2197                                        struct btrfs_root *root,
2198                                        struct list_head *cluster)
2199 {
2200         struct btrfs_delayed_ref_root *delayed_refs;
2201         struct btrfs_delayed_ref_node *ref;
2202         struct btrfs_delayed_ref_head *locked_ref = NULL;
2203         struct btrfs_delayed_extent_op *extent_op;
2204         int ret;
2205         int count = 0;
2206         int must_insert_reserved = 0;
2207
2208         delayed_refs = &trans->transaction->delayed_refs;
2209         while (1) {
2210                 if (!locked_ref) {
2211                         /* pick a new head ref from the cluster list */
2212                         if (list_empty(cluster))
2213                                 break;
2214
2215                         locked_ref = list_entry(cluster->next,
2216                                      struct btrfs_delayed_ref_head, cluster);
2217
2218                         /* grab the lock that says we are going to process
2219                          * all the refs for this head */
2220                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2221
2222                         /*
2223                          * we may have dropped the spin lock to get the head
2224                          * mutex lock, and that might have given someone else
2225                          * time to free the head.  If that's true, it has been
2226                          * removed from our list and we can move on.
2227                          */
2228                         if (ret == -EAGAIN) {
2229                                 locked_ref = NULL;
2230                                 count++;
2231                                 continue;
2232                         }
2233                 }
2234
2235                 /*
2236                  * record the must insert reserved flag before we
2237                  * drop the spin lock.
2238                  */
2239                 must_insert_reserved = locked_ref->must_insert_reserved;
2240                 locked_ref->must_insert_reserved = 0;
2241
2242                 extent_op = locked_ref->extent_op;
2243                 locked_ref->extent_op = NULL;
2244
2245                 /*
2246                  * locked_ref is the head node, so we have to go one
2247                  * node back for any delayed ref updates
2248                  */
2249                 ref = select_delayed_ref(locked_ref);
2250                 if (!ref) {
2251                         /* All delayed refs have been processed, Go ahead
2252                          * and send the head node to run_one_delayed_ref,
2253                          * so that any accounting fixes can happen
2254                          */
2255                         ref = &locked_ref->node;
2256
2257                         if (extent_op && must_insert_reserved) {
2258                                 kfree(extent_op);
2259                                 extent_op = NULL;
2260                         }
2261
2262                         if (extent_op) {
2263                                 spin_unlock(&delayed_refs->lock);
2264
2265                                 ret = run_delayed_extent_op(trans, root,
2266                                                             ref, extent_op);
2267                                 BUG_ON(ret);
2268                                 kfree(extent_op);
2269
2270                                 cond_resched();
2271                                 spin_lock(&delayed_refs->lock);
2272                                 continue;
2273                         }
2274
2275                         list_del_init(&locked_ref->cluster);
2276                         locked_ref = NULL;
2277                 }
2278
2279                 ref->in_tree = 0;
2280                 rb_erase(&ref->rb_node, &delayed_refs->root);
2281                 delayed_refs->num_entries--;
2282
2283                 spin_unlock(&delayed_refs->lock);
2284
2285                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2286                                           must_insert_reserved);
2287                 BUG_ON(ret);
2288
2289                 btrfs_put_delayed_ref(ref);
2290                 kfree(extent_op);
2291                 count++;
2292
2293                 cond_resched();
2294                 spin_lock(&delayed_refs->lock);
2295         }
2296         return count;
2297 }
2298
2299 /*
2300  * this starts processing the delayed reference count updates and
2301  * extent insertions we have queued up so far.  count can be
2302  * 0, which means to process everything in the tree at the start
2303  * of the run (but not newly added entries), or it can be some target
2304  * number you'd like to process.
2305  */
2306 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2307                            struct btrfs_root *root, unsigned long count)
2308 {
2309         struct rb_node *node;
2310         struct btrfs_delayed_ref_root *delayed_refs;
2311         struct btrfs_delayed_ref_node *ref;
2312         struct list_head cluster;
2313         int ret;
2314         int run_all = count == (unsigned long)-1;
2315         int run_most = 0;
2316
2317         if (root == root->fs_info->extent_root)
2318                 root = root->fs_info->tree_root;
2319
2320         delayed_refs = &trans->transaction->delayed_refs;
2321         INIT_LIST_HEAD(&cluster);
2322 again:
2323         spin_lock(&delayed_refs->lock);
2324         if (count == 0) {
2325                 count = delayed_refs->num_entries * 2;
2326                 run_most = 1;
2327         }
2328         while (1) {
2329                 if (!(run_all || run_most) &&
2330                     delayed_refs->num_heads_ready < 64)
2331                         break;
2332
2333                 /*
2334                  * go find something we can process in the rbtree.  We start at
2335                  * the beginning of the tree, and then build a cluster
2336                  * of refs to process starting at the first one we are able to
2337                  * lock
2338                  */
2339                 ret = btrfs_find_ref_cluster(trans, &cluster,
2340                                              delayed_refs->run_delayed_start);
2341                 if (ret)
2342                         break;
2343
2344                 ret = run_clustered_refs(trans, root, &cluster);
2345                 BUG_ON(ret < 0);
2346
2347                 count -= min_t(unsigned long, ret, count);
2348
2349                 if (count == 0)
2350                         break;
2351         }
2352
2353         if (run_all) {
2354                 node = rb_first(&delayed_refs->root);
2355                 if (!node)
2356                         goto out;
2357                 count = (unsigned long)-1;
2358
2359                 while (node) {
2360                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2361                                        rb_node);
2362                         if (btrfs_delayed_ref_is_head(ref)) {
2363                                 struct btrfs_delayed_ref_head *head;
2364
2365                                 head = btrfs_delayed_node_to_head(ref);
2366                                 atomic_inc(&ref->refs);
2367
2368                                 spin_unlock(&delayed_refs->lock);
2369                                 /*
2370                                  * Mutex was contended, block until it's
2371                                  * released and try again
2372                                  */
2373                                 mutex_lock(&head->mutex);
2374                                 mutex_unlock(&head->mutex);
2375
2376                                 btrfs_put_delayed_ref(ref);
2377                                 cond_resched();
2378                                 goto again;
2379                         }
2380                         node = rb_next(node);
2381                 }
2382                 spin_unlock(&delayed_refs->lock);
2383                 schedule_timeout(1);
2384                 goto again;
2385         }
2386 out:
2387         spin_unlock(&delayed_refs->lock);
2388         return 0;
2389 }
2390
2391 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2392                                 struct btrfs_root *root,
2393                                 u64 bytenr, u64 num_bytes, u64 flags,
2394                                 int is_data)
2395 {
2396         struct btrfs_delayed_extent_op *extent_op;
2397         int ret;
2398
2399         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2400         if (!extent_op)
2401                 return -ENOMEM;
2402
2403         extent_op->flags_to_set = flags;
2404         extent_op->update_flags = 1;
2405         extent_op->update_key = 0;
2406         extent_op->is_data = is_data ? 1 : 0;
2407
2408         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2409         if (ret)
2410                 kfree(extent_op);
2411         return ret;
2412 }
2413
2414 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2415                                       struct btrfs_root *root,
2416                                       struct btrfs_path *path,
2417                                       u64 objectid, u64 offset, u64 bytenr)
2418 {
2419         struct btrfs_delayed_ref_head *head;
2420         struct btrfs_delayed_ref_node *ref;
2421         struct btrfs_delayed_data_ref *data_ref;
2422         struct btrfs_delayed_ref_root *delayed_refs;
2423         struct rb_node *node;
2424         int ret = 0;
2425
2426         ret = -ENOENT;
2427         delayed_refs = &trans->transaction->delayed_refs;
2428         spin_lock(&delayed_refs->lock);
2429         head = btrfs_find_delayed_ref_head(trans, bytenr);
2430         if (!head)
2431                 goto out;
2432
2433         if (!mutex_trylock(&head->mutex)) {
2434                 atomic_inc(&head->node.refs);
2435                 spin_unlock(&delayed_refs->lock);
2436
2437                 btrfs_release_path(path);
2438
2439                 /*
2440                  * Mutex was contended, block until it's released and let
2441                  * caller try again
2442                  */
2443                 mutex_lock(&head->mutex);
2444                 mutex_unlock(&head->mutex);
2445                 btrfs_put_delayed_ref(&head->node);
2446                 return -EAGAIN;
2447         }
2448
2449         node = rb_prev(&head->node.rb_node);
2450         if (!node)
2451                 goto out_unlock;
2452
2453         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2454
2455         if (ref->bytenr != bytenr)
2456                 goto out_unlock;
2457
2458         ret = 1;
2459         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2460                 goto out_unlock;
2461
2462         data_ref = btrfs_delayed_node_to_data_ref(ref);
2463
2464         node = rb_prev(node);
2465         if (node) {
2466                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2467                 if (ref->bytenr == bytenr)
2468                         goto out_unlock;
2469         }
2470
2471         if (data_ref->root != root->root_key.objectid ||
2472             data_ref->objectid != objectid || data_ref->offset != offset)
2473                 goto out_unlock;
2474
2475         ret = 0;
2476 out_unlock:
2477         mutex_unlock(&head->mutex);
2478 out:
2479         spin_unlock(&delayed_refs->lock);
2480         return ret;
2481 }
2482
2483 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2484                                         struct btrfs_root *root,
2485                                         struct btrfs_path *path,
2486                                         u64 objectid, u64 offset, u64 bytenr)
2487 {
2488         struct btrfs_root *extent_root = root->fs_info->extent_root;
2489         struct extent_buffer *leaf;
2490         struct btrfs_extent_data_ref *ref;
2491         struct btrfs_extent_inline_ref *iref;
2492         struct btrfs_extent_item *ei;
2493         struct btrfs_key key;
2494         u32 item_size;
2495         int ret;
2496
2497         key.objectid = bytenr;
2498         key.offset = (u64)-1;
2499         key.type = BTRFS_EXTENT_ITEM_KEY;
2500
2501         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2502         if (ret < 0)
2503                 goto out;
2504         BUG_ON(ret == 0);
2505
2506         ret = -ENOENT;
2507         if (path->slots[0] == 0)
2508                 goto out;
2509
2510         path->slots[0]--;
2511         leaf = path->nodes[0];
2512         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2513
2514         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2515                 goto out;
2516
2517         ret = 1;
2518         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2519 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2520         if (item_size < sizeof(*ei)) {
2521                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2522                 goto out;
2523         }
2524 #endif
2525         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2526
2527         if (item_size != sizeof(*ei) +
2528             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2529                 goto out;
2530
2531         if (btrfs_extent_generation(leaf, ei) <=
2532             btrfs_root_last_snapshot(&root->root_item))
2533                 goto out;
2534
2535         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2536         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2537             BTRFS_EXTENT_DATA_REF_KEY)
2538                 goto out;
2539
2540         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2541         if (btrfs_extent_refs(leaf, ei) !=
2542             btrfs_extent_data_ref_count(leaf, ref) ||
2543             btrfs_extent_data_ref_root(leaf, ref) !=
2544             root->root_key.objectid ||
2545             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2546             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2547                 goto out;
2548
2549         ret = 0;
2550 out:
2551         return ret;
2552 }
2553
2554 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2555                           struct btrfs_root *root,
2556                           u64 objectid, u64 offset, u64 bytenr)
2557 {
2558         struct btrfs_path *path;
2559         int ret;
2560         int ret2;
2561
2562         path = btrfs_alloc_path();
2563         if (!path)
2564                 return -ENOENT;
2565
2566         do {
2567                 ret = check_committed_ref(trans, root, path, objectid,
2568                                           offset, bytenr);
2569                 if (ret && ret != -ENOENT)
2570                         goto out;
2571
2572                 ret2 = check_delayed_ref(trans, root, path, objectid,
2573                                          offset, bytenr);
2574         } while (ret2 == -EAGAIN);
2575
2576         if (ret2 && ret2 != -ENOENT) {
2577                 ret = ret2;
2578                 goto out;
2579         }
2580
2581         if (ret != -ENOENT || ret2 != -ENOENT)
2582                 ret = 0;
2583 out:
2584         btrfs_free_path(path);
2585         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2586                 WARN_ON(ret > 0);
2587         return ret;
2588 }
2589
2590 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2591                            struct btrfs_root *root,
2592                            struct extent_buffer *buf,
2593                            int full_backref, int inc)
2594 {
2595         u64 bytenr;
2596         u64 num_bytes;
2597         u64 parent;
2598         u64 ref_root;
2599         u32 nritems;
2600         struct btrfs_key key;
2601         struct btrfs_file_extent_item *fi;
2602         int i;
2603         int level;
2604         int ret = 0;
2605         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2606                             u64, u64, u64, u64, u64, u64);
2607
2608         ref_root = btrfs_header_owner(buf);
2609         nritems = btrfs_header_nritems(buf);
2610         level = btrfs_header_level(buf);
2611
2612         if (!root->ref_cows && level == 0)
2613                 return 0;
2614
2615         if (inc)
2616                 process_func = btrfs_inc_extent_ref;
2617         else
2618                 process_func = btrfs_free_extent;
2619
2620         if (full_backref)
2621                 parent = buf->start;
2622         else
2623                 parent = 0;
2624
2625         for (i = 0; i < nritems; i++) {
2626                 if (level == 0) {
2627                         btrfs_item_key_to_cpu(buf, &key, i);
2628                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2629                                 continue;
2630                         fi = btrfs_item_ptr(buf, i,
2631                                             struct btrfs_file_extent_item);
2632                         if (btrfs_file_extent_type(buf, fi) ==
2633                             BTRFS_FILE_EXTENT_INLINE)
2634                                 continue;
2635                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2636                         if (bytenr == 0)
2637                                 continue;
2638
2639                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2640                         key.offset -= btrfs_file_extent_offset(buf, fi);
2641                         ret = process_func(trans, root, bytenr, num_bytes,
2642                                            parent, ref_root, key.objectid,
2643                                            key.offset);
2644                         if (ret)
2645                                 goto fail;
2646                 } else {
2647                         bytenr = btrfs_node_blockptr(buf, i);
2648                         num_bytes = btrfs_level_size(root, level - 1);
2649                         ret = process_func(trans, root, bytenr, num_bytes,
2650                                            parent, ref_root, level - 1, 0);
2651                         if (ret)
2652                                 goto fail;
2653                 }
2654         }
2655         return 0;
2656 fail:
2657         BUG();
2658         return ret;
2659 }
2660
2661 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2662                   struct extent_buffer *buf, int full_backref)
2663 {
2664         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2665 }
2666
2667 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2668                   struct extent_buffer *buf, int full_backref)
2669 {
2670         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2671 }
2672
2673 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2674                                  struct btrfs_root *root,
2675                                  struct btrfs_path *path,
2676                                  struct btrfs_block_group_cache *cache)
2677 {
2678         int ret;
2679         struct btrfs_root *extent_root = root->fs_info->extent_root;
2680         unsigned long bi;
2681         struct extent_buffer *leaf;
2682
2683         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2684         if (ret < 0)
2685                 goto fail;
2686         BUG_ON(ret);
2687
2688         leaf = path->nodes[0];
2689         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2690         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2691         btrfs_mark_buffer_dirty(leaf);
2692         btrfs_release_path(path);
2693 fail:
2694         if (ret)
2695                 return ret;
2696         return 0;
2697
2698 }
2699
2700 static struct btrfs_block_group_cache *
2701 next_block_group(struct btrfs_root *root,
2702                  struct btrfs_block_group_cache *cache)
2703 {
2704         struct rb_node *node;
2705         spin_lock(&root->fs_info->block_group_cache_lock);
2706         node = rb_next(&cache->cache_node);
2707         btrfs_put_block_group(cache);
2708         if (node) {
2709                 cache = rb_entry(node, struct btrfs_block_group_cache,
2710                                  cache_node);
2711                 btrfs_get_block_group(cache);
2712         } else
2713                 cache = NULL;
2714         spin_unlock(&root->fs_info->block_group_cache_lock);
2715         return cache;
2716 }
2717
2718 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2719                             struct btrfs_trans_handle *trans,
2720                             struct btrfs_path *path)
2721 {
2722         struct btrfs_root *root = block_group->fs_info->tree_root;
2723         struct inode *inode = NULL;
2724         u64 alloc_hint = 0;
2725         int dcs = BTRFS_DC_ERROR;
2726         int num_pages = 0;
2727         int retries = 0;
2728         int ret = 0;
2729
2730         /*
2731          * If this block group is smaller than 100 megs don't bother caching the
2732          * block group.
2733          */
2734         if (block_group->key.offset < (100 * 1024 * 1024)) {
2735                 spin_lock(&block_group->lock);
2736                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2737                 spin_unlock(&block_group->lock);
2738                 return 0;
2739         }
2740
2741 again:
2742         inode = lookup_free_space_inode(root, block_group, path);
2743         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2744                 ret = PTR_ERR(inode);
2745                 btrfs_release_path(path);
2746                 goto out;
2747         }
2748
2749         if (IS_ERR(inode)) {
2750                 BUG_ON(retries);
2751                 retries++;
2752
2753                 if (block_group->ro)
2754                         goto out_free;
2755
2756                 ret = create_free_space_inode(root, trans, block_group, path);
2757                 if (ret)
2758                         goto out_free;
2759                 goto again;
2760         }
2761
2762         /* We've already setup this transaction, go ahead and exit */
2763         if (block_group->cache_generation == trans->transid &&
2764             i_size_read(inode)) {
2765                 dcs = BTRFS_DC_SETUP;
2766                 goto out_put;
2767         }
2768
2769         /*
2770          * We want to set the generation to 0, that way if anything goes wrong
2771          * from here on out we know not to trust this cache when we load up next
2772          * time.
2773          */
2774         BTRFS_I(inode)->generation = 0;
2775         ret = btrfs_update_inode(trans, root, inode);
2776         WARN_ON(ret);
2777
2778         if (i_size_read(inode) > 0) {
2779                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2780                                                       inode);
2781                 if (ret)
2782                         goto out_put;
2783         }
2784
2785         spin_lock(&block_group->lock);
2786         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2787                 /* We're not cached, don't bother trying to write stuff out */
2788                 dcs = BTRFS_DC_WRITTEN;
2789                 spin_unlock(&block_group->lock);
2790                 goto out_put;
2791         }
2792         spin_unlock(&block_group->lock);
2793
2794         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2795         if (!num_pages)
2796                 num_pages = 1;
2797
2798         /*
2799          * Just to make absolutely sure we have enough space, we're going to
2800          * preallocate 12 pages worth of space for each block group.  In
2801          * practice we ought to use at most 8, but we need extra space so we can
2802          * add our header and have a terminator between the extents and the
2803          * bitmaps.
2804          */
2805         num_pages *= 16;
2806         num_pages *= PAGE_CACHE_SIZE;
2807
2808         ret = btrfs_check_data_free_space(inode, num_pages);
2809         if (ret)
2810                 goto out_put;
2811
2812         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2813                                               num_pages, num_pages,
2814                                               &alloc_hint);
2815         if (!ret)
2816                 dcs = BTRFS_DC_SETUP;
2817         btrfs_free_reserved_data_space(inode, num_pages);
2818
2819 out_put:
2820         iput(inode);
2821 out_free:
2822         btrfs_release_path(path);
2823 out:
2824         spin_lock(&block_group->lock);
2825         if (!ret)
2826                 block_group->cache_generation = trans->transid;
2827         block_group->disk_cache_state = dcs;
2828         spin_unlock(&block_group->lock);
2829
2830         return ret;
2831 }
2832
2833 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2834                                    struct btrfs_root *root)
2835 {
2836         struct btrfs_block_group_cache *cache;
2837         int err = 0;
2838         struct btrfs_path *path;
2839         u64 last = 0;
2840
2841         path = btrfs_alloc_path();
2842         if (!path)
2843                 return -ENOMEM;
2844
2845 again:
2846         while (1) {
2847                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2848                 while (cache) {
2849                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2850                                 break;
2851                         cache = next_block_group(root, cache);
2852                 }
2853                 if (!cache) {
2854                         if (last == 0)
2855                                 break;
2856                         last = 0;
2857                         continue;
2858                 }
2859                 err = cache_save_setup(cache, trans, path);
2860                 last = cache->key.objectid + cache->key.offset;
2861                 btrfs_put_block_group(cache);
2862         }
2863
2864         while (1) {
2865                 if (last == 0) {
2866                         err = btrfs_run_delayed_refs(trans, root,
2867                                                      (unsigned long)-1);
2868                         BUG_ON(err);
2869                 }
2870
2871                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2872                 while (cache) {
2873                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2874                                 btrfs_put_block_group(cache);
2875                                 goto again;
2876                         }
2877
2878                         if (cache->dirty)
2879                                 break;
2880                         cache = next_block_group(root, cache);
2881                 }
2882                 if (!cache) {
2883                         if (last == 0)
2884                                 break;
2885                         last = 0;
2886                         continue;
2887                 }
2888
2889                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2890                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2891                 cache->dirty = 0;
2892                 last = cache->key.objectid + cache->key.offset;
2893
2894                 err = write_one_cache_group(trans, root, path, cache);
2895                 BUG_ON(err);
2896                 btrfs_put_block_group(cache);
2897         }
2898
2899         while (1) {
2900                 /*
2901                  * I don't think this is needed since we're just marking our
2902                  * preallocated extent as written, but just in case it can't
2903                  * hurt.
2904                  */
2905                 if (last == 0) {
2906                         err = btrfs_run_delayed_refs(trans, root,
2907                                                      (unsigned long)-1);
2908                         BUG_ON(err);
2909                 }
2910
2911                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2912                 while (cache) {
2913                         /*
2914                          * Really this shouldn't happen, but it could if we
2915                          * couldn't write the entire preallocated extent and
2916                          * splitting the extent resulted in a new block.
