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, int for_cow)
1876 {
1877         int ret;
1878         struct btrfs_fs_info *fs_info = root->fs_info;
1879
1880         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1881                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1882
1883         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1884                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1885                                         num_bytes,
1886                                         parent, root_objectid, (int)owner,
1887                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1888         } else {
1889                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1890                                         num_bytes,
1891                                         parent, root_objectid, owner, offset,
1892                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1893         }
1894         return ret;
1895 }
1896
1897 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1898                                   struct btrfs_root *root,
1899                                   u64 bytenr, u64 num_bytes,
1900                                   u64 parent, u64 root_objectid,
1901                                   u64 owner, u64 offset, int refs_to_add,
1902                                   struct btrfs_delayed_extent_op *extent_op)
1903 {
1904         struct btrfs_path *path;
1905         struct extent_buffer *leaf;
1906         struct btrfs_extent_item *item;
1907         u64 refs;
1908         int ret;
1909         int err = 0;
1910
1911         path = btrfs_alloc_path();
1912         if (!path)
1913                 return -ENOMEM;
1914
1915         path->reada = 1;
1916         path->leave_spinning = 1;
1917         /* this will setup the path even if it fails to insert the back ref */
1918         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1919                                            path, bytenr, num_bytes, parent,
1920                                            root_objectid, owner, offset,
1921                                            refs_to_add, extent_op);
1922         if (ret == 0)
1923                 goto out;
1924
1925         if (ret != -EAGAIN) {
1926                 err = ret;
1927                 goto out;
1928         }
1929
1930         leaf = path->nodes[0];
1931         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1932         refs = btrfs_extent_refs(leaf, item);
1933         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1934         if (extent_op)
1935                 __run_delayed_extent_op(extent_op, leaf, item);
1936
1937         btrfs_mark_buffer_dirty(leaf);
1938         btrfs_release_path(path);
1939
1940         path->reada = 1;
1941         path->leave_spinning = 1;
1942
1943         /* now insert the actual backref */
1944         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1945                                     path, bytenr, parent, root_objectid,
1946                                     owner, offset, refs_to_add);
1947         BUG_ON(ret);
1948 out:
1949         btrfs_free_path(path);
1950         return err;
1951 }
1952
1953 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1954                                 struct btrfs_root *root,
1955                                 struct btrfs_delayed_ref_node *node,
1956                                 struct btrfs_delayed_extent_op *extent_op,
1957                                 int insert_reserved)
1958 {
1959         int ret = 0;
1960         struct btrfs_delayed_data_ref *ref;
1961         struct btrfs_key ins;
1962         u64 parent = 0;
1963         u64 ref_root = 0;
1964         u64 flags = 0;
1965
1966         ins.objectid = node->bytenr;
1967         ins.offset = node->num_bytes;
1968         ins.type = BTRFS_EXTENT_ITEM_KEY;
1969
1970         ref = btrfs_delayed_node_to_data_ref(node);
1971         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1972                 parent = ref->parent;
1973         else
1974                 ref_root = ref->root;
1975
1976         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1977                 if (extent_op) {
1978                         BUG_ON(extent_op->update_key);
1979                         flags |= extent_op->flags_to_set;
1980                 }
1981                 ret = alloc_reserved_file_extent(trans, root,
1982                                                  parent, ref_root, flags,
1983                                                  ref->objectid, ref->offset,
1984                                                  &ins, node->ref_mod);
1985         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1986                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1987                                              node->num_bytes, parent,
1988                                              ref_root, ref->objectid,
1989                                              ref->offset, node->ref_mod,
1990                                              extent_op);
1991         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1992                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1993                                           node->num_bytes, parent,
1994                                           ref_root, ref->objectid,
1995                                           ref->offset, node->ref_mod,
1996                                           extent_op);
1997         } else {
1998                 BUG();
1999         }
2000         return ret;
2001 }
2002
2003 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2004                                     struct extent_buffer *leaf,
2005                                     struct btrfs_extent_item *ei)
2006 {
2007         u64 flags = btrfs_extent_flags(leaf, ei);
2008         if (extent_op->update_flags) {
2009                 flags |= extent_op->flags_to_set;
2010                 btrfs_set_extent_flags(leaf, ei, flags);
2011         }
2012
2013         if (extent_op->update_key) {
2014                 struct btrfs_tree_block_info *bi;
2015                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2016                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2017                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2018         }
2019 }
2020
2021 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2022                                  struct btrfs_root *root,
2023                                  struct btrfs_delayed_ref_node *node,