2917                          */
2918                         if (cache->dirty) {
2919                                 btrfs_put_block_group(cache);
2920                                 goto again;
2921                         }
2922                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2923                                 break;
2924                         cache = next_block_group(root, cache);
2925                 }
2926                 if (!cache) {
2927                         if (last == 0)
2928                                 break;
2929                         last = 0;
2930                         continue;
2931                 }
2932
2933                 btrfs_write_out_cache(root, trans, cache, path);
2934
2935                 /*
2936                  * If we didn't have an error then the cache state is still
2937                  * NEED_WRITE, so we can set it to WRITTEN.
2938                  */
2939                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2940                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2941                 last = cache->key.objectid + cache->key.offset;
2942                 btrfs_put_block_group(cache);
2943         }
2944
2945         btrfs_free_path(path);
2946         return 0;
2947 }
2948
2949 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2950 {
2951         struct btrfs_block_group_cache *block_group;
2952         int readonly = 0;
2953
2954         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2955         if (!block_group || block_group->ro)
2956                 readonly = 1;
2957         if (block_group)
2958                 btrfs_put_block_group(block_group);
2959         return readonly;
2960 }
2961
2962 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2963                              u64 total_bytes, u64 bytes_used,
2964                              struct btrfs_space_info **space_info)
2965 {
2966         struct btrfs_space_info *found;
2967         int i;
2968         int factor;
2969
2970         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2971                      BTRFS_BLOCK_GROUP_RAID10))
2972                 factor = 2;
2973         else
2974                 factor = 1;
2975
2976         found = __find_space_info(info, flags);
2977         if (found) {
2978                 spin_lock(&found->lock);
2979                 found->total_bytes += total_bytes;
2980                 found->disk_total += total_bytes * factor;
2981                 found->bytes_used += bytes_used;
2982                 found->disk_used += bytes_used * factor;
2983                 found->full = 0;
2984                 spin_unlock(&found->lock);
2985                 *space_info = found;
2986                 return 0;
2987         }
2988         found = kzalloc(sizeof(*found), GFP_NOFS);
2989         if (!found)
2990                 return -ENOMEM;
2991
2992         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2993                 INIT_LIST_HEAD(&found->block_groups[i]);
2994         init_rwsem(&found->groups_sem);
2995         spin_lock_init(&found->lock);
2996         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2997                                 BTRFS_BLOCK_GROUP_SYSTEM |
2998                                 BTRFS_BLOCK_GROUP_METADATA);
2999         found->total_bytes = total_bytes;
3000         found->disk_total = total_bytes * factor;
3001         found->bytes_used = bytes_used;
3002         found->disk_used = bytes_used * factor;
3003         found->bytes_pinned = 0;
3004         found->bytes_reserved = 0;
3005         found->bytes_readonly = 0;
3006         found->bytes_may_use = 0;
3007         found->full = 0;
3008         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3009         found->chunk_alloc = 0;
3010         found->flush = 0;
3011         init_waitqueue_head(&found->wait);
3012         *space_info = found;
3013         list_add_rcu(&found->list, &info->space_info);
3014         return 0;
3015 }
3016
3017 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3018 {
3019         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3020                                    BTRFS_BLOCK_GROUP_RAID1 |
3021                                    BTRFS_BLOCK_GROUP_RAID10 |
3022                                    BTRFS_BLOCK_GROUP_DUP);
3023         if (extra_flags) {
3024                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3025                         fs_info->avail_data_alloc_bits |= extra_flags;
3026                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3027                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3028                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3029                         fs_info->avail_system_alloc_bits |= extra_flags;
3030         }
3031 }
3032
3033 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3034 {
3035         /*
3036          * we add in the count of missing devices because we want
3037          * to make sure that any RAID levels on a degraded FS
3038          * continue to be honored.
3039          */
3040         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3041                 root->fs_info->fs_devices->missing_devices;
3042
3043         if (num_devices == 1)
3044                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3045         if (num_devices < 4)
3046                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3047
3048         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3049             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3050                       BTRFS_BLOCK_GROUP_RAID10))) {
3051                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3052         }
3053
3054         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3055             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3056                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3057         }
3058
3059         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3060             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3061              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3062              (flags & BTRFS_BLOCK_GROUP_DUP)))
3063                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3064         return flags;
3065 }
3066
3067 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3068 {
3069         if (flags & BTRFS_BLOCK_GROUP_DATA)
3070                 flags |= root->fs_info->avail_data_alloc_bits &
3071                          root->fs_info->data_alloc_profile;
3072         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3073                 flags |= root->fs_info->avail_system_alloc_bits &
3074                          root->fs_info->system_alloc_profile;
3075         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3076                 flags |= root->fs_info->avail_metadata_alloc_bits &
3077                          root->fs_info->metadata_alloc_profile;
3078         return btrfs_reduce_alloc_profile(root, flags);
3079 }
3080
3081 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3082 {
3083         u64 flags;
3084
3085         if (data)
3086                 flags = BTRFS_BLOCK_GROUP_DATA;
3087         else if (root == root->fs_info->chunk_root)
3088                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3089         else
3090                 flags = BTRFS_BLOCK_GROUP_METADATA;
3091
3092         return get_alloc_profile(root, flags);
3093 }
3094
3095 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3096 {
3097         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3098                                                        BTRFS_BLOCK_GROUP_DATA);
3099 }
3100
3101 /*
3102  * This will check the space that the inode allocates from to make sure we have
3103  * enough space for bytes.
3104  */
3105 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3106 {
3107         struct btrfs_space_info *data_sinfo;
3108         struct btrfs_root *root = BTRFS_I(inode)->root;
3109         u64 used;
3110         int ret = 0, committed = 0, alloc_chunk = 1;
3111
3112         /* make sure bytes are sectorsize aligned */
3113         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3114
3115         if (root == root->fs_info->tree_root ||
3116             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3117                 alloc_chunk = 0;
3118                 committed = 1;
3119         }
3120
3121         data_sinfo = BTRFS_I(inode)->space_info;
3122         if (!data_sinfo)
3123                 goto alloc;
3124
3125 again:
3126         /* make sure we have enough space to handle the data first */
3127         spin_lock(&data_sinfo->lock);
3128         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3129                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3130                 data_sinfo->bytes_may_use;
3131
3132         if (used + bytes > data_sinfo->total_bytes) {
3133                 struct btrfs_trans_handle *trans;
3134
3135                 /*
3136                  * if we don't have enough free bytes in this space then we need
3137                  * to alloc a new chunk.
3138                  */
3139                 if (!data_sinfo->full && alloc_chunk) {
3140                         u64 alloc_target;
3141
3142                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3143                         spin_unlock(&data_sinfo->lock);
3144 alloc:
3145                         alloc_target = btrfs_get_alloc_profile(root, 1);
3146                         trans = btrfs_join_transaction(root);
3147                         if (IS_ERR(trans))
3148                                 return PTR_ERR(trans);
3149
3150                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3151                                              bytes + 2 * 1024 * 1024,
3152                                              alloc_target,
3153                                              CHUNK_ALLOC_NO_FORCE);
3154                         btrfs_end_transaction(trans, root);
3155                         if (ret < 0) {
3156                                 if (ret != -ENOSPC)
3157                                         return ret;
3158                                 else
3159                                         goto commit_trans;
3160                         }
3161
3162                         if (!data_sinfo) {
3163                                 btrfs_set_inode_space_info(root, inode);
3164                                 data_sinfo = BTRFS_I(inode)->space_info;
3165                         }
3166                         goto again;
3167                 }
3168
3169                 /*
3170                  * If we have less pinned bytes than we want to allocate then
3171                  * don't bother committing the transaction, it won't help us.
3172                  */
3173                 if (data_sinfo->bytes_pinned < bytes)
3174                         committed = 1;
3175                 spin_unlock(&data_sinfo->lock);
3176
3177                 /* commit the current transaction and try again */
3178 commit_trans:
3179                 if (!committed &&
3180                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3181                         committed = 1;
3182                         trans = btrfs_join_transaction(root);
3183                         if (IS_ERR(trans))
3184                                 return PTR_ERR(trans);
3185                         ret = btrfs_commit_transaction(trans, root);
3186                         if (ret)
3187                                 return ret;
3188                         goto again;
3189                 }
3190
3191                 return -ENOSPC;
3192         }
3193         data_sinfo->bytes_may_use += bytes;
3194         spin_unlock(&data_sinfo->lock);
3195
3196         return 0;
3197 }
3198
3199 /*
3200  * Called if we need to clear a data reservation for this inode.
3201  */
3202 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3203 {
3204         struct btrfs_root *root = BTRFS_I(inode)->root;
3205         struct btrfs_space_info *data_sinfo;
3206
3207         /* make sure bytes are sectorsize aligned */
3208         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3209
3210         data_sinfo = BTRFS_I(inode)->space_info;
3211         spin_lock(&data_sinfo->lock);
3212         data_sinfo->bytes_may_use -= bytes;
3213         spin_unlock(&data_sinfo->lock);
3214 }
3215
3216 static void force_metadata_allocation(struct btrfs_fs_info *info)
3217 {
3218         struct list_head *head = &info->space_info;
3219         struct btrfs_space_info *found;
3220
3221         rcu_read_lock();
3222         list_for_each_entry_rcu(found, head, list) {
3223                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3224                         found->force_alloc = CHUNK_ALLOC_FORCE;
3225         }
3226         rcu_read_unlock();
3227 }
3228
3229 static int should_alloc_chunk(struct btrfs_root *root,
3230                               struct btrfs_space_info *sinfo, u64 alloc_bytes,
3231                               int force)
3232 {
3233         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3234         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3235         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3236         u64 thresh;
3237
3238         if (force == CHUNK_ALLOC_FORCE)
3239                 return 1;
3240
3241         /*
3242          * We need to take into account the global rsv because for all intents
3243          * and purposes it's used space.  Don't worry about locking the
3244          * global_rsv, it doesn't change except when the transaction commits.
3245          */
3246         num_allocated += global_rsv->size;
3247
3248         /*
3249          * in limited mode, we want to have some free space up to
3250          * about 1% of the FS size.
3251          */
3252         if (force == CHUNK_ALLOC_LIMITED) {
3253                 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3254                 thresh = max_t(u64, 64 * 1024 * 1024,
3255                                div_factor_fine(thresh, 1));
3256
3257                 if (num_bytes - num_allocated < thresh)
3258                         return 1;
3259         }
3260
3261         /*
3262          * we have two similar checks here, one based on percentage
3263          * and once based on a hard number of 256MB.  The idea
3264          * is that if we have a good amount of free
3265          * room, don't allocate a chunk.  A good mount is
3266          * less than 80% utilized of the chunks we have allocated,
3267          * or more than 256MB free
3268          */
3269         if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3270                 return 0;
3271
3272         if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3273                 return 0;
3274
3275         thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3276
3277         /* 256MB or 5% of the FS */
3278         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3279
3280         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3281                 return 0;
3282         return 1;
3283 }
3284
3285 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3286                           struct btrfs_root *extent_root, u64 alloc_bytes,
3287                           u64 flags, int force)
3288 {
3289         struct btrfs_space_info *space_info;
3290         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3291         int wait_for_alloc = 0;
3292         int ret = 0;
3293
3294         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3295
3296         space_info = __find_space_info(extent_root->fs_info, flags);
3297         if (!space_info) {
3298                 ret = update_space_info(extent_root->fs_info, flags,
3299                                         0, 0, &space_info);
3300                 BUG_ON(ret);
3301         }
3302         BUG_ON(!space_info);
3303
3304 again:
3305         spin_lock(&space_info->lock);
3306         if (space_info->force_alloc)
3307                 force = space_info->force_alloc;
3308         if (space_info->full) {
3309                 spin_unlock(&space_info->lock);
3310                 return 0;
3311         }
3312
3313         if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3314                 spin_unlock(&space_info->lock);
3315                 return 0;
3316         } else if (space_info->chunk_alloc) {
3317                 wait_for_alloc = 1;
3318         } else {
3319                 space_info->chunk_alloc = 1;
3320         }
3321
3322         spin_unlock(&space_info->lock);
3323
3324         mutex_lock(&fs_info->chunk_mutex);
3325
3326         /*
3327          * The chunk_mutex is held throughout the entirety of a chunk
3328          * allocation, so once we've acquired the chunk_mutex we know that the
3329          * other guy is done and we need to recheck and see if we should
3330          * allocate.
3331          */
3332         if (wait_for_alloc) {
3333                 mutex_unlock(&fs_info->chunk_mutex);
3334                 wait_for_alloc = 0;
3335                 goto again;
3336         }
3337
3338         /*
3339          * If we have mixed data/metadata chunks we want to make sure we keep
3340          * allocating mixed chunks instead of individual chunks.
3341          */
3342         if (btrfs_mixed_space_info(space_info))
3343                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3344
3345         /*
3346          * if we're doing a data chunk, go ahead and make sure that
3347          * we keep a reasonable number of metadata chunks allocated in the
3348          * FS as well.
3349          */
3350         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3351                 fs_info->data_chunk_allocations++;
3352                 if (!(fs_info->data_chunk_allocations %
3353                       fs_info->metadata_ratio))
3354                         force_metadata_allocation(fs_info);
3355         }
3356
3357         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3358         if (ret < 0 && ret != -ENOSPC)
3359                 goto out;
3360
3361         spin_lock(&space_info->lock);
3362         if (ret)
3363                 space_info->full = 1;
3364         else
3365                 ret = 1;
3366
3367         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3368         space_info->chunk_alloc = 0;
3369         spin_unlock(&space_info->lock);
3370 out:
3371         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3372         return ret;
3373 }
3374
3375 /*
3376  * shrink metadata reservation for delalloc
3377  */
3378 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3379                            bool wait_ordered)
3380 {
3381         struct btrfs_block_rsv *block_rsv;
3382         struct btrfs_space_info *space_info;
3383         struct btrfs_trans_handle *trans;
3384         u64 reserved;
3385         u64 max_reclaim;
3386         u64 reclaimed = 0;
3387         long time_left;
3388         unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3389         int loops = 0;
3390         unsigned long progress;
3391
3392         trans = (struct btrfs_trans_handle *)current->journal_info;
3393         block_rsv = &root->fs_info->delalloc_block_rsv;
3394         space_info = block_rsv->space_info;
3395
3396         smp_mb();
3397         reserved = space_info->bytes_may_use;
3398         progress = space_info->reservation_progress;
3399
3400         if (reserved == 0)
3401                 return 0;
3402
3403         smp_mb();
3404         if (root->fs_info->delalloc_bytes == 0) {
3405                 if (trans)
3406                         return 0;
3407                 btrfs_wait_ordered_extents(root, 0, 0);
3408                 return 0;
3409         }
3410
3411         max_reclaim = min(reserved, to_reclaim);
3412         nr_pages = max_t(unsigned long, nr_pages,
3413                          max_reclaim >> PAGE_CACHE_SHIFT);
3414         while (loops < 1024) {
3415                 /* have the flusher threads jump in and do some IO */
3416                 smp_mb();
3417                 nr_pages = min_t(unsigned long, nr_pages,
3418                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3419                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3420                                                 WB_REASON_FS_FREE_SPACE);
3421
3422                 spin_lock(&space_info->lock);
3423                 if (reserved > space_info->bytes_may_use)
3424                         reclaimed += reserved - space_info->bytes_may_use;
3425                 reserved = space_info->bytes_may_use;
3426                 spin_unlock(&space_info->lock);
3427
3428                 loops++;
3429
3430                 if (reserved == 0 || reclaimed >= max_reclaim)
3431                         break;
3432
3433                 if (trans && trans->transaction->blocked)
3434                         return -EAGAIN;
3435
3436                 if (wait_ordered && !trans) {
3437                         btrfs_wait_ordered_extents(root, 0, 0);
3438                 } else {
3439                         time_left = schedule_timeout_interruptible(1);
3440
3441                         /* We were interrupted, exit */
3442                         if (time_left)
3443                                 break;
3444                 }
3445
3446                 /* we've kicked the IO a few times, if anything has been freed,
3447                  * exit.  There is no sense in looping here for a long time
3448                  * when we really need to commit the transaction, or there are
3449                  * just too many writers without enough free space
3450                  */
3451
3452                 if (loops > 3) {
3453                         smp_mb();
3454                         if (progress != space_info->reservation_progress)
3455                                 break;
3456                 }
3457
3458         }
3459
3460         return reclaimed >= to_reclaim;
3461 }
3462
3463 /**
3464  * maybe_commit_transaction - possibly commit the transaction if its ok to
3465  * @root - the root we're allocating for
3466  * @bytes - the number of bytes we want to reserve
3467  * @force - force the commit
3468  *
3469  * This will check to make sure that committing the transaction will actually
3470  * get us somewhere and then commit the transaction if it does.  Otherwise it
3471  * will return -ENOSPC.
3472  */
3473 static int may_commit_transaction(struct btrfs_root *root,
3474                                   struct btrfs_space_info *space_info,
3475                                   u64 bytes, int force)
3476 {
3477         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3478         struct btrfs_trans_handle *trans;
3479
3480         trans = (struct btrfs_trans_handle *)current->journal_info;
3481         if (trans)
3482                 return -EAGAIN;
3483
3484         if (force)
3485                 goto commit;
3486
3487         /* See if there is enough pinned space to make this reservation */
3488         spin_lock(&space_info->lock);
3489         if (space_info->bytes_pinned >= bytes) {
3490                 spin_unlock(&space_info->lock);
3491                 goto commit;
3492         }
3493         spin_unlock(&space_info->lock);
3494
3495         /*
3496          * See if there is some space in the delayed insertion reservation for
3497          * this reservation.
3498          */
3499         if (space_info != delayed_rsv->space_info)
3500                 return -ENOSPC;
3501
3502         spin_lock(&delayed_rsv->lock);
3503         if (delayed_rsv->size < bytes) {
3504                 spin_unlock(&delayed_rsv->lock);
3505                 return -ENOSPC;
3506         }
3507         spin_unlock(&delayed_rsv->lock);
3508
3509 commit:
3510         trans = btrfs_join_transaction(root);
3511         if (IS_ERR(trans))
3512                 return -ENOSPC;
3513
3514         return btrfs_commit_transaction(trans, root);
3515 }
3516
3517 /**
3518  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3519  * @root - the root we're allocating for
3520  * @block_rsv - the block_rsv we're allocating for
3521  * @orig_bytes - the number of bytes we want
3522  * @flush - wether or not we can flush to make our reservation
3523  *
3524  * This will reserve orgi_bytes number of bytes from the space info associated
3525  * with the block_rsv.  If there is not enough space it will make an attempt to
3526  * flush out space to make room.  It will do this by flushing delalloc if
3527  * possible or committing the transaction.  If flush is 0 then no attempts to
3528  * regain reservations will be made and this will fail if there is not enough
3529  * space already.
3530  */
3531 static int reserve_metadata_bytes(struct btrfs_root *root,
3532                                   struct btrfs_block_rsv *block_rsv,
3533                                   u64 orig_bytes, int flush)
3534 {
3535         struct btrfs_space_info *space_info = block_rsv->space_info;
3536         u64 used;
3537         u64 num_bytes = orig_bytes;
3538         int retries = 0;
3539         int ret = 0;
3540         bool committed = false;
3541         bool flushing = false;
3542         bool wait_ordered = false;
3543
3544 again:
3545         ret = 0;
3546         spin_lock(&space_info->lock);
3547         /*
3548          * We only want to wait if somebody other than us is flushing and we are
3549          * actually alloed to flush.
3550          */
3551         while (flush && !flushing && space_info->flush) {
3552                 spin_unlock(&space_info->lock);
3553                 /*
3554                  * If we have a trans handle we can't wait because the flusher
3555                  * may have to commit the transaction, which would mean we would
3556                  * deadlock since we are waiting for the flusher to finish, but
3557                  * hold the current transaction open.
3558                  */
3559                 if (current->journal_info)
3560                         return -EAGAIN;
3561                 ret = wait_event_interruptible(space_info->wait,
3562                                                !space_info->flush);
3563                 /* Must have been interrupted, return */
3564                 if (ret)
3565                         return -EINTR;
3566
3567                 spin_lock(&space_info->lock);
3568         }
3569
3570         ret = -ENOSPC;
3571         used = space_info->bytes_used + space_info->bytes_reserved +
3572                 space_info->bytes_pinned + space_info->bytes_readonly +
3573                 space_info->bytes_may_use;
3574
3575         /*
3576          * The idea here is that we've not already over-reserved the block group
3577          * then we can go ahead and save our reservation first and then start
3578          * flushing if we need to.  Otherwise if we've already overcommitted
3579          * lets start flushing stuff first and then come back and try to make
3580          * our reservation.
3581          */
3582         if (used <= space_info->total_bytes) {
3583                 if (used + orig_bytes <= space_info->total_bytes) {
3584                         space_info->bytes_may_use += orig_bytes;
3585                         ret = 0;
3586                 } else {
3587                         /*
3588                          * Ok set num_bytes to orig_bytes since we aren't
3589                          * overocmmitted, this way we only try and reclaim what
3590                          * we need.
3591                          */
3592                         num_bytes = orig_bytes;
3593                 }
3594         } else {
3595                 /*
3596                  * Ok we're over committed, set num_bytes to the overcommitted
3597                  * amount plus the amount of bytes that we need for this
3598                  * reservation.
3599                  */
3600                 wait_ordered = true;
3601                 num_bytes = used - space_info->total_bytes +
3602                         (orig_bytes * (retries + 1));
3603         }
3604
3605         if (ret) {
3606                 u64 profile = btrfs_get_alloc_profile(root, 0);
3607                 u64 avail;
3608
3609                 /*
3610                  * If we have a lot of space that's pinned, don't bother doing
3611                  * the overcommit dance yet and just commit the transaction.
3612                  */
3613                 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3614                 do_div(avail, 10);
3615                 if (space_info->bytes_pinned >= avail && flush && !committed) {
3616                         space_info->flush = 1;
3617                         flushing = true;
3618                         spin_unlock(&space_info->lock);
3619                         ret = may_commit_transaction(root, space_info,
3620                                                      orig_bytes, 1);
3621                         if (ret)
3622                                 goto out;
3623                         committed = true;
3624                         goto again;
3625                 }
3626
3627                 spin_lock(&root->fs_info->free_chunk_lock);
3628                 avail = root->fs_info->free_chunk_space;
3629
3630                 /*
3631                  * If we have dup, raid1 or raid10 then only half of the free
3632                  * space is actually useable.
3633                  */
3634                 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3635                                BTRFS_BLOCK_GROUP_RAID1 |
3636                                BTRFS_BLOCK_GROUP_RAID10))
3637                         avail >>= 1;
3638
3639                 /*
3640                  * If we aren't flushing don't let us overcommit too much, say
3641                  * 1/8th of the space.  If we can flush, let it overcommit up to
3642                  * 1/2 of the space.
3643                  */
3644                 if (flush)
3645                         avail >>= 3;
3646                 else
3647                         avail >>= 1;
3648                  spin_unlock(&root->fs_info->free_chunk_lock);
3649
3650                 if (used + num_bytes < space_info->total_bytes + avail) {
3651                         space_info->bytes_may_use += orig_bytes;
3652                         ret = 0;
3653                 } else {
3654                         wait_ordered = true;
3655                 }
3656         }
3657
3658         /*
3659          * Couldn't make our reservation, save our place so while we're trying
3660          * to reclaim space we can actually use it instead of somebody else
3661          * stealing it from us.
3662          */
3663         if (ret && flush) {
3664                 flushing = true;
3665                 space_info->flush = 1;
3666         }
3667
3668         spin_unlock(&space_info->lock);
3669
3670         if (!ret || !flush)
3671                 goto out;
3672
3673         /*
3674          * We do synchronous shrinking since we don't actually unreserve
3675          * metadata until after the IO is completed.
3676          */
3677         ret = shrink_delalloc(root, num_bytes, wait_ordered);
3678         if (ret < 0)
3679                 goto out;
3680
3681         ret = 0;
3682
3683         /*
3684          * So if we were overcommitted it's possible that somebody else flushed
3685          * out enough space and we simply didn't have enough space to reclaim,
3686          * so go back around and try again.
3687          */
3688         if (retries < 2) {
3689                 wait_ordered = true;
3690                 retries++;
3691                 goto again;
3692         }
3693
3694         ret = -ENOSPC;
3695         if (committed)
3696                 goto out;
3697
3698         ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3699         if (!ret) {
3700                 committed = true;
3701                 goto again;
3702         }
3703
3704 out:
3705         if (flushing) {
3706                 spin_lock(&space_info->lock);
3707                 space_info->flush = 0;
3708                 wake_up_all(&space_info->wait);
3709                 spin_unlock(&space_info->lock);
3710         }
3711         return ret;
3712 }
3713
3714 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3715                                              struct btrfs_root *root)
3716 {
3717         struct btrfs_block_rsv *block_rsv = NULL;
3718
3719         if (root->ref_cows || root == root->fs_info->csum_root)
3720                 block_rsv = trans->block_rsv;
3721
3722         if (!block_rsv)
3723                 block_rsv = root->block_rsv;
3724
3725         if (!block_rsv)
3726                 block_rsv = &root->fs_info->empty_block_rsv;
3727
3728         return block_rsv;
3729 }
3730
3731 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3732                                u64 num_bytes)
3733 {
3734         int ret = -ENOSPC;
3735         spin_lock(&block_rsv->lock);
3736         if (block_rsv->reserved >= num_bytes) {
3737                 block_rsv->reserved -= num_bytes;
3738                 if (block_rsv->reserved < block_rsv->size)
3739                         block_rsv->full = 0;
3740                 ret = 0;
3741         }
3742         spin_unlock(&block_rsv->lock);
3743         return ret;
3744 }
3745
3746 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3747                                 u64 num_bytes, int update_size)
3748 {
3749         spin_lock(&block_rsv->lock);
3750         block_rsv->reserved += num_bytes;
3751         if (update_size)
3752                 block_rsv->size += num_bytes;
3753         else if (block_rsv->reserved >= block_rsv->size)
3754                 block_rsv->full = 1;
3755         spin_unlock(&block_rsv->lock);
3756 }
3757
3758 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3759                                     struct btrfs_block_rsv *dest, u64 num_bytes)
3760 {
3761         struct btrfs_space_info *space_info = block_rsv->space_info;
3762
3763         spin_lock(&block_rsv->lock);
3764         if (num_bytes == (u64)-1)
3765                 num_bytes = block_rsv->size;
3766         block_rsv->size -= num_bytes;
3767         if (block_rsv->reserved >= block_rsv->size) {
3768                 num_bytes = block_rsv->reserved - block_rsv->size;
3769                 block_rsv->reserved = block_rsv->size;
3770                 block_rsv->full = 1;
3771         } else {
3772                 num_bytes = 0;
3773         }
3774         spin_unlock(&block_rsv->lock);
3775
3776         if (num_bytes > 0) {
3777                 if (dest) {
3778                         spin_lock(&dest->lock);
3779                         if (!dest->full) {
3780                                 u64 bytes_to_add;
3781
3782                                 bytes_to_add = dest->size - dest->reserved;
3783                                 bytes_to_add = min(num_bytes, bytes_to_add);
3784                                 dest->reserved += bytes_to_add;
3785                                 if (dest->reserved >= dest->size)
3786                                         dest->full = 1;
3787                                 num_bytes -= bytes_to_add;
3788                         }
3789                         spin_unlock(&dest->lock);
3790                 }
3791                 if (num_bytes) {
3792                         spin_lock(&space_info->lock);
3793                         space_info->bytes_may_use -= num_bytes;
3794                         space_info->reservation_progress++;
3795                         spin_unlock(&space_info->lock);
3796                 }
3797         }
3798 }
3799
3800 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3801                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3802 {
3803         int ret;
3804
3805         ret = block_rsv_use_bytes(src, num_bytes);
3806         if (ret)
3807                 return ret;
3808
3809         block_rsv_add_bytes(dst, num_bytes, 1);
3810         return 0;
3811 }
3812
3813 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3814 {
3815         memset(rsv, 0, sizeof(*rsv));
3816         spin_lock_init(&rsv->lock);
3817 }
3818
3819 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3820 {
3821         struct btrfs_block_rsv *block_rsv;
3822         struct btrfs_fs_info *fs_info = root->fs_info;
3823
3824         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3825         if (!block_rsv)
3826                 return NULL;
3827
3828         btrfs_init_block_rsv(block_rsv);
3829         block_rsv->space_info = __find_space_info(fs_info,
3830                                                   BTRFS_BLOCK_GROUP_METADATA);
3831         return block_rsv;
3832 }
3833
3834 void btrfs_free_block_rsv(struct btrfs_root *root,
3835                           struct btrfs_block_rsv *rsv)
3836 {
3837         btrfs_block_rsv_release(root, rsv, (u64)-1);
3838         kfree(rsv);
3839 }
3840
3841 static inline int __block_rsv_add(struct btrfs_root *root,
3842                                   struct btrfs_block_rsv *block_rsv,
3843                                   u64 num_bytes, int flush)
3844 {
3845         int ret;
3846
3847         if (num_bytes == 0)
3848                 return 0;
3849
3850         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3851         if (!ret) {
3852                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3853                 return 0;
3854         }
3855
3856         return ret;
3857 }
3858
3859 int btrfs_block_rsv_add(struct btrfs_root *root,
3860                         struct btrfs_block_rsv *block_rsv,
3861                         u64 num_bytes)
3862 {
3863         return __block_rsv_add(root, block_rsv, num_bytes, 1);
3864 }
3865
3866 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3867                                 struct btrfs_block_rsv *block_rsv,
3868                                 u64 num_bytes)
3869 {
3870         return __block_rsv_add(root, block_rsv, num_bytes, 0);
3871 }
3872
3873 int btrfs_block_rsv_check(struct btrfs_root *root,
3874                           struct btrfs_block_rsv *block_rsv, int min_factor)
3875 {
3876         u64 num_bytes = 0;
3877         int ret = -ENOSPC;
3878
3879         if (!block_rsv)
3880                 return 0;
3881
3882         spin_lock(&block_rsv->lock);
3883         num_bytes = div_factor(block_rsv->size, min_factor);
3884         if (block_rsv->reserved >= num_bytes)
3885                 ret = 0;
3886         spin_unlock(&block_rsv->lock);
3887
3888         return ret;
3889 }
3890
3891 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
3892                                            struct btrfs_block_rsv *block_rsv,
3893                                            u64 min_reserved, int flush)
3894 {
3895         u64 num_bytes = 0;
3896         int ret = -ENOSPC;
3897
3898         if (!block_rsv)
3899                 return 0;
3900
3901         spin_lock(&block_rsv->lock);
3902         num_bytes = min_reserved;
3903         if (block_rsv->reserved >= num_bytes)
3904                 ret = 0;
3905         else
3906                 num_bytes -= block_rsv->reserved;
3907         spin_unlock(&block_rsv->lock);
3908
3909         if (!ret)
3910                 return 0;
3911
3912         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3913         if (!ret) {
3914                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3915                 return 0;
3916         }
3917
3918         return ret;
3919 }
3920
3921 int btrfs_block_rsv_refill(struct btrfs_root *root,
3922                            struct btrfs_block_rsv *block_rsv,
3923                            u64 min_reserved)
3924 {
3925         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
3926 }
3927
3928 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
3929                                    struct btrfs_block_rsv *block_rsv,
3930                                    u64 min_reserved)
3931 {
3932         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
3933 }
3934
3935 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3936                             struct btrfs_block_rsv *dst_rsv,
3937                             u64 num_bytes)
3938 {
3939         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3940 }
3941
3942 void btrfs_block_rsv_release(struct btrfs_root *root,
3943                              struct btrfs_block_rsv *block_rsv,
3944                              u64 num_bytes)
3945 {
3946         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3947         if (global_rsv->full || global_rsv == block_rsv ||
3948             block_rsv->space_info != global_rsv->space_info)
3949                 global_rsv = NULL;
3950         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3951 }
3952
3953 /*
3954  * helper to calculate size of global block reservation.
3955  * the desired value is sum of space used by extent tree,
3956  * checksum tree and root tree
3957  */
3958 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3959 {
3960         struct btrfs_space_info *sinfo;
3961         u64 num_bytes;
3962         u64 meta_used;
3963         u64 data_used;
3964         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
3965
3966         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3967         spin_lock(&sinfo->lock);
3968         data_used = sinfo->bytes_used;
3969         spin_unlock(&sinfo->lock);
3970
3971         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3972         spin_lock(&sinfo->lock);
3973         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3974                 data_used = 0;
3975         meta_used = sinfo->bytes_used;
3976         spin_unlock(&sinfo->lock);
3977
3978         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3979                     csum_size * 2;
3980         num_bytes += div64_u64(data_used + meta_used, 50);
3981
3982         if (num_bytes * 3 > meta_used)
3983                 num_bytes = div64_u64(meta_used, 3);
3984
3985         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3986 }
3987
3988 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3989 {
3990         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3991         struct btrfs_space_info *sinfo = block_rsv->space_info;
3992         u64 num_bytes;
3993
3994         num_bytes = calc_global_metadata_size(fs_info);
3995
3996         spin_lock(&block_rsv->lock);
3997         spin_lock(&sinfo->lock);
3998
3999         block_rsv->size = num_bytes;
4000
4001         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4002                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4003                     sinfo->bytes_may_use;
4004
4005         if (sinfo->total_bytes > num_bytes) {
4006                 num_bytes = sinfo->total_bytes - num_bytes;
4007                 block_rsv->reserved += num_bytes;
4008                 sinfo->bytes_may_use += num_bytes;
4009         }
4010
4011         if (block_rsv->reserved >= block_rsv->size) {
4012                 num_bytes = block_rsv->reserved - block_rsv->size;
4013                 sinfo->bytes_may_use -= num_bytes;
4014                 sinfo->reservation_progress++;
4015                 block_rsv->reserved = block_rsv->size;
4016                 block_rsv->full = 1;
4017         }
4018
4019         spin_unlock(&sinfo->lock);
4020         spin_unlock(&block_rsv->lock);
4021 }
4022
4023 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4024 {
4025         struct btrfs_space_info *space_info;
4026
4027         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4028         fs_info->chunk_block_rsv.space_info = space_info;
4029
4030         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4031         fs_info->global_block_rsv.space_info = space_info;
4032         fs_info->delalloc_block_rsv.space_info = space_info;
4033         fs_info->trans_block_rsv.space_info = space_info;
4034         fs_info->empty_block_rsv.space_info = space_info;
4035         fs_info->delayed_block_rsv.space_info = space_info;
4036
4037         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4038         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4039         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4040         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4041         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4042
4043         update_global_block_rsv(fs_info);
4044 }
4045
4046 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4047 {
4048         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
4049         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4050         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4051         WARN_ON(fs_info->trans_block_rsv.size > 0);
4052         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4053         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4054         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4055         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4056         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4057 }
4058
4059 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4060                                   struct btrfs_root *root)
4061 {
4062         if (!trans->bytes_reserved)
4063                 return;
4064
4065         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4066         trans->bytes_reserved = 0;
4067 }
4068
4069 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4070                                   struct inode *inode)
4071 {
4072         struct btrfs_root *root = BTRFS_I(inode)->root;
4073         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4074         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4075
4076         /*
4077          * We need to hold space in order to delete our orphan item once we've
4078          * added it, so this takes the reservation so we can release it later
4079          * when we are truly done with the orphan item.
4080          */
4081         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4082         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4083 }
4084
4085 void btrfs_orphan_release_metadata(struct inode *inode)
4086 {
4087         struct btrfs_root *root = BTRFS_I(inode)->root;
4088         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4089         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4090 }
4091
4092 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4093                                 struct btrfs_pending_snapshot *pending)
4094 {
4095         struct btrfs_root *root = pending->root;
4096         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4097         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4098         /*
4099          * two for root back/forward refs, two for directory entries
4100          * and one for root of the snapshot.
4101          */
4102         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4103         dst_rsv->space_info = src_rsv->space_info;
4104         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4105 }
4106
4107 /**
4108  * drop_outstanding_extent - drop an outstanding extent
4109  * @inode: the inode we're dropping the extent for
4110  *
4111  * This is called when we are freeing up an outstanding extent, either called
4112  * after an error or after an extent is written.  This will return the number of
4113  * reserved extents that need to be freed.  This must be called with
4114  * BTRFS_I(inode)->lock held.
4115  */
4116 static unsigned drop_outstanding_extent(struct inode *inode)
4117 {
4118         unsigned drop_inode_space = 0;
4119         unsigned dropped_extents = 0;
4120
4121         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4122         BTRFS_I(inode)->outstanding_extents--;
4123
4124         if (BTRFS_I(inode)->outstanding_extents == 0 &&
4125             BTRFS_I(inode)->delalloc_meta_reserved) {
4126                 drop_inode_space = 1;
4127                 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4128         }
4129
4130         /*
4131          * If we have more or the same amount of outsanding extents than we have
4132          * reserved then we need to leave the reserved extents count alone.
4133          */
4134         if (BTRFS_I(inode)->outstanding_extents >=
4135             BTRFS_I(inode)->reserved_extents)
4136                 return drop_inode_space;
4137
4138         dropped_extents = BTRFS_I(inode)->reserved_extents -
4139                 BTRFS_I(inode)->outstanding_extents;
4140         BTRFS_I(inode)->reserved_extents -= dropped_extents;
4141         return dropped_extents + drop_inode_space;
4142 }
4143
4144 /**
4145  * calc_csum_metadata_size - return the amount of metada space that must be
4146  *      reserved/free'd for the given bytes.
4147  * @inode: the inode we're manipulating
4148  * @num_bytes: the number of bytes in question
4149  * @reserve: 1 if we are reserving space, 0 if we are freeing space
4150  *
4151  * This adjusts the number of csum_bytes in the inode and then returns the
4152  * correct amount of metadata that must either be reserved or freed.  We
4153  * calculate how many checksums we can fit into one leaf and then divide the
4154  * number of bytes that will need to be checksumed by this value to figure out
4155  * how many checksums will be required.  If we are adding bytes then the number
4156  * may go up and we will return the number of additional bytes that must be
4157  * reserved.  If it is going down we will return the number of bytes that must
4158  * be freed.
4159  *
4160  * This must be called with BTRFS_I(inode)->lock held.
4161  */
4162 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4163                                    int reserve)
4164 {
4165         struct btrfs_root *root = BTRFS_I(inode)->root;
4166         u64 csum_size;
4167         int num_csums_per_leaf;
4168         int num_csums;
4169         int old_csums;
4170
4171         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4172             BTRFS_I(inode)->csum_bytes == 0)
4173                 return 0;
4174
4175         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4176         if (reserve)
4177                 BTRFS_I(inode)->csum_bytes += num_bytes;
4178         else
4179                 BTRFS_I(inode)->csum_bytes -= num_bytes;
4180         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4181         num_csums_per_leaf = (int)div64_u64(csum_size,
4182                                             sizeof(struct btrfs_csum_item) +
4183                                             sizeof(struct btrfs_disk_key));
4184         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4185         num_csums = num_csums + num_csums_per_leaf - 1;
4186         num_csums = num_csums / num_csums_per_leaf;
4187
4188         old_csums = old_csums + num_csums_per_leaf - 1;
4189         old_csums = old_csums / num_csums_per_leaf;
4190
4191         /* No change, no need to reserve more */
4192         if (old_csums == num_csums)
4193                 return 0;
4194
4195         if (reserve)
4196                 return btrfs_calc_trans_metadata_size(root,
4197                                                       num_csums - old_csums);
4198
4199         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4200 }
4201
4202 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4203 {
4204         struct btrfs_root *root = BTRFS_I(inode)->root;
4205         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4206         u64 to_reserve = 0;
4207         unsigned nr_extents = 0;
4208         int flush = 1;
4209         int ret;
4210
4211         if (btrfs_is_free_space_inode(root, inode))
4212                 flush = 0;
4213
4214         if (flush && btrfs_transaction_in_commit(root->fs_info))
4215                 schedule_timeout(1);
4216
4217         num_bytes = ALIGN(num_bytes, root->sectorsize);
4218
4219         spin_lock(&BTRFS_I(inode)->lock);
4220         BTRFS_I(inode)->outstanding_extents++;
4221
4222         if (BTRFS_I(inode)->outstanding_extents >
4223             BTRFS_I(inode)->reserved_extents) {
4224                 nr_extents = BTRFS_I(inode)->outstanding_extents -
4225                         BTRFS_I(inode)->reserved_extents;
4226                 BTRFS_I(inode)->reserved_extents += nr_extents;
4227         }
4228
4229         /*
4230          * Add an item to reserve for updating the inode when we complete the
4231          * delalloc io.
4232          */
4233         if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4234                 nr_extents++;
4235                 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4236         }
4237
4238         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4239         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4240         spin_unlock(&BTRFS_I(inode)->lock);
4241
4242         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4243         if (ret) {
4244                 u64 to_free = 0;
4245                 unsigned dropped;
4246
4247                 spin_lock(&BTRFS_I(inode)->lock);
4248                 dropped = drop_outstanding_extent(inode);
4249                 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4250                 spin_unlock(&BTRFS_I(inode)->lock);
4251                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4252
4253                 /*
4254                  * Somebody could have come in and twiddled with the
4255                  * reservation, so if we have to free more than we would have
4256                  * reserved from this reservation go ahead and release those
4257                  * bytes.
4258                  */
4259                 to_free -= to_reserve;
4260                 if (to_free)
4261                         btrfs_block_rsv_release(root, block_rsv, to_free);
4262                 return ret;
4263         }
4264
4265         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4266
4267         return 0;
4268 }
4269
4270 /**
4271  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4272  * @inode: the inode to release the reservation for
4273  * @num_bytes: the number of bytes we're releasing
4274  *
4275  * This will release the metadata reservation for an inode.  This can be called
4276  * once we complete IO for a given set of bytes to release their metadata
4277  * reservations.
4278  */
4279 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4280 {
4281         struct btrfs_root *root = BTRFS_I(inode)->root;
4282         u64 to_free = 0;
4283         unsigned dropped;
4284
4285         num_bytes = ALIGN(num_bytes, root->sectorsize);
4286         spin_lock(&BTRFS_I(inode)->lock);
4287         dropped = drop_outstanding_extent(inode);
4288
4289         to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4290         spin_unlock(&BTRFS_I(inode)->lock);
4291         if (dropped > 0)
4292                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4293
4294         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4295                                 to_free);
4296 }
4297
4298 /**
4299  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4300  * @inode: inode we're writing to
4301  * @num_bytes: the number of bytes we want to allocate
4302  *
4303  * This will do the following things
4304  *
4305  * o reserve space in the data space info for num_bytes
4306  * o reserve space in the metadata space info based on number of outstanding
4307  *   extents and how much csums will be needed
4308  * o add to the inodes ->delalloc_bytes
4309  * o add it to the fs_info's delalloc inodes list.
4310  *
4311  * This will return 0 for success and -ENOSPC if there is no space left.
4312  */
4313 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4314 {
4315         int ret;
4316
4317         ret = btrfs_check_data_free_space(inode, num_bytes);
4318         if (ret)
4319                 return ret;
4320
4321         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4322         if (ret) {
4323                 btrfs_free_reserved_data_space(inode, num_bytes);
4324                 return ret;
4325         }
4326
4327         return 0;
4328 }
4329
4330 /**
4331  * btrfs_delalloc_release_space - release data and metadata space for delalloc
4332  * @inode: inode we're releasing space for
4333  * @num_bytes: the number of bytes we want to free up
4334  *
4335  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
4336  * called in the case that we don't need the metadata AND data reservations
4337  * anymore.  So if there is an error or we insert an inline extent.
4338  *
4339  * This function will release the metadata space that was not used and will
4340  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4341  * list if there are no delalloc bytes left.
4342  */
4343 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4344 {
4345         btrfs_delalloc_release_metadata(inode, num_bytes);
4346         btrfs_free_reserved_data_space(inode, num_bytes);
4347 }
4348
4349 static int update_block_group(struct btrfs_trans_handle *trans,
4350                               struct btrfs_root *root,
4351                               u64 bytenr, u64 num_bytes, int alloc)
4352 {
4353         struct btrfs_block_group_cache *cache = NULL;
4354         struct btrfs_fs_info *info = root->fs_info;
4355         u64 total = num_bytes;
4356         u64 old_val;
4357         u64 byte_in_group;
4358         int factor;
4359
4360         /* block accounting for super block */
4361         spin_lock(&info->delalloc_lock);
4362         old_val = btrfs_super_bytes_used(info->super_copy);
4363         if (alloc)
4364                 old_val += num_bytes;
4365         else
4366                 old_val -= num_bytes;
4367         btrfs_set_super_bytes_used(info->super_copy, old_val);
4368         spin_unlock(&info->delalloc_lock);
4369
4370         while (total) {
4371                 cache = btrfs_lookup_block_group(info, bytenr);
4372                 if (!cache)
4373                         return -1;
4374                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4375                                     BTRFS_BLOCK_GROUP_RAID1 |
4376                                     BTRFS_BLOCK_GROUP_RAID10))
4377                         factor = 2;
4378                 else
4379                         factor = 1;
4380                 /*
4381                  * If this block group has free space cache written out, we
4382                  * need to make sure to load it if we are removing space.  This
4383                  * is because we need the unpinning stage to actually add the
4384                  * space back to the block group, otherwise we will leak space.
4385                  */
4386                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4387                         cache_block_group(cache, trans, NULL, 1);
4388
4389                 byte_in_group = bytenr - cache->key.objectid;
4390                 WARN_ON(byte_in_group > cache->key.offset);
4391
4392                 spin_lock(&cache->space_info->lock);
4393                 spin_lock(&cache->lock);
4394
4395                 if (btrfs_test_opt(root, SPACE_CACHE) &&
4396                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4397                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4398
4399                 cache->dirty = 1;
4400                 old_val = btrfs_block_group_used(&cache->item);
4401                 num_bytes = min(total, cache->key.offset - byte_in_group);
4402                 if (alloc) {
4403                         old_val += num_bytes;
4404                         btrfs_set_block_group_used(&cache->item, old_val);
4405                         cache->reserved -= num_bytes;
4406                         cache->space_info->bytes_reserved -= num_bytes;
4407                         cache->space_info->bytes_used += num_bytes;
4408                         cache->space_info->disk_used += num_bytes * factor;
4409                         spin_unlock(&cache->lock);
4410                         spin_unlock(&cache->space_info->lock);
4411                 } else {
4412                         old_val -= num_bytes;
4413                         btrfs_set_block_group_used(&cache->item, old_val);
4414                         cache->pinned += num_bytes;
4415                         cache->space_info->bytes_pinned += num_bytes;
4416                         cache->space_info->bytes_used -= num_bytes;
4417                         cache->space_info->disk_used -= num_bytes * factor;
4418                         spin_unlock(&cache->lock);
4419                         spin_unlock(&cache->space_info->lock);
4420
4421                         set_extent_dirty(info->pinned_extents,
4422                                          bytenr, bytenr + num_bytes - 1,
4423                                          GFP_NOFS | __GFP_NOFAIL);
4424                 }
4425                 btrfs_put_block_group(cache);
4426                 total -= num_bytes;
4427                 bytenr += num_bytes;
4428         }
4429         return 0;
4430 }
4431
4432 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4433 {
4434         struct btrfs_block_group_cache *cache;
4435         u64 bytenr;
4436
4437         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4438         if (!cache)
4439                 return 0;
4440
4441         bytenr = cache->key.objectid;
4442         btrfs_put_block_group(cache);
4443
4444         return bytenr;
4445 }
4446
4447 static int pin_down_extent(struct btrfs_root *root,
4448                            struct btrfs_block_group_cache *cache,
4449                            u64 bytenr, u64 num_bytes, int reserved)
4450 {
4451         spin_lock(&cache->space_info->lock);
4452         spin_lock(&cache->lock);
4453         cache->pinned += num_bytes;
4454         cache->space_info->bytes_pinned += num_bytes;
4455         if (reserved) {
4456                 cache->reserved -= num_bytes;
4457                 cache->space_info->bytes_reserved -= num_bytes;
4458         }
4459         spin_unlock(&cache->lock);
4460         spin_unlock(&cache->space_info->lock);
4461
4462         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4463                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4464         return 0;
4465 }
4466
4467 /*
4468  * this function must be called within transaction
4469  */
4470 int btrfs_pin_extent(struct btrfs_root *root,
4471                      u64 bytenr, u64 num_bytes, int reserved)
4472 {
4473         struct btrfs_block_group_cache *cache;
4474
4475         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4476         BUG_ON(!cache);
4477
4478         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4479
4480         btrfs_put_block_group(cache);
4481         return 0;
4482 }
4483
4484 /*
4485  * this function must be called within transaction
4486  */
4487 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4488                                     struct btrfs_root *root,
4489                                     u64 bytenr, u64 num_bytes)
4490 {
4491         struct btrfs_block_group_cache *cache;
4492
4493         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4494         BUG_ON(!cache);
4495
4496         /*
4497          * pull in the free space cache (if any) so that our pin
4498          * removes the free space from the cache.  We have load_only set
4499          * to one because the slow code to read in the free extents does check
4500          * the pinned extents.
4501          */
4502         cache_block_group(cache, trans, root, 1);
4503
4504         pin_down_extent(root, cache, bytenr, num_bytes, 0);
4505
4506         /* remove us from the free space cache (if we're there at all) */
4507         btrfs_remove_free_space(cache, bytenr, num_bytes);
4508         btrfs_put_block_group(cache);
4509         return 0;
4510 }
4511
4512 /**
4513  * btrfs_update_reserved_bytes - update the block_group and space info counters
4514  * @cache:      The cache we are manipulating
4515  * @num_bytes:  The number of bytes in question
4516  * @reserve:    One of the reservation enums
4517  *
4518  * This is called by the allocator when it reserves space, or by somebody who is
4519  * freeing space that was never actually used on disk.  For example if you
4520  * reserve some space for a new leaf in transaction A and before transaction A
4521  * commits you free that leaf, you call this with reserve set to 0 in order to
4522  * clear the reservation.
4523  *
4524  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4525  * ENOSPC accounting.  For data we handle the reservation through clearing the
4526  * delalloc bits in the io_tree.  We have to do this since we could end up
4527  * allocating less disk space for the amount of data we have reserved in the
4528  * case of compression.
4529  *
4530  * If this is a reservation and the block group has become read only we cannot
4531  * make the reservation and return -EAGAIN, otherwise this function always
4532  * succeeds.
4533  */
4534 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4535                                        u64 num_bytes, int reserve)
4536 {
4537         struct btrfs_space_info *space_info = cache->space_info;
4538         int ret = 0;
4539         spin_lock(&space_info->lock);
4540         spin_lock(&cache->lock);
4541         if (reserve != RESERVE_FREE) {
4542                 if (cache->ro) {
4543                         ret = -EAGAIN;
4544                 } else {
4545                         cache->reserved += num_bytes;
4546                         space_info->bytes_reserved += num_bytes;
4547                         if (reserve == RESERVE_ALLOC) {
4548                                 BUG_ON(space_info->bytes_may_use < num_bytes);
4549                                 space_info->bytes_may_use -= num_bytes;
4550                         }
4551                 }
4552         } else {
4553                 if (cache->ro)
4554                         space_info->bytes_readonly += num_bytes;
4555                 cache->reserved -= num_bytes;
4556                 space_info->bytes_reserved -= num_bytes;
4557                 space_info->reservation_progress++;
4558         }
4559         spin_unlock(&cache->lock);
4560         spin_unlock(&space_info->lock);
4561         return ret;
4562 }
4563
4564 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4565                                 struct btrfs_root *root)
4566 {
4567         struct btrfs_fs_info *fs_info = root->fs_info;
4568         struct btrfs_caching_control *next;
4569         struct btrfs_caching_control *caching_ctl;
4570         struct btrfs_block_group_cache *cache;
4571
4572         down_write(&fs_info->extent_commit_sem);
4573
4574         list_for_each_entry_safe(caching_ctl, next,
4575                                  &fs_info->caching_block_groups, list) {
4576                 cache = caching_ctl->block_group;
4577                 if (block_group_cache_done(cache)) {
4578                         cache->last_byte_to_unpin = (u64)-1;
4579                         list_del_init(&caching_ctl->list);
4580                         put_caching_control(caching_ctl);
4581                 } else {
4582                         cache->last_byte_to_unpin = caching_ctl->progress;
4583                 }
4584         }
4585
4586         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4587                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4588         else
4589                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4590
4591         up_write(&fs_info->extent_commit_sem);
4592
4593         update_global_block_rsv(fs_info);
4594         return 0;
4595 }
4596
4597 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4598 {
4599         struct btrfs_fs_info *fs_info = root->fs_info;
4600         struct btrfs_block_group_cache *cache = NULL;
4601         u64 len;
4602
4603         while (start <= end) {
4604                 if (!cache ||
4605                     start >= cache->key.objectid + cache->key.offset) {
4606                         if (cache)
4607                                 btrfs_put_block_group(cache);
4608                         cache = btrfs_lookup_block_group(fs_info, start);
4609                         BUG_ON(!cache);
4610                 }
4611
4612                 len = cache->key.objectid + cache->key.offset - start;
4613                 len = min(len, end + 1 - start);
4614
4615                 if (start < cache->last_byte_to_unpin) {
4616                         len = min(len, cache->last_byte_to_unpin - start);
4617                         btrfs_add_free_space(cache, start, len);
4618                 }
4619
4620                 start += len;
4621
4622                 spin_lock(&cache->space_info->lock);
4623                 spin_lock(&cache->lock);
4624                 cache->pinned -= len;
4625                 cache->space_info->bytes_pinned -= len;
4626                 if (cache->ro)
4627                         cache->space_info->bytes_readonly += len;
4628                 spin_unlock(&cache->lock);
4629                 spin_unlock(&cache->space_info->lock);
4630         }
4631
4632         if (cache)
4633                 btrfs_put_block_group(cache);
4634         return 0;
4635 }
4636
4637 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4638                                struct btrfs_root *root)
4639 {
4640         struct btrfs_fs_info *fs_info = root->fs_info;
4641         struct extent_io_tree *unpin;
4642         u64 start;
4643         u64 end;
4644         int ret;
4645
4646         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4647                 unpin = &fs_info->freed_extents[1];
4648         else
4649                 unpin = &fs_info->freed_extents[0];
4650
4651         while (1) {
4652                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4653                                             EXTENT_DIRTY);
4654                 if (ret)
4655                         break;
4656
4657                 if (btrfs_test_opt(root, DISCARD))
4658                         ret = btrfs_discard_extent(root, start,
4659                                                    end + 1 - start, NULL);
4660
4661                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4662                 unpin_extent_range(root, start, end);
4663                 cond_resched();
4664         }
4665
4666         return 0;
4667 }
4668
4669 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4670                                 struct btrfs_root *root,
4671                                 u64 bytenr, u64 num_bytes, u64 parent,
4672                                 u64 root_objectid, u64 owner_objectid,
4673                                 u64 owner_offset, int refs_to_drop,
4674                                 struct btrfs_delayed_extent_op *extent_op)
4675 {
4676         struct btrfs_key key;
4677         struct btrfs_path *path;
4678         struct btrfs_fs_info *info = root->fs_info;
4679         struct btrfs_root *extent_root = info->extent_root;
4680         struct extent_buffer *leaf;
4681         struct btrfs_extent_item *ei;
4682         struct btrfs_extent_inline_ref *iref;
4683         int ret;
4684         int is_data;
4685         int extent_slot = 0;
4686         int found_extent = 0;
4687         int num_to_del = 1;
4688         u32 item_size;
4689         u64 refs;
4690
4691         path = btrfs_alloc_path();
4692         if (!path)
4693                 return -ENOMEM;
4694
4695         path->reada = 1;
4696         path->leave_spinning = 1;
4697
4698         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4699         BUG_ON(!is_data && refs_to_drop != 1);
4700
4701         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4702                                     bytenr, num_bytes, parent,
4703                                     root_objectid, owner_objectid,
4704                                     owner_offset);
4705         if (ret == 0) {
4706                 extent_slot = path->slots[0];
4707                 while (extent_slot >= 0) {
4708                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4709                                               extent_slot);
4710                         if (key.objectid != bytenr)
4711                                 break;
4712                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4713                             key.offset == num_bytes) {
4714                                 found_extent = 1;
4715                                 break;
4716                         }
4717                         if (path->slots[0] - extent_slot > 5)
4718                                 break;
4719                         extent_slot--;
4720                 }
4721 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4722                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4723                 if (found_extent && item_size < sizeof(*ei))
4724                         found_extent = 0;
4725 #endif
4726                 if (!found_extent) {
4727                         BUG_ON(iref);
4728                         ret = remove_extent_backref(trans, extent_root, path,
4729                                                     NULL, refs_to_drop,
4730                                                     is_data);
4731                         BUG_ON(ret);
4732                         btrfs_release_path(path);
4733                         path->leave_spinning = 1;
4734
4735                         key.objectid = bytenr;
4736                         key.type = BTRFS_EXTENT_ITEM_KEY;
4737                         key.offset = num_bytes;
4738
4739                         ret = btrfs_search_slot(trans, extent_root,
4740                                                 &key, path, -1, 1);
4741                         if (ret) {
4742                                 printk(KERN_ERR "umm, got %d back from search"
4743                                        ", was looking for %llu\n", ret,
4744                                        (unsigned long long)bytenr);
4745                                 if (ret > 0)
4746                                         btrfs_print_leaf(extent_root,
4747                                                          path->nodes[0]);
4748                         }
4749                         BUG_ON(ret);
4750                         extent_slot = path->slots[0];
4751                 }
4752         } else {
4753                 btrfs_print_leaf(extent_root, path->nodes[0]);
4754                 WARN_ON(1);
4755                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4756                        "parent %llu root %llu  owner %llu offset %llu\n",
4757                        (unsigned long long)bytenr,
4758                        (unsigned long long)parent,
4759                        (unsigned long long)root_objectid,
4760                        (unsigned long long)owner_objectid,
4761                        (unsigned long long)owner_offset);
4762         }
4763
4764         leaf = path->nodes[0];
4765         item_size = btrfs_item_size_nr(leaf, extent_slot);
4766 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4767         if (item_size < sizeof(*ei)) {
4768                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4769                 ret = convert_extent_item_v0(trans, extent_root, path,
4770                                              owner_objectid, 0);
4771                 BUG_ON(ret < 0);
4772
4773                 btrfs_release_path(path);
4774                 path->leave_spinning = 1;
4775
4776                 key.objectid = bytenr;
4777                 key.type = BTRFS_EXTENT_ITEM_KEY;
4778                 key.offset = num_bytes;
4779
4780                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4781                                         -1, 1);
4782                 if (ret) {
4783                         printk(KERN_ERR "umm, got %d back from search"
4784                                ", was looking for %llu\n", ret,
4785                                (unsigned long long)bytenr);
4786                         btrfs_print_leaf(extent_root, path->nodes[0]);
4787                 }
4788                 BUG_ON(ret);
4789                 extent_slot = path->slots[0];
4790                 leaf = path->nodes[0];
4791                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4792         }
4793 #endif
4794         BUG_ON(item_size < sizeof(*ei));
4795         ei = btrfs_item_ptr(leaf, extent_slot,
4796                             struct btrfs_extent_item);
4797         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4798                 struct btrfs_tree_block_info *bi;
4799                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4800                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4801                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4802         }
4803
4804         refs = btrfs_extent_refs(leaf, ei);
4805         BUG_ON(refs < refs_to_drop);
4806         refs -= refs_to_drop;
4807
4808         if (refs > 0) {
4809                 if (extent_op)
4810                         __run_delayed_extent_op(extent_op, leaf, ei);
4811                 /*
4812                  * In the case of inline back ref, reference count will
4813                  * be updated by remove_extent_backref
4814                  */
4815                 if (iref) {
4816                         BUG_ON(!found_extent);
4817                 } else {
4818                         btrfs_set_extent_refs(leaf, ei, refs);
4819                         btrfs_mark_buffer_dirty(leaf);
4820                 }
4821                 if (found_extent) {
4822                         ret = remove_extent_backref(trans, extent_root, path,
4823                                                     iref, refs_to_drop,
4824                                                     is_data);
4825                         BUG_ON(ret);
4826                 }
4827         } else {
4828                 if (found_extent) {
4829                         BUG_ON(is_data && refs_to_drop !=
4830                                extent_data_ref_count(root, path, iref));
4831                         if (iref) {
4832                                 BUG_ON(path->slots[0] != extent_slot);
4833                         } else {
4834                                 BUG_ON(path->slots[0] != extent_slot + 1);
4835                                 path->slots[0] = extent_slot;
4836                                 num_to_del = 2;
4837                         }
4838                 }
4839
4840                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4841                                       num_to_del);
4842                 BUG_ON(ret);
4843                 btrfs_release_path(path);
4844
4845                 if (is_data) {
4846                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4847                         BUG_ON(ret);
4848                 } else {
4849                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4850                              bytenr >> PAGE_CACHE_SHIFT,
4851                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4852                 }
4853
4854                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4855                 BUG_ON(ret);
4856         }
4857         btrfs_free_path(path);
4858         return ret;
4859 }
4860
4861 /*
4862  * when we free an block, it is possible (and likely) that we free the last
4863  * delayed ref for that extent as well.  This searches the delayed ref tree for
4864  * a given extent, and if there are no other delayed refs to be processed, it
4865  * removes it from the tree.
4866  */
4867 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4868                                       struct btrfs_root *root, u64 bytenr)
4869 {
4870         struct btrfs_delayed_ref_head *head;
4871         struct btrfs_delayed_ref_root *delayed_refs;
4872         struct btrfs_delayed_ref_node *ref;
4873         struct rb_node *node;
4874         int ret = 0;
4875
4876         delayed_refs = &trans->transaction->delayed_refs;
4877         spin_lock(&delayed_refs->lock);
4878         head = btrfs_find_delayed_ref_head(trans, bytenr);
4879         if (!head)
4880                 goto out;
4881
4882         node = rb_prev(&head->node.rb_node);
4883         if (!node)
4884                 goto out;
4885
4886         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4887
4888         /* there are still entries for this ref, we can't drop it */
4889         if (ref->bytenr == bytenr)
4890                 goto out;
4891
4892         if (head->extent_op) {
4893                 if (!head->must_insert_reserved)
4894                         goto out;
4895                 kfree(head->extent_op);
4896                 head->extent_op = NULL;
4897         }
4898
4899         /*
4900          * waiting for the lock here would deadlock.  If someone else has it
4901          * locked they are already in the process of dropping it anyway
4902          */
4903         if (!mutex_trylock(&head->mutex))
4904                 goto out;
4905
4906         /*
4907          * at this point we have a head with no other entries.  Go
4908          * ahead and process it.
4909          */
4910         head->node.in_tree = 0;
4911         rb_erase(&head->node.rb_node, &delayed_refs->root);
4912
4913         delayed_refs->num_entries--;
4914
4915         /*
4916          * we don't take a ref on the node because we're removing it from the
4917          * tree, so we just steal the ref the tree was holding.
4918          */
4919         delayed_refs->num_heads--;
4920         if (list_empty(&head->cluster))
4921                 delayed_refs->num_heads_ready--;
4922
4923         list_del_init(&head->cluster);
4924         spin_unlock(&delayed_refs->lock);
4925
4926         BUG_ON(head->extent_op);
4927         if (head->must_insert_reserved)
4928                 ret = 1;
4929
4930         mutex_unlock(&head->mutex);
4931         btrfs_put_delayed_ref(&head->node);
4932         return ret;
4933 out:
4934         spin_unlock(&delayed_refs->lock);
4935         return 0;
4936 }
4937
4938 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4939                            struct btrfs_root *root,
4940                            struct extent_buffer *buf,
4941                            u64 parent, int last_ref)
4942 {
4943         struct btrfs_block_group_cache *cache = NULL;
4944         int ret;
4945
4946         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4947                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4948                                                 parent, root->root_key.objectid,
4949                                                 btrfs_header_level(buf),
4950                                                 BTRFS_DROP_DELAYED_REF, NULL);
4951                 BUG_ON(ret);
4952         }
4953
4954         if (!last_ref)
4955                 return;
4956
4957         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4958
4959         if (btrfs_header_generation(buf) == trans->transid) {
4960                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4961                         ret = check_ref_cleanup(trans, root, buf->start);
4962                         if (!ret)
4963                                 goto out;
4964                 }
4965
4966                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4967                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4968                         goto out;
4969                 }
4970
4971                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4972
4973                 btrfs_add_free_space(cache, buf->start, buf->len);
4974                 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
4975         }
4976 out:
4977         /*
4978          * Deleting the buffer, clear the corrupt flag since it doesn't matter
4979          * anymore.
4980          */
4981         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4982         btrfs_put_block_group(cache);
4983 }
4984
4985 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4986                       struct btrfs_root *root,
4987                       u64 bytenr, u64 num_bytes, u64 parent,
4988                       u64 root_objectid, u64 owner, u64 offset)
4989 {
4990         int ret;
4991
4992         /*
4993          * tree log blocks never actually go into the extent allocation
4994          * tree, just update pinning info and exit early.
4995          */
4996         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4997                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4998                 /* unlocks the pinned mutex */
4999                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5000                 ret = 0;
5001         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5002                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
5003                                         parent, root_objectid, (int)owner,
5004                                         BTRFS_DROP_DELAYED_REF, NULL);
5005                 BUG_ON(ret);
5006         } else {
5007                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
5008                                         parent, root_objectid, owner,
5009                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
5010                 BUG_ON(ret);
5011         }
5012         return ret;
5013 }
5014
5015 static u64 stripe_align(struct btrfs_root *root, u64 val)
5016 {
5017         u64 mask = ((u64)root->stripesize - 1);
5018         u64 ret = (val + mask) & ~mask;
5019         return ret;
5020 }
5021
5022 /*
5023  * when we wait for progress in the block group caching, its because
5024  * our allocation attempt failed at least once.  So, we must sleep
5025  * and let some progress happen before we try again.
5026  *
5027  * This function will sleep at least once waiting for new free space to
5028  * show up, and then it will check the block group free space numbers
5029  * for our min num_bytes.  Another option is to have it go ahead
5030  * and look in the rbtree for a free extent of a given size, but this
5031  * is a good start.
5032  */
5033 static noinline int
5034 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5035                                 u64 num_bytes)
5036 {
5037         struct btrfs_caching_control *caching_ctl;
5038         DEFINE_WAIT(wait);
5039
5040         caching_ctl = get_caching_control(cache);
5041         if (!caching_ctl)
5042                 return 0;
5043
5044         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5045                    (cache->free_space_ctl->free_space >= num_bytes));
5046
5047         put_caching_control(caching_ctl);
5048         return 0;
5049 }
5050
5051 static noinline int
5052 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5053 {
5054         struct btrfs_caching_control *caching_ctl;
5055         DEFINE_WAIT(wait);
5056
5057         caching_ctl = get_caching_control(cache);
5058         if (!caching_ctl)
5059                 return 0;
5060
5061         wait_event(caching_ctl->wait, block_group_cache_done(cache));
5062
5063         put_caching_control(caching_ctl);
5064         return 0;
5065 }
5066
5067 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5068 {
5069         int index;
5070         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5071                 index = 0;
5072         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5073                 index = 1;
5074         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5075                 index = 2;
5076         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5077                 index = 3;
5078         else
5079                 index = 4;
5080         return index;
5081 }
5082
5083 enum btrfs_loop_type {
5084         LOOP_FIND_IDEAL = 0,
5085         LOOP_CACHING_NOWAIT = 1,
5086         LOOP_CACHING_WAIT = 2,
5087         LOOP_ALLOC_CHUNK = 3,
5088         LOOP_NO_EMPTY_SIZE = 4,
5089 };
5090
5091 /*
5092  * walks the btree of allocated extents and find a hole of a given size.
5093  * The key ins is changed to record the hole:
5094  * ins->objectid == block start
5095  * ins->flags = BTRFS_EXTENT_ITEM_KEY
5096  * ins->offset == number of blocks
5097  * Any available blocks before search_start are skipped.
5098  */
5099 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5100                                      struct btrfs_root *orig_root,
5101                                      u64 num_bytes, u64 empty_size,
5102                                      u64 search_start, u64 search_end,
5103                                      u64 hint_byte, struct btrfs_key *ins,
5104                                      u64 data)
5105 {
5106         int ret = 0;
5107         struct btrfs_root *root = orig_root->fs_info->extent_root;
5108         struct btrfs_free_cluster *last_ptr = NULL;
5109         struct btrfs_block_group_cache *block_group = NULL;
5110         struct btrfs_block_group_cache *used_block_group;
5111         int empty_cluster = 2 * 1024 * 1024;
5112         int allowed_chunk_alloc = 0;
5113         int done_chunk_alloc = 0;
5114         struct btrfs_space_info *space_info;
5115         int loop = 0;
5116         int index = 0;
5117         int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5118                 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5119         bool found_uncached_bg = false;
5120         bool failed_cluster_refill = false;
5121         bool failed_alloc = false;
5122         bool use_cluster = true;
5123         bool have_caching_bg = false;
5124         u64 ideal_cache_percent = 0;
5125         u64 ideal_cache_offset = 0;
5126
5127         WARN_ON(num_bytes < root->sectorsize);
5128         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5129         ins->objectid = 0;
5130         ins->offset = 0;
5131
5132         space_info = __find_space_info(root->fs_info, data);
5133         if (!space_info) {
5134                 printk(KERN_ERR "No space info for %llu\n", data);
5135                 return -ENOSPC;
5136         }
5137
5138         /*
5139          * If the space info is for both data and metadata it means we have a
5140          * small filesystem and we can't use the clustering stuff.
5141          */
5142         if (btrfs_mixed_space_info(space_info))
5143                 use_cluster = false;
5144
5145         if (orig_root->ref_cows || empty_size)
5146                 allowed_chunk_alloc = 1;
5147
5148         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5149                 last_ptr = &root->fs_info->meta_alloc_cluster;
5150                 if (!btrfs_test_opt(root, SSD))
5151                         empty_cluster = 64 * 1024;
5152         }
5153
5154         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5155             btrfs_test_opt(root, SSD)) {
5156                 last_ptr = &root->fs_info->data_alloc_cluster;
5157         }
5158
5159         if (last_ptr) {
5160                 spin_lock(&last_ptr->lock);
5161                 if (last_ptr->block_group)
5162                         hint_byte = last_ptr->window_start;
5163                 spin_unlock(&last_ptr->lock);
5164         }
5165
5166         search_start = max(search_start, first_logical_byte(root, 0));
5167         search_start = max(search_start, hint_byte);
5168
5169         if (!last_ptr)
5170                 empty_cluster = 0;
5171
5172         if (search_start == hint_byte) {
5173 ideal_cache:
5174                 block_group = btrfs_lookup_block_group(root->fs_info,
5175                                                        search_start);
5176                 used_block_group = block_group;
5177                 /*
5178                  * we don't want to use the block group if it doesn't match our
5179                  * allocation bits, or if its not cached.
5180                  *
5181                  * However if we are re-searching with an ideal block group
5182                  * picked out then we don't care that the block group is cached.
5183                  */
5184                 if (block_group && block_group_bits(block_group, data) &&
5185                     (block_group->cached != BTRFS_CACHE_NO ||
5186                      search_start == ideal_cache_offset)) {
5187                         down_read(&space_info->groups_sem);
5188                         if (list_empty(&block_group->list) ||
5189                             block_group->ro) {
5190                                 /*
5191                                  * someone is removing this block group,
5192                                  * we can't jump into the have_block_group
5193                                  * target because our list pointers are not
5194                                  * valid
5195                                  */
5196                                 btrfs_put_block_group(block_group);
5197                                 up_read(&space_info->groups_sem);
5198                         } else {
5199                                 index = get_block_group_index(block_group);
5200                                 goto have_block_group;
5201                         }
5202                 } else if (block_group) {
5203                         btrfs_put_block_group(block_group);
5204                 }
5205         }
5206 search:
5207         have_caching_bg = false;
5208         down_read(&space_info->groups_sem);
5209         list_for_each_entry(block_group, &space_info->block_groups[index],
5210                             list) {
5211                 u64 offset;
5212                 int cached;
5213
5214                 used_block_group = block_group;
5215                 btrfs_get_block_group(block_group);
5216                 search_start = block_group->key.objectid;
5217
5218                 /*
5219                  * this can happen if we end up cycling through all the
5220                  * raid types, but we want to make sure we only allocate
5221                  * for the proper type.
5222                  */
5223                 if (!block_group_bits(block_group, data)) {
5224                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
5225                                 BTRFS_BLOCK_GROUP_RAID1 |
5226                                 BTRFS_BLOCK_GROUP_RAID10;
5227
5228                         /*
5229                          * if they asked for extra copies and this block group
5230                          * doesn't provide them, bail.  This does allow us to
5231                          * fill raid0 from raid1.
5232                          */
5233                         if ((data & extra) && !(block_group->flags & extra))
5234                                 goto loop;
5235                 }
5236
5237 have_block_group:
5238                 cached = block_group_cache_done(block_group);
5239                 if (unlikely(!cached)) {
5240                         u64 free_percent;
5241
5242                         found_uncached_bg = true;
5243                         ret = cache_block_group(block_group, trans,
5244                                                 orig_root, 1);
5245                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5246                                 goto alloc;
5247
5248                         free_percent = btrfs_block_group_used(&block_group->item);
5249                         free_percent *= 100;
5250                         free_percent = div64_u64(free_percent,
5251                                                  block_group->key.offset);
5252                         free_percent = 100 - free_percent;
5253                         if (free_percent > ideal_cache_percent &&
5254                             likely(!block_group->ro)) {
5255                                 ideal_cache_offset = block_group->key.objectid;
5256                                 ideal_cache_percent = free_percent;
5257                         }
5258
5259                         /*
5260                          * The caching workers are limited to 2 threads, so we
5261                          * can queue as much work as we care to.
5262                          */
5263                         if (loop > LOOP_FIND_IDEAL) {
5264                                 ret = cache_block_group(block_group, trans,
5265                                                         orig_root, 0);
5266                                 BUG_ON(ret);
5267                         }
5268
5269                         /*
5270                          * If loop is set for cached only, try the next block
5271                          * group.
5272                          */
5273                         if (loop == LOOP_FIND_IDEAL)
5274                                 goto loop;
5275                 }
5276
5277 alloc:
5278                 if (unlikely(block_group->ro))
5279                         goto loop;
5280
5281                 spin_lock(&block_group->free_space_ctl->tree_lock);
5282                 if (cached &&
5283                     block_group->free_space_ctl->free_space <
5284                     num_bytes + empty_cluster + empty_size) {
5285                         spin_unlock(&block_group->free_space_ctl->tree_lock);
5286                         goto loop;
5287                 }
5288                 spin_unlock(&block_group->free_space_ctl->tree_lock);
5289
5290                 /*
5291                  * Ok we want to try and use the cluster allocator, so
5292                  * lets look there
5293                  */
5294                 if (last_ptr) {
5295                         /*
5296                          * the refill lock keeps out other
5297                          * people trying to start a new cluster
5298                          */
5299                         spin_lock(&last_ptr->refill_lock);
5300                         used_block_group = last_ptr->block_group;
5301                         if (used_block_group != block_group &&
5302                             (!used_block_group ||
5303                              used_block_group->ro ||
5304                              !block_group_bits(used_block_group, data))) {
5305                                 used_block_group = block_group;
5306                                 goto refill_cluster;
5307                         }
5308
5309                         if (used_block_group != block_group)
5310                                 btrfs_get_block_group(used_block_group);
5311
5312                         offset = btrfs_alloc_from_cluster(used_block_group,
5313                           last_ptr, num_bytes, used_block_group->key.objectid);
5314                         if (offset) {
5315                                 /* we have a block, we're done */
5316                                 spin_unlock(&last_ptr->refill_lock);
5317                                 goto checks;
5318                         }
5319
5320                         WARN_ON(last_ptr->block_group != used_block_group);
5321                         if (used_block_group != block_group) {
5322                                 btrfs_put_block_group(used_block_group);
5323                                 used_block_group = block_group;
5324                         }
5325 refill_cluster:
5326                         BUG_ON(used_block_group != block_group);
5327                         /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5328                          * set up a new clusters, so lets just skip it
5329                          * and let the allocator find whatever block
5330                          * it can find.  If we reach this point, we
5331                          * will have tried the cluster allocator
5332                          * plenty of times and not have found
5333                          * anything, so we are likely way too
5334                          * fragmented for the clustering stuff to find
5335                          * anything.  */
5336                         if (loop >= LOOP_NO_EMPTY_SIZE) {
5337                                 spin_unlock(&last_ptr->refill_lock);
5338                                 goto unclustered_alloc;
5339                         }
5340
5341                         /*
5342                          * this cluster didn't work out, free it and
5343                          * start over
5344                          */
5345                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5346
5347                         /* allocate a cluster in this block group */
5348                         ret = btrfs_find_space_cluster(trans, root,
5349                                                block_group, last_ptr,
5350                                                search_start, num_bytes,
5351                                                empty_cluster + empty_size);
5352                         if (ret == 0) {
5353                                 /*
5354                                  * now pull our allocation out of this
5355                                  * cluster
5356                                  */
5357                                 offset = btrfs_alloc_from_cluster(block_group,
5358                                                   last_ptr, num_bytes,
5359                                                   search_start);
5360                                 if (offset) {
5361                                         /* we found one, proceed */
5362                                         spin_unlock(&last_ptr->refill_lock);
5363                                         goto checks;
5364                                 }
5365                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5366                                    && !failed_cluster_refill) {
5367                                 spin_unlock(&last_ptr->refill_lock);
5368
5369                                 failed_cluster_refill = true;
5370                                 wait_block_group_cache_progress(block_group,
5371                                        num_bytes + empty_cluster + empty_size);
5372                                 goto have_block_group;
5373                         }
5374
5375                         /*
5376                          * at this point we either didn't find a cluster
5377                          * or we weren't able to allocate a block from our
5378                          * cluster.  Free the cluster we've been trying
5379                          * to use, and go to the next block group
5380                          */
5381                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5382                         spin_unlock(&last_ptr->refill_lock);
5383                         goto loop;
5384                 }
5385
5386 unclustered_alloc:
5387                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5388                                                     num_bytes, empty_size);
5389                 /*
5390                  * If we didn't find a chunk, and we haven't failed on this
5391                  * block group before, and this block group is in the middle of
5392                  * caching and we are ok with waiting, then go ahead and wait
5393                  * for progress to be made, and set failed_alloc to true.
5394                  *
5395                  * If failed_alloc is true then we've already waited on this
5396                  * block group once and should move on to the next block group.
5397                  */
5398                 if (!offset && !failed_alloc && !cached &&
5399                     loop > LOOP_CACHING_NOWAIT) {
5400                         wait_block_group_cache_progress(block_group,
5401                                                 num_bytes + empty_size);
5402                         failed_alloc = true;
5403                         goto have_block_group;
5404                 } else if (!offset) {
5405                         if (!cached)
5406                                 have_caching_bg = true;
5407                         goto loop;
5408                 }
5409 checks:
5410                 search_start = stripe_align(root, offset);
5411                 /* move on to the next group */
5412                 if (search_start + num_bytes >= search_end) {
5413                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5414                         goto loop;
5415                 }
5416
5417                 /* move on to the next group */
5418                 if (search_start + num_bytes >
5419                     used_block_group->key.objectid + used_block_group->key.offset) {
5420                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5421                         goto loop;
5422                 }
5423
5424                 ins->objectid = search_start;
5425                 ins->offset = num_bytes;
5426
5427                 if (offset < search_start)
5428                         btrfs_add_free_space(used_block_group, offset,
5429                                              search_start - offset);
5430                 BUG_ON(offset > search_start);
5431
5432                 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5433                                                   alloc_type);
5434                 if (ret == -EAGAIN) {
5435                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5436                         goto loop;
5437                 }
5438
5439                 /* we are all good, lets return */
5440                 ins->objectid = search_start;
5441                 ins->offset = num_bytes;
5442
5443                 if (offset < search_start)
5444                         btrfs_add_free_space(used_block_group, offset,
5445                                              search_start - offset);
5446                 BUG_ON(offset > search_start);
5447                 if (used_block_group != block_group)
5448                         btrfs_put_block_group(used_block_group);
5449                 btrfs_put_block_group(block_group);
5450                 break;
5451 loop:
5452                 failed_cluster_refill = false;
5453                 failed_alloc = false;
5454                 BUG_ON(index != get_block_group_index(block_group));
5455                 if (used_block_group != block_group)
5456                         btrfs_put_block_group(used_block_group);
5457                 btrfs_put_block_group(block_group);
5458         }
5459         up_read(&space_info->groups_sem);
5460
5461         if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5462                 goto search;
5463
5464         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5465                 goto search;
5466
5467         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5468          *                      for them to make caching progress.  Also
5469          *                      determine the best possible bg to cache
5470          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5471          *                      caching kthreads as we move along
5472          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5473          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5474          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5475          *                      again
5476          */
5477         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5478                 index = 0;
5479                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5480                         found_uncached_bg = false;
5481                         loop++;
5482                         if (!ideal_cache_percent)
5483                                 goto search;
5484
5485                         /*
5486                          * 1 of the following 2 things have happened so far
5487                          *
5488                          * 1) We found an ideal block group for caching that
5489                          * is mostly full and will cache quickly, so we might
5490                          * as well wait for it.
5491                          *
5492                          * 2) We searched for cached only and we didn't find
5493                          * anything, and we didn't start any caching kthreads
5494                          * either, so chances are we will loop through and
5495                          * start a couple caching kthreads, and then come back
5496                          * around and just wait for them.  This will be slower
5497                          * because we will have 2 caching kthreads reading at
5498                          * the same time when we could have just started one
5499                          * and waited for it to get far enough to give us an
5500                          * allocation, so go ahead and go to the wait caching
5501                          * loop.
5502                          */
5503                         loop = LOOP_CACHING_WAIT;
5504                         search_start = ideal_cache_offset;
5505                         ideal_cache_percent = 0;
5506                         goto ideal_cache;
5507                 } else if (loop == LOOP_FIND_IDEAL) {
5508                         /*
5509                          * Didn't find a uncached bg, wait on anything we find
5510                          * next.
5511                          */
5512                         loop = LOOP_CACHING_WAIT;
5513                         goto search;
5514                 }
5515
5516                 loop++;
5517
5518                 if (loop == LOOP_ALLOC_CHUNK) {
5519                        if (allowed_chunk_alloc) {
5520                                 ret = do_chunk_alloc(trans, root, num_bytes +
5521                                                      2 * 1024 * 1024, data,
5522                                                      CHUNK_ALLOC_LIMITED);
5523                                 allowed_chunk_alloc = 0;
5524                                 if (ret == 1)
5525                                         done_chunk_alloc = 1;
5526                         } else if (!done_chunk_alloc &&
5527                                    space_info->force_alloc ==
5528                                    CHUNK_ALLOC_NO_FORCE) {
5529                                 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5530                         }
5531
5532                        /*
5533                         * We didn't allocate a chunk, go ahead and drop the
5534                         * empty size and loop again.
5535                         */
5536                        if (!done_chunk_alloc)
5537                                loop = LOOP_NO_EMPTY_SIZE;
5538                 }
5539
5540                 if (loop == LOOP_NO_EMPTY_SIZE) {
5541                         empty_size = 0;
5542                         empty_cluster = 0;
5543                 }
5544
5545                 goto search;
5546         } else if (!ins->objectid) {
5547                 ret = -ENOSPC;
5548         } else if (ins->objectid) {
5549                 ret = 0;
5550         }
5551
5552         return ret;
5553 }
5554
5555 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5556                             int dump_block_groups)
5557 {
5558         struct btrfs_block_group_cache *cache;
5559         int index = 0;
5560
5561         spin_lock(&info->lock);
5562         printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5563                (unsigned long long)info->flags,
5564                (unsigned long long)(info->total_bytes - info->bytes_used -
5565                                     info->bytes_pinned - info->bytes_reserved -
5566                                     info->bytes_readonly),
5567                (info->full) ? "" : "not ");
5568         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5569                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5570                (unsigned long long)info->total_bytes,
5571                (unsigned long long)info->bytes_used,
5572                (unsigned long long)info->bytes_pinned,
5573                (unsigned long long)info->bytes_reserved,
5574                (unsigned long long)info->bytes_may_use,
5575                (unsigned long long)info->bytes_readonly);
5576         spin_unlock(&info->lock);
5577
5578         if (!dump_block_groups)
5579                 return;
5580
5581         down_read(&info->groups_sem);
5582 again:
5583         list_for_each_entry(cache, &info->block_groups[index], list) {
5584                 spin_lock(&cache->lock);
5585                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5586                        "%llu pinned %llu reserved\n",
5587                        (unsigned long long)cache->key.objectid,
5588                        (unsigned long long)cache->key.offset,
5589                        (unsigned long long)btrfs_block_group_used(&cache->item),
5590                        (unsigned long long)cache->pinned,
5591                        (unsigned long long)cache->reserved);
5592                 btrfs_dump_free_space(cache, bytes);
5593                 spin_unlock(&cache->lock);
5594         }
5595         if (++index < BTRFS_NR_RAID_TYPES)
5596                 goto again;
5597         up_read(&info->groups_sem);
5598 }
5599
5600 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5601                          struct btrfs_root *root,
5602                          u64 num_bytes, u64 min_alloc_size,
5603                          u64 empty_size, u64 hint_byte,
5604                          u64 search_end, struct btrfs_key *ins,
5605                          u64 data)
5606 {
5607         int ret;
5608         u64 search_start = 0;
5609
5610         data = btrfs_get_alloc_profile(root, data);
5611 again:
5612         /*
5613          * the only place that sets empty_size is btrfs_realloc_node, which
5614          * is not called recursively on allocations
5615          */
5616         if (empty_size || root->ref_cows)
5617                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5618                                      num_bytes + 2 * 1024 * 1024, data,
5619                                      CHUNK_ALLOC_NO_FORCE);
5620
5621         WARN_ON(num_bytes < root->sectorsize);
5622         ret = find_free_extent(trans, root, num_bytes, empty_size,
5623                                search_start, search_end, hint_byte,
5624                                ins, data);
5625
5626         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5627                 num_bytes = num_bytes >> 1;
5628                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5629                 num_bytes = max(num_bytes, min_alloc_size);
5630                 do_chunk_alloc(trans, root->fs_info->extent_root,
5631                                num_bytes, data, CHUNK_ALLOC_FORCE);
5632                 goto again;
5633         }
5634         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5635                 struct btrfs_space_info *sinfo;
5636
5637                 sinfo = __find_space_info(root->fs_info, data);
5638                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5639                        "wanted %llu\n", (unsigned long long)data,
5640                        (unsigned long long)num_bytes);
5641                 dump_space_info(sinfo, num_bytes, 1);
5642         }
5643
5644         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5645
5646         return ret;
5647 }
5648
5649 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5650                                         u64 start, u64 len, int pin)
5651 {
5652         struct btrfs_block_group_cache *cache;
5653         int ret = 0;
5654
5655         cache = btrfs_lookup_block_group(root->fs_info, start);
5656         if (!cache) {
5657                 printk(KERN_ERR "Unable to find block group for %llu\n",
5658                        (unsigned long long)start);
5659                 return -ENOSPC;
5660         }
5661
5662         if (btrfs_test_opt(root, DISCARD))
5663                 ret = btrfs_discard_extent(root, start, len, NULL);
5664
5665         if (pin)
5666                 pin_down_extent(root, cache, start, len, 1);
5667         else {
5668                 btrfs_add_free_space(cache, start, len);
5669                 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5670         }
5671         btrfs_put_block_group(cache);
5672
5673         trace_btrfs_reserved_extent_free(root, start, len);
5674
5675         return ret;
5676 }
5677
5678 int btrfs_free_reserved_extent(struct btrfs_root *root,
5679                                         u64 start, u64 len)
5680 {
5681         return __btrfs_free_reserved_extent(root, start, len, 0);
5682 }
5683
5684 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5685                                        u64 start, u64 len)
5686 {
5687         return __btrfs_free_reserved_extent(root, start, len, 1);
5688 }
5689
5690 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5691                                       struct btrfs_root *root,
5692                                       u64 parent, u64 root_objectid,
5693                                       u64 flags, u64 owner, u64 offset,
5694                                       struct btrfs_key *ins, int ref_mod)
5695 {
5696         int ret;
5697         struct btrfs_fs_info *fs_info = root->fs_info;
5698         struct btrfs_extent_item *extent_item;
5699         struct btrfs_extent_inline_ref *iref;
5700         struct btrfs_path *path;
5701         struct extent_buffer *leaf;
5702         int type;
5703         u32 size;
5704
5705         if (parent > 0)
5706                 type = BTRFS_SHARED_DATA_REF_KEY;
5707         else
5708                 type = BTRFS_EXTENT_DATA_REF_KEY;
5709
5710         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5711
5712         path = btrfs_alloc_path();
5713         if (!path)
5714                 return -ENOMEM;
5715
5716         path->leave_spinning = 1;
5717         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5718                                       ins, size);
5719         BUG_ON(ret);
5720
5721         leaf = path->nodes[0];
5722         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5723                                      struct btrfs_extent_item);
5724         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5725         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5726         btrfs_set_extent_flags(leaf, extent_item,
5727                                flags | BTRFS_EXTENT_FLAG_DATA);
5728
5729         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5730         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5731         if (parent > 0) {
5732                 struct btrfs_shared_data_ref *ref;
5733                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5734                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5735                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5736         } else {
5737                 struct btrfs_extent_data_ref *ref;
5738                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5739                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5740                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5741                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5742                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5743         }
5744
5745         btrfs_mark_buffer_dirty(path->nodes[0]);
5746         btrfs_free_path(path);
5747
5748         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5749         if (ret) {
5750                 printk(KERN_ERR "btrfs update block group failed for %llu "
5751                        "%llu\n", (unsigned long long)ins->objectid,
5752                        (unsigned long long)ins->offset);
5753                 BUG();
5754         }
5755         return ret;
5756 }
5757
5758 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5759                                      struct btrfs_root *root,
5760                                      u64 parent, u64 root_objectid,
5761                                      u64 flags, struct btrfs_disk_key *key,
5762                                      int level, struct btrfs_key *ins)
5763 {
5764         int ret;
5765         struct btrfs_fs_info *fs_info = root->fs_info;
5766         struct btrfs_extent_item *extent_item;
5767         struct btrfs_tree_block_info *block_info;
5768         struct btrfs_extent_inline_ref *iref;
5769         struct btrfs_path *path;
5770         struct extent_buffer *leaf;
5771         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5772
5773         path = btrfs_alloc_path();
5774         if (!path)
5775                 return -ENOMEM;
5776
5777         path->leave_spinning = 1;
5778         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5779                                       ins, size);
5780         BUG_ON(ret);
5781
5782         leaf = path->nodes[0];
5783         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5784                                      struct btrfs_extent_item);
5785         btrfs_set_extent_refs(leaf, extent_item, 1);
5786         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5787         btrfs_set_extent_flags(leaf, extent_item,
5788                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5789         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5790
5791         btrfs_set_tree_block_key(leaf, block_info, key);
5792         btrfs_set_tree_block_level(leaf, block_info, level);
5793
5794         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5795         if (parent > 0) {
5796                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5797                 btrfs_set_extent_inline_ref_type(leaf, iref,
5798                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5799                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5800         } else {
5801                 btrfs_set_extent_inline_ref_type(leaf, iref,
5802                                                  BTRFS_TREE_BLOCK_REF_KEY);
5803                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5804         }
5805
5806         btrfs_mark_buffer_dirty(leaf);
5807         btrfs_free_path(path);
5808
5809         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5810         if (ret) {
5811                 printk(KERN_ERR "btrfs update block group failed for %llu "
5812                        "%llu\n", (unsigned long long)ins->objectid,
5813                        (unsigned long long)ins->offset);
5814                 BUG();
5815         }
5816         return ret;
5817 }
5818
5819 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5820                                      struct btrfs_root *root,
5821                                      u64 root_objectid, u64 owner,
5822                                      u64 offset, struct btrfs_key *ins)
5823 {
5824         int ret;
5825
5826         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5827
5828         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5829                                          0, root_objectid, owner, offset,
5830                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5831         return ret;
5832 }
5833
5834 /*
5835  * this is used by the tree logging recovery code.  It records that
5836  * an extent has been allocated and makes sure to clear the free
5837  * space cache bits as well
5838  */
5839 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5840                                    struct btrfs_root *root,
5841                                    u64 root_objectid, u64 owner, u64 offset,
5842                                    struct btrfs_key *ins)
5843 {
5844         int ret;
5845         struct btrfs_block_group_cache *block_group;
5846         struct btrfs_caching_control *caching_ctl;
5847         u64 start = ins->objectid;
5848         u64 num_bytes = ins->offset;
5849
5850         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5851         cache_block_group(block_group, trans, NULL, 0);
5852         caching_ctl = get_caching_control(block_group);
5853
5854         if (!caching_ctl) {
5855                 BUG_ON(!block_group_cache_done(block_group));
5856                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5857                 BUG_ON(ret);
5858         } else {
5859                 mutex_lock(&caching_ctl->mutex);
5860
5861                 if (start >= caching_ctl->progress) {
5862                         ret = add_excluded_extent(root, start, num_bytes);
5863                         BUG_ON(ret);
5864                 } else if (start + num_bytes <= caching_ctl->progress) {
5865                         ret = btrfs_remove_free_space(block_group,
5866                                                       start, num_bytes);
5867                         BUG_ON(ret);
5868                 } else {
5869                         num_bytes = caching_ctl->progress - start;
5870                         ret = btrfs_remove_free_space(block_group,
5871                                                       start, num_bytes);
5872                         BUG_ON(ret);
5873
5874                         start = caching_ctl->progress;
5875                         num_bytes = ins->objectid + ins->offset -
5876                                     caching_ctl->progress;
5877                         ret = add_excluded_extent(root, start, num_bytes);
5878                         BUG_ON(ret);
5879                 }
5880
5881                 mutex_unlock(&caching_ctl->mutex);
5882                 put_caching_control(caching_ctl);
5883         }
5884
5885         ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5886                                           RESERVE_ALLOC_NO_ACCOUNT);
5887         BUG_ON(ret);
5888         btrfs_put_block_group(block_group);
5889         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5890                                          0, owner, offset, ins, 1);
5891         return ret;
5892 }
5893
5894 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5895                                             struct btrfs_root *root,
5896                                             u64 bytenr, u32 blocksize,
5897                                             int level)
5898 {
5899         struct extent_buffer *buf;
5900
5901         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5902         if (!buf)
5903                 return ERR_PTR(-ENOMEM);
5904         btrfs_set_header_generation(buf, trans->transid);
5905         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5906         btrfs_tree_lock(buf);
5907         clean_tree_block(trans, root, buf);
5908
5909         btrfs_set_lock_blocking(buf);
5910         btrfs_set_buffer_uptodate(buf);
5911
5912         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5913                 /*
5914                  * we allow two log transactions at a time, use different
5915                  * EXENT bit to differentiate dirty pages.
5916                  */
5917                 if (root->log_transid % 2 == 0)
5918                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5919                                         buf->start + buf->len - 1, GFP_NOFS);
5920                 else
5921                         set_extent_new(&root->dirty_log_pages, buf->start,
5922                                         buf->start + buf->len - 1, GFP_NOFS);
5923         } else {
5924                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5925                          buf->start + buf->len - 1, GFP_NOFS);
5926         }
5927         trans->blocks_used++;
5928         /* this returns a buffer locked for blocking */
5929         return buf;
5930 }
5931
5932 static struct btrfs_block_rsv *
5933 use_block_rsv(struct btrfs_trans_handle *trans,
5934               struct btrfs_root *root, u32 blocksize)
5935 {
5936         struct btrfs_block_rsv *block_rsv;
5937         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5938         int ret;
5939
5940         block_rsv = get_block_rsv(trans, root);
5941
5942         if (block_rsv->size == 0) {
5943                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5944                 /*
5945                  * If we couldn't reserve metadata bytes try and use some from
5946                  * the global reserve.
5947                  */
5948                 if (ret && block_rsv != global_rsv) {
5949                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5950                         if (!ret)
5951                                 return global_rsv;
5952                         return ERR_PTR(ret);
5953                 } else if (ret) {
5954                         return ERR_PTR(ret);
5955                 }
5956                 return block_rsv;
5957         }
5958
5959         ret = block_rsv_use_bytes(block_rsv, blocksize);
5960         if (!ret)
5961                 return block_rsv;
5962         if (ret) {
5963                 static DEFINE_RATELIMIT_STATE(_rs,
5964                                 DEFAULT_RATELIMIT_INTERVAL,
5965                                 /*DEFAULT_RATELIMIT_BURST*/ 2);
5966                 if (__ratelimit(&_rs)) {
5967                         printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
5968                         WARN_ON(1);
5969                 }
5970                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5971                 if (!ret) {
5972                         return block_rsv;
5973                 } else if (ret && block_rsv != global_rsv) {
5974                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5975                         if (!ret)
5976                                 return global_rsv;
5977                 }
5978         }
5979
5980         return ERR_PTR(-ENOSPC);
5981 }
5982
5983 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5984 {
5985         block_rsv_add_bytes(block_rsv, blocksize, 0);
5986         block_rsv_release_bytes(block_rsv, NULL, 0);
5987 }
5988
5989 /*
5990  * finds a free extent and does all the dirty work required for allocation
5991  * returns the key for the extent through ins, and a tree buffer for
5992  * the first block of the extent through buf.
5993  *
5994  * returns the tree buffer or NULL.
5995  */
5996 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5997                                         struct btrfs_root *root, u32 blocksize,
5998                                         u64 parent, u64 root_objectid,
5999                                         struct btrfs_disk_key *key, int level,
6000                                         u64 hint, u64 empty_size)
6001 {
6002         struct btrfs_key ins;
6003         struct btrfs_block_rsv *block_rsv;
6004         struct extent_buffer *buf;
6005         u64 flags = 0;
6006         int ret;
6007
6008
6009         block_rsv = use_block_rsv(trans, root, blocksize);
6010         if (IS_ERR(block_rsv))
6011                 return ERR_CAST(block_rsv);
6012
6013         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6014                                    empty_size, hint, (u64)-1, &ins, 0);
6015         if (ret) {
6016                 unuse_block_rsv(block_rsv, blocksize);
6017                 return ERR_PTR(ret);
6018         }
6019
6020         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6021                                     blocksize, level);
6022         BUG_ON(IS_ERR(buf));
6023
6024         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6025                 if (parent == 0)
6026                         parent = ins.objectid;
6027                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6028         } else
6029                 BUG_ON(parent > 0);
6030
6031         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6032                 struct btrfs_delayed_extent_op *extent_op;
6033                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6034                 BUG_ON(!extent_op);
6035                 if (key)
6036                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
6037                 else
6038                         memset(&extent_op->key, 0, sizeof(extent_op->key));
6039                 extent_op->flags_to_set = flags;
6040                 extent_op->update_key = 1;
6041                 extent_op->update_flags = 1;
6042                 extent_op->is_data = 0;
6043
6044                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
6045                                         ins.offset, parent, root_objectid,
6046                                         level, BTRFS_ADD_DELAYED_EXTENT,
6047                                         extent_op);
6048                 BUG_ON(ret);
6049         }
6050         return buf;
6051 }
6052
6053 struct walk_control {
6054         u64 refs[BTRFS_MAX_LEVEL];
6055         u64 flags[BTRFS_MAX_LEVEL];
6056         struct btrfs_key update_progress;
6057         int stage;
6058         int level;
6059         int shared_level;
6060         int update_ref;
6061         int keep_locks;
6062         int reada_slot;
6063         int reada_count;
6064 };
6065
6066 #define DROP_REFERENCE  1
6067 #define UPDATE_BACKREF  2
6068
6069 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6070                                      struct btrfs_root *root,
6071                                      struct walk_control *wc,
6072                                      struct btrfs_path *path)
6073 {
6074         u64 bytenr;
6075         u64 generation;
6076         u64 refs;
6077         u64 flags;
6078         u32 nritems;
6079         u32 blocksize;
6080         struct btrfs_key key;
6081         struct extent_buffer *eb;
6082         int ret;
6083         int slot;
6084         int nread = 0;
6085
6086         if (path->slots[wc->level] < wc->reada_slot) {
6087                 wc->reada_count = wc->reada_count * 2 / 3;
6088                 wc->reada_count = max(wc->reada_count, 2);
6089         } else {
6090                 wc->reada_count = wc->reada_count * 3 / 2;
6091                 wc->reada_count = min_t(int, wc->reada_count,
6092                                         BTRFS_NODEPTRS_PER_BLOCK(root));
6093         }
6094
6095         eb = path->nodes[wc->level];
6096         nritems = btrfs_header_nritems(eb);
6097         blocksize = btrfs_level_size(root, wc->level - 1);
6098
6099         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6100                 if (nread >= wc->reada_count)
6101                         break;
6102
6103                 cond_resched();
6104                 bytenr = btrfs_node_blockptr(eb, slot);
6105                 generation = btrfs_node_ptr_generation(eb, slot);
6106
6107                 if (slot == path->slots[wc->level])
6108                         goto reada;
6109
6110                 if (wc->stage == UPDATE_BACKREF &&
6111                     generation <= root->root_key.offset)
6112                         continue;
6113
6114                 /* We don't lock the tree block, it's OK to be racy here */
6115                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6116                                                &refs, &flags);
6117                 BUG_ON(ret);
6118                 BUG_ON(refs == 0);
6119
6120                 if (wc->stage == DROP_REFERENCE) {
6121                         if (refs == 1)
6122                                 goto reada;
6123
6124                         if (wc->level == 1 &&
6125                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6126                                 continue;
6127                         if (!wc->update_ref ||
6128                             generation <= root->root_key.offset)
6129                                 continue;
6130                         btrfs_node_key_to_cpu(eb, &key, slot);
6131                         ret = btrfs_comp_cpu_keys(&key,
6132                                                   &wc->update_progress);
6133                         if (ret < 0)
6134                                 continue;
6135                 } else {
6136                         if (wc->level == 1 &&
6137                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6138                                 continue;
6139                 }
6140 reada:
6141                 ret = readahead_tree_block(root, bytenr, blocksize,
6142                                            generation);
6143                 if (ret)
6144                         break;
6145                 nread++;
6146         }
6147         wc->reada_slot = slot;
6148 }
6149
6150 /*
6151  * hepler to process tree block while walking down the tree.
6152  *
6153  * when wc->stage == UPDATE_BACKREF, this function updates
6154  * back refs for pointers in the block.
6155  *
6156  * NOTE: return value 1 means we should stop walking down.
6157  */
6158 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6159                                    struct btrfs_root *root,
6160                                    struct btrfs_path *path,
6161                                    struct walk_control *wc, int lookup_info)
6162 {
6163         int level = wc->level;
6164         struct extent_buffer *eb = path->nodes[level];
6165         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6166         int ret;
6167
6168         if (wc->stage == UPDATE_BACKREF &&
6169             btrfs_header_owner(eb) != root->root_key.objectid)
6170                 return 1;
6171
6172         /*
6173          * when reference count of tree block is 1, it won't increase
6174          * again. once full backref flag is set, we never clear it.
6175          */
6176         if (lookup_info &&
6177             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6178              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6179                 BUG_ON(!path->locks[level]);
6180                 ret = btrfs_lookup_extent_info(trans, root,
6181                                                eb->start, eb->len,
6182                                                &wc->refs[level],
6183                                                &wc->flags[level]);
6184                 BUG_ON(ret);
6185                 BUG_ON(wc->refs[level] == 0);
6186         }
6187
6188         if (wc->stage == DROP_REFERENCE) {
6189                 if (wc->refs[level] > 1)
6190                         return 1;
6191
6192                 if (path->locks[level] && !wc->keep_locks) {
6193                         btrfs_tree_unlock_rw(eb, path->locks[level]);
6194                         path->locks[level] = 0;
6195                 }
6196                 return 0;
6197         }
6198
6199         /* wc->stage == UPDATE_BACKREF */
6200         if (!(wc->flags[level] & flag)) {
6201                 BUG_ON(!path->locks[level]);
6202                 ret = btrfs_inc_ref(trans, root, eb, 1);
6203                 BUG_ON(ret);
6204                 ret = btrfs_dec_ref(trans, root, eb, 0);
6205                 BUG_ON(ret);
6206                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6207                                                   eb->len, flag, 0);
6208                 BUG_ON(ret);
6209                 wc->flags[level] |= flag;
6210         }
6211
6212         /*
6213          * the block is shared by multiple trees, so it's not good to
6214          * keep the tree lock
6215          */
6216         if (path->locks[level] && level > 0) {
6217                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6218                 path->locks[level] = 0;
6219         }
6220         return 0;
6221 }
6222
6223 /*
6224  * hepler to process tree block pointer.
6225  *
6226  * when wc->stage == DROP_REFERENCE, this function checks
6227  * reference count of the block pointed to. if the block
6228  * is shared and we need update back refs for the subtree
6229  * rooted at the block, this function changes wc->stage to
6230  * UPDATE_BACKREF. if the block is shared and there is no
6231  * need to update back, this function drops the reference
6232  * to the block.
6233  *
6234  * NOTE: return value 1 means we should stop walking down.
6235  */
6236 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6237                                  struct btrfs_root *root,
6238                                  struct btrfs_path *path,
6239                                  struct walk_control *wc, int *lookup_info)
6240 {
6241         u64 bytenr;
6242         u64 generation;
6243         u64 parent;
6244         u32 blocksize;
6245         struct btrfs_key key;
6246         struct extent_buffer *next;
6247         int level = wc->level;
6248         int reada = 0;
6249         int ret = 0;
6250
6251         generation = btrfs_node_ptr_generation(path->nodes[level],
6252                                                path->slots[level]);
6253         /*
6254          * if the lower level block was created before the snapshot
6255          * was created, we know there is no need to update back refs
6256          * for the subtree
6257          */
6258         if (wc->stage == UPDATE_BACKREF &&
6259             generation <= root->root_key.offset) {
6260                 *lookup_info = 1;
6261                 return 1;
6262         }
6263
6264         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6265         blocksize = btrfs_level_size(root, level - 1);
6266
6267         next = btrfs_find_tree_block(root, bytenr, blocksize);
6268         if (!next) {
6269                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6270                 if (!next)
6271                         return -ENOMEM;
6272                 reada = 1;
6273         }
6274         btrfs_tree_lock(next);
6275         btrfs_set_lock_blocking(next);
6276
6277         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6278                                        &wc->refs[level - 1],
6279                                        &wc->flags[level - 1]);
6280         BUG_ON(ret);
6281         BUG_ON(wc->refs[level - 1] == 0);
6282         *lookup_info = 0;
6283
6284         if (wc->stage == DROP_REFERENCE) {
6285                 if (wc->refs[level - 1] > 1) {
6286                         if (level == 1 &&
6287                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6288                                 goto skip;
6289
6290                         if (!wc->update_ref ||
6291                             generation <= root->root_key.offset)
6292                                 goto skip;
6293
6294                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6295                                               path->slots[level]);
6296                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6297                         if (ret < 0)
6298                                 goto skip;
6299
6300                         wc->stage = UPDATE_BACKREF;
6301                         wc->shared_level = level - 1;
6302                 }
6303         } else {
6304                 if (level == 1 &&
6305                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6306                         goto skip;
6307         }
6308
6309         if (!btrfs_buffer_uptodate(next, generation)) {
6310                 btrfs_tree_unlock(next);
6311                 free_extent_buffer(next);
6312                 next = NULL;
6313                 *lookup_info = 1;
6314         }
6315
6316         if (!next) {
6317                 if (reada && level == 1)
6318                         reada_walk_down(trans, root, wc, path);
6319                 next = read_tree_block(root, bytenr, blocksize, generation);
6320                 if (!next)
6321                         return -EIO;
6322                 btrfs_tree_lock(next);
6323                 btrfs_set_lock_blocking(next);
6324         }
6325
6326         level--;
6327         BUG_ON(level != btrfs_header_level(next));
6328         path->nodes[level] = next;
6329         path->slots[level] = 0;
6330         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6331         wc->level = level;
6332         if (wc->level == 1)
6333                 wc->reada_slot = 0;
6334         return 0;
6335 skip:
6336         wc->refs[level - 1] = 0;
6337         wc->flags[level - 1] = 0;
6338         if (wc->stage == DROP_REFERENCE) {
6339                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6340                         parent = path->nodes[level]->start;
6341                 } else {
6342                         BUG_ON(root->root_key.objectid !=
6343                                btrfs_header_owner(path->nodes[level]));
6344                         parent = 0;
6345                 }
6346
6347                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6348                                         root->root_key.objectid, level - 1, 0);
6349                 BUG_ON(ret);
6350         }
6351         btrfs_tree_unlock(next);
6352         free_extent_buffer(next);
6353         *lookup_info = 1;
6354         return 1;
6355 }
6356
6357 /*
6358  * hepler to process tree block while walking up the tree.
6359  *
6360  * when wc->stage == DROP_REFERENCE, this function drops
6361  * reference count on the block.
6362  *
6363  * when wc->stage == UPDATE_BACKREF, this function changes
6364  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6365  * to UPDATE_BACKREF previously while processing the block.
6366  *
6367  * NOTE: return value 1 means we should stop walking up.
6368  */
6369 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6370                                  struct btrfs_root *root,
6371                                  struct btrfs_path *path,
6372                                  struct walk_control *wc)
6373 {
6374         int ret;
6375         int level = wc->level;
6376         struct extent_buffer *eb = path->nodes[level];
6377         u64 parent = 0;
6378
6379         if (wc->stage == UPDATE_BACKREF) {
6380                 BUG_ON(wc->shared_level < level);
6381                 if (level < wc->shared_level)
6382                         goto out;
6383
6384                 ret = find_next_key(path, level + 1, &wc->update_progress);
6385                 if (ret > 0)
6386                         wc->update_ref = 0;
6387
6388                 wc->stage = DROP_REFERENCE;
6389                 wc->shared_level = -1;
6390                 path->slots[level] = 0;
6391
6392                 /*
6393                  * check reference count again if the block isn't locked.
6394                  * we should start walking down the tree again if reference
6395                  * count is one.
6396                  */
6397                 if (!path->locks[level]) {
6398                         BUG_ON(level == 0);
6399                         btrfs_tree_lock(eb);
6400                         btrfs_set_lock_blocking(eb);
6401                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6402
6403                         ret = btrfs_lookup_extent_info(trans, root,
6404                                                        eb->start, eb->len,
6405                                                        &wc->refs[level],
6406                                                        &wc->flags[level]);
6407                         BUG_ON(ret);
6408                         BUG_ON(wc->refs[level] == 0);
6409                         if (wc->refs[level] == 1) {
6410                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6411                                 return 1;
6412                         }
6413                 }
6414         }
6415
6416         /* wc->stage == DROP_REFERENCE */
6417         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6418
6419         if (wc->refs[level] == 1) {
6420                 if (level == 0) {
6421                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6422                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6423                         else
6424                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6425                         BUG_ON(ret);
6426                 }
6427                 /* make block locked assertion in clean_tree_block happy */
6428                 if (!path->locks[level] &&
6429                     btrfs_header_generation(eb) == trans->transid) {
6430                         btrfs_tree_lock(eb);
6431                         btrfs_set_lock_blocking(eb);
6432                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6433                 }
6434                 clean_tree_block(trans, root, eb);
6435         }
6436
6437         if (eb == root->node) {
6438                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6439                         parent = eb->start;
6440                 else
6441                         BUG_ON(root->root_key.objectid !=
6442                                btrfs_header_owner(eb));
6443         } else {
6444                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6445                         parent = path->nodes[level + 1]->start;
6446                 else
6447                         BUG_ON(root->root_key.objectid !=
6448                                btrfs_header_owner(path->nodes[level + 1]));
6449         }
6450
6451         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6452 out:
6453         wc->refs[level] = 0;
6454         wc->flags[level] = 0;
6455         return 0;
6456 }
6457
6458 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6459                                    struct btrfs_root *root,
6460                                    struct btrfs_path *path,
6461                                    struct walk_control *wc)
6462 {
6463         int level = wc->level;
6464         int lookup_info = 1;
6465         int ret;
6466
6467         while (level >= 0) {
6468                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6469                 if (ret > 0)
6470                         break;
6471
6472                 if (level == 0)
6473                         break;
6474
6475                 if (path->slots[level] >=
6476                     btrfs_header_nritems(path->nodes[level]))
6477                         break;
6478
6479                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6480                 if (ret > 0) {
6481                         path->slots[level]++;
6482                         continue;
6483                 } else if (ret < 0)
6484                         return ret;
6485                 level = wc->level;
6486         }
6487         return 0;
6488 }
6489
6490 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6491                                  struct btrfs_root *root,
6492                                  struct btrfs_path *path,
6493                                  struct walk_control *wc, int max_level)
6494 {
6495         int level = wc->level;
6496         int ret;
6497
6498         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6499         while (level < max_level && path->nodes[level]) {
6500                 wc->level = level;
6501                 if (path->slots[level] + 1 <
6502                     btrfs_header_nritems(path->nodes[level])) {
6503                         path->slots[level]++;
6504                         return 0;
6505                 } else {
6506                         ret = walk_up_proc(trans, root, path, wc);
6507                         if (ret > 0)
6508                                 return 0;
6509
6510                         if (path->locks[level]) {
6511                                 btrfs_tree_unlock_rw(path->nodes[level],
6512                                                      path->locks[level]);
6513                                 path->locks[level] = 0;
6514                         }
6515                         free_extent_buffer(path->nodes[level]);
6516                         path->nodes[level] = NULL;
6517                         level++;
6518                 }
6519         }
6520         return 1;
6521 }
6522
6523 /*
6524  * drop a subvolume tree.
6525  *
6526  * this function traverses the tree freeing any blocks that only
6527  * referenced by the tree.
6528  *
6529  * when a shared tree block is found. this function decreases its
6530  * reference count by one. if update_ref is true, this function
6531  * also make sure backrefs for the shared block and all lower level
6532  * blocks are properly updated.
6533  */
6534 void btrfs_drop_snapshot(struct btrfs_root *root,
6535                          struct btrfs_block_rsv *block_rsv, int update_ref)
6536 {
6537         struct btrfs_path *path;
6538         struct btrfs_trans_handle *trans;
6539         struct btrfs_root *tree_root = root->fs_info->tree_root;
6540         struct btrfs_root_item *root_item = &root->root_item;
6541         struct walk_control *wc;
6542         struct btrfs_key key;
6543         int err = 0;
6544         int ret;
6545         int level;
6546
6547         path = btrfs_alloc_path();
6548         if (!path) {
6549                 err = -ENOMEM;
6550                 goto out;
6551         }
6552
6553         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6554         if (!wc) {
6555                 btrfs_free_path(path);
6556                 err = -ENOMEM;
6557                 goto out;
6558         }
6559
6560         trans = btrfs_start_transaction(tree_root, 0);
6561         BUG_ON(IS_ERR(trans));
6562
6563         if (block_rsv)
6564                 trans->block_rsv = block_rsv;
6565
6566         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6567                 level = btrfs_header_level(root->node);
6568                 path->nodes[level] = btrfs_lock_root_node(root);
6569                 btrfs_set_lock_blocking(path->nodes[level]);
6570                 path->slots[level] = 0;
6571                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6572                 memset(&wc->update_progress, 0,
6573                        sizeof(wc->update_progress));
6574         } else {
6575                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6576                 memcpy(&wc->update_progress, &key,
6577                        sizeof(wc->update_progress));
6578
6579                 level = root_item->drop_level;
6580                 BUG_ON(level == 0);
6581                 path->lowest_level = level;
6582                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6583                 path->lowest_level = 0;
6584                 if (ret < 0) {
6585                         err = ret;
6586                         goto out_free;
6587                 }
6588                 WARN_ON(ret > 0);
6589
6590                 /*
6591                  * unlock our path, this is safe because only this
6592                  * function is allowed to delete this snapshot
6593                  */
6594                 btrfs_unlock_up_safe(path, 0);
6595
6596                 level = btrfs_header_level(root->node);
6597                 while (1) {
6598                         btrfs_tree_lock(path->nodes[level]);
6599                         btrfs_set_lock_blocking(path->nodes[level]);
6600
6601                         ret = btrfs_lookup_extent_info(trans, root,
6602                                                 path->nodes[level]->start,
6603                                                 path->nodes[level]->len,
6604                                                 &wc->refs[level],
6605                                                 &wc->flags[level]);
6606                         BUG_ON(ret);
6607                         BUG_ON(wc->refs[level] == 0);
6608
6609                         if (level == root_item->drop_level)
6610                                 break;
6611
6612                         btrfs_tree_unlock(path->nodes[level]);
6613                         WARN_ON(wc->refs[level] != 1);
6614                         level--;
6615                 }
6616         }
6617
6618         wc->level = level;
6619         wc->shared_level = -1;
6620         wc->stage = DROP_REFERENCE;
6621         wc->update_ref = update_ref;
6622         wc->keep_locks = 0;
6623         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6624
6625         while (1) {
6626                 ret = walk_down_tree(trans, root, path, wc);
6627                 if (ret < 0) {
6628                         err = ret;
6629                         break;
6630                 }
6631
6632                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6633                 if (ret < 0) {
6634                         err = ret;
6635                         break;
6636                 }
6637
6638                 if (ret > 0) {
6639                         BUG_ON(wc->stage != DROP_REFERENCE);
6640                         break;
6641                 }
6642
6643                 if (wc->stage == DROP_REFERENCE) {
6644                         level = wc->level;
6645                         btrfs_node_key(path->nodes[level],
6646                                        &root_item->drop_progress,
6647                                        path->slots[level]);
6648                         root_item->drop_level = level;
6649                 }
6650
6651                 BUG_ON(wc->level == 0);
6652                 if (btrfs_should_end_transaction(trans, tree_root)) {
6653                         ret = btrfs_update_root(trans, tree_root,
6654                                                 &root->root_key,
6655                                                 root_item);
6656                         BUG_ON(ret);
6657
6658                         btrfs_end_transaction_throttle(trans, tree_root);
6659                         trans = btrfs_start_transaction(tree_root, 0);
6660                         BUG_ON(IS_ERR(trans));
6661                         if (block_rsv)
6662                                 trans->block_rsv = block_rsv;
6663                 }
6664         }
6665         btrfs_release_path(path);
6666         BUG_ON(err);
6667
6668         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6669         BUG_ON(ret);
6670
6671         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6672                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6673                                            NULL, NULL);
6674                 BUG_ON(ret < 0);
6675                 if (ret > 0) {
6676                         /* if we fail to delete the orphan item this time
6677                          * around, it'll get picked up the next time.
6678                          *
6679                          * The most common failure here is just -ENOENT.
6680                          */
6681                         btrfs_del_orphan_item(trans, tree_root,
6682                                               root->root_key.objectid);
6683                 }
6684         }
6685
6686         if (root->in_radix) {
6687                 btrfs_free_fs_root(tree_root->fs_info, root);
6688         } else {
6689                 free_extent_buffer(root->node);
6690                 free_extent_buffer(root->commit_root);
6691                 kfree(root);
6692         }
6693 out_free:
6694         btrfs_end_transaction_throttle(trans, tree_root);
6695         kfree(wc);
6696         btrfs_free_path(path);
6697 out:
6698         if (err)
6699                 btrfs_std_error(root->fs_info, err);
6700         return;
6701 }
6702
6703 /*
6704  * drop subtree rooted at tree block 'node'.
6705  *
6706  * NOTE: this function will unlock and release tree block 'node'
6707  */
6708 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6709                         struct btrfs_root *root,
6710                         struct extent_buffer *node,
6711                         struct extent_buffer *parent)
6712 {
6713         struct btrfs_path *path;
6714         struct walk_control *wc;
6715         int level;
6716         int parent_level;
6717         int ret = 0;
6718         int wret;
6719
6720         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6721
6722         path = btrfs_alloc_path();
6723         if (!path)
6724                 return -ENOMEM;
6725
6726         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6727         if (!wc) {
6728                 btrfs_free_path(path);
6729                 return -ENOMEM;
6730         }
6731
6732         btrfs_assert_tree_locked(parent);
6733         parent_level = btrfs_header_level(parent);
6734         extent_buffer_get(parent);
6735         path->nodes[parent_level] = parent;
6736         path->slots[parent_level] = btrfs_header_nritems(parent);
6737
6738         btrfs_assert_tree_locked(node);
6739         level = btrfs_header_level(node);
6740         path->nodes[level] = node;
6741         path->slots[level] = 0;
6742         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6743
6744         wc->refs[parent_level] = 1;
6745         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6746         wc->level = level;
6747         wc->shared_level = -1;
6748         wc->stage = DROP_REFERENCE;
6749         wc->update_ref = 0;
6750         wc->keep_locks = 1;
6751         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6752
6753         while (1) {
6754                 wret = walk_down_tree(trans, root, path, wc);
6755                 if (wret < 0) {
6756                         ret = wret;
6757                         break;
6758                 }
6759
6760                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6761                 if (wret < 0)
6762                         ret = wret;
6763                 if (wret != 0)
6764                         break;
6765         }
6766
6767         kfree(wc);
6768         btrfs_free_path(path);
6769         return ret;
6770 }
6771
6772 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6773 {
6774         u64 num_devices;
6775         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6776                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6777
6778         /*
6779          * we add in the count of missing devices because we want
6780          * to make sure that any RAID levels on a degraded FS
6781          * continue to be honored.
6782          */
6783         num_devices = root->fs_info->fs_devices->rw_devices +
6784                 root->fs_info->fs_devices->missing_devices;
6785
6786         if (num_devices == 1) {
6787                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6788                 stripped = flags & ~stripped;
6789
6790                 /* turn raid0 into single device chunks */
6791                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6792                         return stripped;
6793
6794                 /* turn mirroring into duplication */
6795                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6796                              BTRFS_BLOCK_GROUP_RAID10))
6797                         return stripped | BTRFS_BLOCK_GROUP_DUP;
6798                 return flags;
6799         } else {
6800                 /* they already had raid on here, just return */
6801                 if (flags & stripped)
6802                         return flags;
6803
6804                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6805                 stripped = flags & ~stripped;
6806
6807                 /* switch duplicated blocks with raid1 */
6808                 if (flags & BTRFS_BLOCK_GROUP_DUP)
6809                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
6810
6811                 /* turn single device chunks into raid0 */
6812                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6813         }
6814         return flags;
6815 }
6816
6817 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6818 {
6819         struct btrfs_space_info *sinfo = cache->space_info;
6820         u64 num_bytes;
6821         u64 min_allocable_bytes;
6822         int ret = -ENOSPC;
6823
6824
6825         /*
6826          * We need some metadata space and system metadata space for
6827          * allocating chunks in some corner cases until we force to set
6828          * it to be readonly.
6829          */
6830         if ((sinfo->flags &
6831              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6832             !force)
6833                 min_allocable_bytes = 1 * 1024 * 1024;
6834         else
6835                 min_allocable_bytes = 0;
6836
6837         spin_lock(&sinfo->lock);
6838         spin_lock(&cache->lock);
6839
6840         if (cache->ro) {
6841                 ret = 0;
6842                 goto out;
6843         }
6844
6845         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6846                     cache->bytes_super - btrfs_block_group_used(&cache->item);
6847
6848         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6849             sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6850             min_allocable_bytes <= sinfo->total_bytes) {
6851                 sinfo->bytes_readonly += num_bytes;
6852                 cache->ro = 1;
6853                 ret = 0;
6854         }
6855 out:
6856         spin_unlock(&cache->lock);
6857         spin_unlock(&sinfo->lock);
6858         return ret;
6859 }
6860
6861 int btrfs_set_block_group_ro(struct btrfs_root *root,
6862                              struct btrfs_block_group_cache *cache)
6863
6864 {
6865         struct btrfs_trans_handle *trans;
6866         u64 alloc_flags;
6867         int ret;
6868
6869         BUG_ON(cache->ro);
6870
6871         trans = btrfs_join_transaction(root);
6872         BUG_ON(IS_ERR(trans));
6873
6874         alloc_flags = update_block_group_flags(root, cache->flags);
6875         if (alloc_flags != cache->flags)
6876                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6877                                CHUNK_ALLOC_FORCE);
6878
6879         ret = set_block_group_ro(cache, 0);
6880         if (!ret)
6881                 goto out;
6882         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6883         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6884                              CHUNK_ALLOC_FORCE);
6885         if (ret < 0)
6886                 goto out;
6887         ret = set_block_group_ro(cache, 0);
6888 out:
6889         btrfs_end_transaction(trans, root);
6890         return ret;
6891 }
6892
6893 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6894                             struct btrfs_root *root, u64 type)
6895 {
6896         u64 alloc_flags = get_alloc_profile(root, type);
6897         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6898                               CHUNK_ALLOC_FORCE);
6899 }
6900
6901 /*
6902  * helper to account the unused space of all the readonly block group in the
6903  * list. takes mirrors into account.
6904  */
6905 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6906 {
6907         struct btrfs_block_group_cache *block_group;
6908         u64 free_bytes = 0;
6909         int factor;
6910
6911         list_for_each_entry(block_group, groups_list, list) {
6912                 spin_lock(&block_group->lock);
6913
6914                 if (!block_group->ro) {
6915                         spin_unlock(&block_group->lock);
6916                         continue;
6917                 }
6918
6919                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6920                                           BTRFS_BLOCK_GROUP_RAID10 |
6921                                           BTRFS_BLOCK_GROUP_DUP))
6922                         factor = 2;
6923                 else
6924                         factor = 1;
6925
6926                 free_bytes += (block_group->key.offset -
6927                                btrfs_block_group_used(&block_group->item)) *
6928                                factor;
6929
6930                 spin_unlock(&block_group->lock);
6931         }
6932
6933         return free_bytes;
6934 }
6935
6936 /*
6937  * helper to account the unused space of all the readonly block group in the
6938  * space_info. takes mirrors into account.
6939  */
6940 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6941 {
6942         int i;
6943         u64 free_bytes = 0;
6944
6945         spin_lock(&sinfo->lock);
6946
6947         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6948                 if (!list_empty(&sinfo->block_groups[i]))
6949                         free_bytes += __btrfs_get_ro_block_group_free_space(
6950                                                 &sinfo->block_groups[i]);
6951
6952         spin_unlock(&sinfo->lock);
6953
6954         return free_bytes;
6955 }
6956
6957 int btrfs_set_block_group_rw(struct btrfs_root *root,
6958                               struct btrfs_block_group_cache *cache)
6959 {
6960         struct btrfs_space_info *sinfo = cache->space_info;
6961         u64 num_bytes;
6962
6963         BUG_ON(!cache->ro);
6964
6965         spin_lock(&sinfo->lock);
6966         spin_lock(&cache->lock);
6967         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6968                     cache->bytes_super - btrfs_block_group_used(&cache->item);
6969         sinfo->bytes_readonly -= num_bytes;
6970         cache->ro = 0;
6971         spin_unlock(&cache->lock);
6972         spin_unlock(&sinfo->lock);
6973         return 0;
6974 }
6975
6976 /*
6977  * checks to see if its even possible to relocate this block group.
6978  *
6979  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6980  * ok to go ahead and try.
6981  */
6982 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6983 {
6984         struct btrfs_block_group_cache *block_group;
6985         struct btrfs_space_info *space_info;
6986         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6987         struct btrfs_device *device;
6988         u64 min_free;
6989         u64 dev_min = 1;
6990         u64 dev_nr = 0;
6991         int index;
6992         int full = 0;
6993         int ret = 0;
6994
6995         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6996
6997         /* odd, couldn't find the block group, leave it alone */
6998         if (!block_group)
6999                 return -1;
7000
7001         min_free = btrfs_block_group_used(&block_group->item);
7002
7003         /* no bytes used, we're good */
7004         if (!min_free)
7005                 goto out;
7006
7007         space_info = block_group->space_info;
7008         spin_lock(&space_info->lock);
7009
7010         full = space_info->full;
7011
7012         /*
7013          * if this is the last block group we have in this space, we can't
7014          * relocate it unless we're able to allocate a new chunk below.
7015          *
7016          * Otherwise, we need to make sure we have room in the space to handle
7017          * all of the extents from this block group.  If we can, we're good
7018          */
7019         if ((space_info->total_bytes != block_group->key.offset) &&
7020             (space_info->bytes_used + space_info->bytes_reserved +
7021              space_info->bytes_pinned + space_info->bytes_readonly +
7022              min_free < space_info->total_bytes)) {
7023                 spin_unlock(&space_info->lock);
7024                 goto out;
7025         }
7026         spin_unlock(&space_info->lock);
7027
7028         /*
7029          * ok we don't have enough space, but maybe we have free space on our
7030          * devices to allocate new chunks for relocation, so loop through our
7031          * alloc devices and guess if we have enough space.  However, if we
7032          * were marked as full, then we know there aren't enough chunks, and we
7033          * can just return.
7034          */
7035         ret = -1;
7036         if (full)
7037                 goto out;
7038
7039         /*
7040          * index:
7041          *      0: raid10
7042          *      1: raid1
7043          *      2: dup
7044          *      3: raid0
7045          *      4: single
7046          */
7047         index = get_block_group_index(block_group);
7048         if (index == 0) {
7049                 dev_min = 4;
7050                 /* Divide by 2 */
7051                 min_free >>= 1;
7052         } else if (index == 1) {
7053                 dev_min = 2;
7054         } else if (index == 2) {
7055                 /* Multiply by 2 */
7056                 min_free <<= 1;
7057         } else if (index == 3) {
7058                 dev_min = fs_devices->rw_devices;
7059                 do_div(min_free, dev_min);
7060         }
7061
7062         mutex_lock(&root->fs_info->chunk_mutex);
7063         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7064                 u64 dev_offset;
7065
7066                 /*
7067                  * check to make sure we can actually find a chunk with enough
7068                  * space to fit our block group in.
7069                  */
7070                 if (device->total_bytes > device->bytes_used + min_free) {
7071                         ret = find_free_dev_extent(NULL, device, min_free,
7072                                                    &dev_offset, NULL);
7073                         if (!ret)
7074                                 dev_nr++;
7075
7076                         if (dev_nr >= dev_min)
7077                                 break;
7078
7079                         ret = -1;
7080                 }
7081         }
7082         mutex_unlock(&root->fs_info->chunk_mutex);
7083 out:
7084         btrfs_put_block_group(block_group);
7085         return ret;
7086 }
7087
7088 static int find_first_block_group(struct btrfs_root *root,
7089                 struct btrfs_path *path, struct btrfs_key *key)
7090 {
7091         int ret = 0;
7092         struct btrfs_key found_key;
7093         struct extent_buffer *leaf;
7094         int slot;
7095
7096         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7097         if (ret < 0)
7098                 goto out;
7099
7100         while (1) {
7101                 slot = path->slots[0];
7102                 leaf = path->nodes[0];
7103                 if (slot >= btrfs_header_nritems(leaf)) {
7104                         ret = btrfs_next_leaf(root, path);
7105                         if (ret == 0)
7106                                 continue;
7107                         if (ret < 0)
7108                                 goto out;
7109                         break;
7110                 }
7111                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7112
7113                 if (found_key.objectid >= key->objectid &&
7114                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7115                         ret = 0;
7116                         goto out;
7117                 }
7118                 path->slots[0]++;
7119         }
7120 out:
7121         return ret;
7122 }
7123
7124 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7125 {
7126         struct btrfs_block_group_cache *block_group;
7127         u64 last = 0;
7128
7129         while (1) {
7130                 struct inode *inode;
7131
7132                 block_group = btrfs_lookup_first_block_group(info, last);
7133                 while (block_group) {
7134                         spin_lock(&block_group->lock);
7135                         if (block_group->iref)
7136                                 break;
7137                         spin_unlock(&block_group->lock);
7138                         block_group = next_block_group(info->tree_root,
7139                                                        block_group);
7140                 }
7141                 if (!block_group) {
7142                         if (last == 0)
7143                                 break;
7144                         last = 0;
7145                         continue;
7146                 }
7147
7148                 inode = block_group->inode;
7149                 block_group->iref = 0;
7150                 block_group->inode = NULL;
7151                 spin_unlock(&block_group->lock);
7152                 iput(inode);
7153                 last = block_group->key.objectid + block_group->key.offset;
7154                 btrfs_put_block_group(block_group);
7155         }
7156 }
7157
7158 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7159 {
7160         struct btrfs_block_group_cache *block_group;
7161         struct btrfs_space_info *space_info;
7162         struct btrfs_caching_control *caching_ctl;
7163         struct rb_node *n;
7164
7165         down_write(&info->extent_commit_sem);
7166         while (!list_empty(&info->caching_block_groups)) {
7167                 caching_ctl = list_entry(info->caching_block_groups.next,
7168                                          struct btrfs_caching_control, list);
7169                 list_del(&caching_ctl->list);
7170                 put_caching_control(caching_ctl);
7171         }
7172         up_write(&info->extent_commit_sem);
7173
7174         spin_lock(&info->block_group_cache_lock);
7175         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7176                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7177                                        cache_node);
7178                 rb_erase(&block_group->cache_node,
7179                          &info->block_group_cache_tree);
7180                 spin_unlock(&info->block_group_cache_lock);
7181
7182                 down_write(&block_group->space_info->groups_sem);
7183                 list_del(&block_group->list);
7184                 up_write(&block_group->space_info->groups_sem);
7185
7186                 if (block_group->cached == BTRFS_CACHE_STARTED)
7187                         wait_block_group_cache_done(block_group);
7188
7189                 /*
7190                  * We haven't cached this block group, which means we could
7191                  * possibly have excluded extents on this block group.
7192                  */
7193                 if (block_group->cached == BTRFS_CACHE_NO)
7194                         free_excluded_extents(info->extent_root, block_group);
7195
7196                 btrfs_remove_free_space_cache(block_group);
7197                 btrfs_put_block_group(block_group);
7198
7199                 spin_lock(&info->block_group_cache_lock);
7200         }
7201         spin_unlock(&info->block_group_cache_lock);
7202
7203         /* now that all the block groups are freed, go through and
7204          * free all the space_info structs.  This is only called during
7205          * the final stages of unmount, and so we know nobody is
7206          * using them.  We call synchronize_rcu() once before we start,
7207          * just to be on the safe side.
7208          */
7209         synchronize_rcu();
7210
7211         release_global_block_rsv(info);
7212
7213         while(!list_empty(&info->space_info)) {
7214                 space_info = list_entry(info->space_info.next,
7215                                         struct btrfs_space_info,
7216                                         list);
7217                 if (space_info->bytes_pinned > 0 ||
7218                     space_info->bytes_reserved > 0 ||
7219                     space_info->bytes_may_use > 0) {
7220                         WARN_ON(1);
7221                         dump_space_info(space_info, 0, 0);
7222                 }
7223                 list_del(&space_info->list);
7224                 kfree(space_info);
7225         }
7226         return 0;
7227 }
7228
7229 static void __link_block_group(struct btrfs_space_info *space_info,
7230                                struct btrfs_block_group_cache *cache)
7231 {
7232         int index = get_block_group_index(cache);
7233
7234         down_write(&space_info->groups_sem);
7235         list_add_tail(&cache->list, &space_info->block_groups[index]);
7236         up_write(&space_info->groups_sem);
7237 }
7238
7239 int btrfs_read_block_groups(struct btrfs_root *root)
7240 {
7241         struct btrfs_path *path;
7242         int ret;
7243         struct btrfs_block_group_cache *cache;
7244         struct btrfs_fs_info *info = root->fs_info;
7245         struct btrfs_space_info *space_info;
7246         struct btrfs_key key;
7247         struct btrfs_key found_key;
7248         struct extent_buffer *leaf;
7249         int need_clear = 0;
7250         u64 cache_gen;
7251
7252         root = info->extent_root;
7253         key.objectid = 0;
7254         key.offset = 0;
7255         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7256         path = btrfs_alloc_path();
7257         if (!path)
7258                 return -ENOMEM;
7259         path->reada = 1;
7260
7261         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7262         if (btrfs_test_opt(root, SPACE_CACHE) &&
7263             btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7264                 need_clear = 1;
7265         if (btrfs_test_opt(root, CLEAR_CACHE))
7266                 need_clear = 1;
7267
7268         while (1) {
7269                 ret = find_first_block_group(root, path, &key);
7270                 if (ret > 0)
7271                         break;
7272                 if (ret != 0)
7273                         goto error;
7274                 leaf = path->nodes[0];
7275                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7276                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7277                 if (!cache) {
7278                         ret = -ENOMEM;
7279                         goto error;
7280                 }
7281                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7282                                                 GFP_NOFS);
7283                 if (!cache->free_space_ctl) {
7284                         kfree(cache);
7285                         ret = -ENOMEM;
7286                         goto error;
7287                 }
7288
7289                 atomic_set(&cache->count, 1);
7290                 spin_lock_init(&cache->lock);
7291                 cache->fs_info = info;
7292                 INIT_LIST_HEAD(&cache->list);
7293                 INIT_LIST_HEAD(&cache->cluster_list);
7294
7295                 if (need_clear)
7296                         cache->disk_cache_state = BTRFS_DC_CLEAR;
7297
7298                 read_extent_buffer(leaf, &cache->item,
7299                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7300                                    sizeof(cache->item));
7301                 memcpy(&cache->key, &found_key, sizeof(found_key));
7302
7303                 key.objectid = found_key.objectid + found_key.offset;
7304                 btrfs_release_path(path);
7305                 cache->flags = btrfs_block_group_flags(&cache->item);
7306                 cache->sectorsize = root->sectorsize;
7307
7308                 btrfs_init_free_space_ctl(cache);
7309
7310                 /*
7311                  * We need to exclude the super stripes now so that the space
7312                  * info has super bytes accounted for, otherwise we'll think
7313                  * we have more space than we actually do.
7314                  */
7315                 exclude_super_stripes(root, cache);
7316
7317                 /*
7318                  * check for two cases, either we are full, and therefore
7319                  * don't need to bother with the caching work since we won't
7320                  * find any space, or we are empty, and we can just add all
7321                  * the space in and be done with it.  This saves us _alot_ of
7322                  * time, particularly in the full case.
7323                  */
7324                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7325                         cache->last_byte_to_unpin = (u64)-1;
7326                         cache->cached = BTRFS_CACHE_FINISHED;
7327                         free_excluded_extents(root, cache);
7328                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7329                         cache->last_byte_to_unpin = (u64)-1;
7330                         cache->cached = BTRFS_CACHE_FINISHED;
7331                         add_new_free_space(cache, root->fs_info,
7332                                            found_key.objectid,
7333                                            found_key.objectid +
7334                                            found_key.offset);
7335                         free_excluded_extents(root, cache);
7336                 }
7337
7338                 ret = update_space_info(info, cache->flags, found_key.offset,
7339                                         btrfs_block_group_used(&cache->item),
7340                                         &space_info);
7341                 BUG_ON(ret);
7342                 cache->space_info = space_info;
7343                 spin_lock(&cache->space_info->lock);
7344                 cache->space_info->bytes_readonly += cache->bytes_super;
7345                 spin_unlock(&cache->space_info->lock);
7346
7347                 __link_block_group(space_info, cache);
7348
7349                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7350                 BUG_ON(ret);
7351
7352                 set_avail_alloc_bits(root->fs_info, cache->flags);
7353                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7354                         set_block_group_ro(cache, 1);
7355         }
7356
7357         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7358                 if (!(get_alloc_profile(root, space_info->flags) &
7359                       (BTRFS_BLOCK_GROUP_RAID10 |
7360                        BTRFS_BLOCK_GROUP_RAID1 |
7361                        BTRFS_BLOCK_GROUP_DUP)))
7362                         continue;
7363                 /*
7364                  * avoid allocating from un-mirrored block group if there are
7365                  * mirrored block groups.
7366                  */
7367                 list_for_each_entry(cache, &space_info->block_groups[3], list)
7368                         set_block_group_ro(cache, 1);
7369                 list_for_each_entry(cache, &space_info->block_groups[4], list)
7370                         set_block_group_ro(cache, 1);
7371         }
7372
7373         init_global_block_rsv(info);
7374         ret = 0;
7375 error:
7376         btrfs_free_path(path);
7377         return ret;
7378 }
7379
7380 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7381                            struct btrfs_root *root, u64 bytes_used,
7382                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7383                            u64 size)
7384 {
7385         int ret;
7386         struct btrfs_root *extent_root;
7387         struct btrfs_block_group_cache *cache;
7388
7389         extent_root = root->fs_info->extent_root;
7390
7391         root->fs_info->last_trans_log_full_commit = trans->transid;
7392
7393         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7394         if (!cache)
7395                 return -ENOMEM;
7396         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7397                                         GFP_NOFS);
7398         if (!cache->free_space_ctl) {
7399                 kfree(cache);
7400                 return -ENOMEM;
7401         }
7402
7403         cache->key.objectid = chunk_offset;
7404         cache->key.offset = size;
7405         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7406         cache->sectorsize = root->sectorsize;
7407         cache->fs_info = root->fs_info;
7408
7409         atomic_set(&cache->count, 1);
7410         spin_lock_init(&cache->lock);
7411         INIT_LIST_HEAD(&cache->list);
7412         INIT_LIST_HEAD(&cache->cluster_list);
7413
7414         btrfs_init_free_space_ctl(cache);
7415
7416         btrfs_set_block_group_used(&cache->item, bytes_used);
7417         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7418         cache->flags = type;
7419         btrfs_set_block_group_flags(&cache->item, type);
7420
7421         cache->last_byte_to_unpin = (u64)-1;
7422         cache->cached = BTRFS_CACHE_FINISHED;
7423         exclude_super_stripes(root, cache);
7424
7425         add_new_free_space(cache, root->fs_info, chunk_offset,
7426                            chunk_offset + size);
7427
7428         free_excluded_extents(root, cache);
7429
7430         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7431                                 &cache->space_info);
7432         BUG_ON(ret);
7433
7434         spin_lock(&cache->space_info->lock);
7435         cache->space_info->bytes_readonly += cache->bytes_super;
7436         spin_unlock(&cache->space_info->lock);
7437
7438         __link_block_group(cache->space_info, cache);
7439
7440         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7441         BUG_ON(ret);
7442
7443         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7444                                 sizeof(cache->item));
7445         BUG_ON(ret);
7446
7447         set_avail_alloc_bits(extent_root->fs_info, type);
7448
7449         return 0;
7450 }
7451
7452 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7453                              struct btrfs_root *root, u64 group_start)
7454 {
7455         struct btrfs_path *path;
7456         struct btrfs_block_group_cache *block_group;
7457         struct btrfs_free_cluster *cluster;
7458         struct btrfs_root *tree_root = root->fs_info->tree_root;
7459         struct btrfs_key key;
7460         struct inode *inode;
7461         int ret;
7462         int factor;
7463
7464         root = root->fs_info->extent_root;
7465
7466         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7467         BUG_ON(!block_group);
7468         BUG_ON(!block_group->ro);
7469
7470         /*
7471          * Free the reserved super bytes from this block group before
7472          * remove it.
7473          */
7474         free_excluded_extents(root, block_group);
7475
7476         memcpy(&key, &block_group->key, sizeof(key));
7477         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7478                                   BTRFS_BLOCK_GROUP_RAID1 |
7479                                   BTRFS_BLOCK_GROUP_RAID10))
7480                 factor = 2;
7481         else
7482                 factor = 1;
7483
7484         /* make sure this block group isn't part of an allocation cluster */
7485         cluster = &root->fs_info->data_alloc_cluster;
7486         spin_lock(&cluster->refill_lock);
7487         btrfs_return_cluster_to_free_space(block_group, cluster);
7488         spin_unlock(&cluster->refill_lock);
7489
7490         /*
7491          * make sure this block group isn't part of a metadata
7492          * allocation cluster
7493          */
7494         cluster = &root->fs_info->meta_alloc_cluster;
7495         spin_lock(&cluster->refill_lock);
7496         btrfs_return_cluster_to_free_space(block_group, cluster);
7497         spin_unlock(&cluster->refill_lock);
7498
7499         path = btrfs_alloc_path();
7500         if (!path) {
7501                 ret = -ENOMEM;
7502                 goto out;
7503         }
7504
7505         inode = lookup_free_space_inode(tree_root, block_group, path);
7506         if (!IS_ERR(inode)) {
7507                 ret = btrfs_orphan_add(trans, inode);
7508                 BUG_ON(ret);
7509                 clear_nlink(inode);
7510                 /* One for the block groups ref */
7511                 spin_lock(&block_group->lock);
7512                 if (block_group->iref) {
7513                         block_group->iref = 0;
7514                         block_group->inode = NULL;
7515                         spin_unlock(&block_group->lock);
7516                         iput(inode);
7517                 } else {
7518                         spin_unlock(&block_group->lock);
7519                 }
7520                 /* One for our lookup ref */
7521                 btrfs_add_delayed_iput(inode);
7522         }
7523
7524         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7525         key.offset = block_group->key.objectid;
7526         key.type = 0;
7527
7528         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7529         if (ret < 0)
7530                 goto out;
7531         if (ret > 0)
7532                 btrfs_release_path(path);
7533         if (ret == 0) {
7534                 ret = btrfs_del_item(trans, tree_root, path);
7535                 if (ret)
7536                         goto out;
7537                 btrfs_release_path(path);
7538         }
7539
7540         spin_lock(&root->fs_info->block_group_cache_lock);
7541         rb_erase(&block_group->cache_node,
7542                  &root->fs_info->block_group_cache_tree);
7543         spin_unlock(&root->fs_info->block_group_cache_lock);
7544
7545         down_write(&block_group->space_info->groups_sem);
7546         /*
7547          * we must use list_del_init so people can check to see if they
7548          * are still on the list after taking the semaphore
7549          */
7550         list_del_init(&block_group->list);
7551         up_write(&block_group->space_info->groups_sem);
7552
7553         if (block_group->cached == BTRFS_CACHE_STARTED)
7554                 wait_block_group_cache_done(block_group);
7555
7556         btrfs_remove_free_space_cache(block_group);
7557
7558         spin_lock(&block_group->space_info->lock);
7559         block_group->space_info->total_bytes -= block_group->key.offset;
7560         block_group->space_info->bytes_readonly -= block_group->key.offset;
7561         block_group->space_info->disk_total -= block_group->key.offset * factor;
7562         spin_unlock(&block_group->space_info->lock);
7563
7564         memcpy(&key, &block_group->key, sizeof(key));
7565
7566         btrfs_clear_space_info_full(root->fs_info);
7567
7568         btrfs_put_block_group(block_group);
7569         btrfs_put_block_group(block_group);
7570
7571         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7572         if (ret > 0)
7573                 ret = -EIO;
7574         if (ret < 0)
7575                 goto out;
7576
7577         ret = btrfs_del_item(trans, root, path);
7578 out:
7579         btrfs_free_path(path);
7580         return ret;
7581 }
7582
7583 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7584 {
7585         struct btrfs_space_info *space_info;
7586         struct btrfs_super_block *disk_super;
7587         u64 features;
7588         u64 flags;
7589         int mixed = 0;
7590         int ret;
7591
7592         disk_super = fs_info->super_copy;
7593         if (!btrfs_super_root(disk_super))
7594                 return 1;
7595
7596         features = btrfs_super_incompat_flags(disk_super);
7597         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7598                 mixed = 1;
7599
7600         flags = BTRFS_BLOCK_GROUP_SYSTEM;
7601         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7602         if (ret)
7603                 goto out;
7604
7605         if (mixed) {
7606                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7607                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7608         } else {
7609                 flags = BTRFS_BLOCK_GROUP_METADATA;
7610                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7611                 if (ret)
7612                         goto out;
7613
7614                 flags = BTRFS_BLOCK_GROUP_DATA;
7615                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7616         }
7617 out:
7618         return ret;
7619 }
7620
7621 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7622 {
7623         return unpin_extent_range(root, start, end);
7624 }
7625
7626 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7627                                u64 num_bytes, u64 *actual_bytes)
7628 {
7629         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7630 }
7631
7632 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7633 {
7634         struct btrfs_fs_info *fs_info = root->fs_info;
7635         struct btrfs_block_group_cache *cache = NULL;
7636         u64 group_trimmed;
7637         u64 start;
7638         u64 end;
7639         u64 trimmed = 0;
7640         int ret = 0;
7641
7642         cache = btrfs_lookup_block_group(fs_info, range->start);
7643
7644         while (cache) {
7645                 if (cache->key.objectid >= (range->start + range->len)) {
7646                         btrfs_put_block_group(cache);
7647                         break;
7648                 }
7649
7650                 start = max(range->start, cache->key.objectid);
7651                 end = min(range->start + range->len,
7652                                 cache->key.objectid + cache->key.offset);
7653
7654                 if (end - start >= range->minlen) {
7655                         if (!block_group_cache_done(cache)) {
7656                                 ret = cache_block_group(cache, NULL, root, 0);
7657                                 if (!ret)
7658                                         wait_block_group_cache_done(cache);
7659                         }
7660                         ret = btrfs_trim_block_group(cache,
7661                                                      &group_trimmed,
7662                                                      start,
7663                                                      end,
7664                                                      range->minlen);
7665
7666                         trimmed += group_trimmed;
7667                         if (ret) {
7668                                 btrfs_put_block_group(cache);
7669                                 break;
7670                         }
7671                 }
7672
7673                 cache = next_block_group(fs_info->tree_root, cache);
7674         }
7675
7676         range->len = trimmed;
7677         return ret;
7678 }