2024                                  struct btrfs_delayed_extent_op *extent_op)
2025 {
2026         struct btrfs_key key;
2027         struct btrfs_path *path;
2028         struct btrfs_extent_item *ei;
2029         struct extent_buffer *leaf;
2030         u32 item_size;
2031         int ret;
2032         int err = 0;
2033
2034         path = btrfs_alloc_path();
2035         if (!path)
2036                 return -ENOMEM;
2037
2038         key.objectid = node->bytenr;
2039         key.type = BTRFS_EXTENT_ITEM_KEY;
2040         key.offset = node->num_bytes;
2041
2042         path->reada = 1;
2043         path->leave_spinning = 1;
2044         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2045                                 path, 0, 1);
2046         if (ret < 0) {
2047                 err = ret;
2048                 goto out;
2049         }
2050         if (ret > 0) {
2051                 err = -EIO;
2052                 goto out;
2053         }
2054
2055         leaf = path->nodes[0];
2056         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2057 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2058         if (item_size < sizeof(*ei)) {
2059                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2060                                              path, (u64)-1, 0);
2061                 if (ret < 0) {
2062                         err = ret;
2063                         goto out;
2064                 }
2065                 leaf = path->nodes[0];
2066                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2067         }
2068 #endif
2069         BUG_ON(item_size < sizeof(*ei));
2070         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2071         __run_delayed_extent_op(extent_op, leaf, ei);
2072
2073         btrfs_mark_buffer_dirty(leaf);
2074 out:
2075         btrfs_free_path(path);
2076         return err;
2077 }
2078
2079 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2080                                 struct btrfs_root *root,
2081                                 struct btrfs_delayed_ref_node *node,
2082                                 struct btrfs_delayed_extent_op *extent_op,
2083                                 int insert_reserved)
2084 {
2085         int ret = 0;
2086         struct btrfs_delayed_tree_ref *ref;
2087         struct btrfs_key ins;
2088         u64 parent = 0;
2089         u64 ref_root = 0;
2090
2091         ins.objectid = node->bytenr;
2092         ins.offset = node->num_bytes;
2093         ins.type = BTRFS_EXTENT_ITEM_KEY;
2094
2095         ref = btrfs_delayed_node_to_tree_ref(node);
2096         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2097                 parent = ref->parent;
2098         else
2099                 ref_root = ref->root;
2100
2101         BUG_ON(node->ref_mod != 1);
2102         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2103                 BUG_ON(!extent_op || !extent_op->update_flags ||
2104                        !extent_op->update_key);
2105                 ret = alloc_reserved_tree_block(trans, root,
2106                                                 parent, ref_root,
2107                                                 extent_op->flags_to_set,
2108                                                 &extent_op->key,
2109                                                 ref->level, &ins);
2110         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2111                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2112                                              node->num_bytes, parent, ref_root,
2113                                              ref->level, 0, 1, extent_op);
2114         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2115                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2116                                           node->num_bytes, parent, ref_root,
2117                                           ref->level, 0, 1, extent_op);
2118         } else {
2119                 BUG();
2120         }
2121         return ret;
2122 }
2123
2124 /* helper function to actually process a single delayed ref entry */
2125 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2126                                struct btrfs_root *root,
2127                                struct btrfs_delayed_ref_node *node,
2128                                struct btrfs_delayed_extent_op *extent_op,
2129                                int insert_reserved)
2130 {
2131         int ret;
2132         if (btrfs_delayed_ref_is_head(node)) {
2133                 struct btrfs_delayed_ref_head *head;
2134                 /*
2135                  * we've hit the end of the chain and we were supposed
2136                  * to insert this extent into the tree.  But, it got
2137                  * deleted before we ever needed to insert it, so all
2138                  * we have to do is clean up the accounting
2139                  */
2140                 BUG_ON(extent_op);
2141                 head = btrfs_delayed_node_to_head(node);
2142                 if (insert_reserved) {
2143                         btrfs_pin_extent(root, node->bytenr,
2144                                          node->num_bytes, 1);
2145                         if (head->is_data) {
2146                                 ret = btrfs_del_csums(trans, root,
2147                                                       node->bytenr,
2148                                                       node->num_bytes);
2149                                 BUG_ON(ret);
2150                         }
2151                 }
2152                 mutex_unlock(&head->mutex);
2153                 return 0;
2154         }
2155
2156         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2157             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2158                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2159                                            insert_reserved);
2160         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2161                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2162                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2163                                            insert_reserved);
2164         else
2165                 BUG();
2166         return ret;
2167 }
2168
2169 static noinline struct btrfs_delayed_ref_node *
2170 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2171 {
2172         struct rb_node *node;
2173         struct btrfs_delayed_ref_node *ref;
2174         int action = BTRFS_ADD_DELAYED_REF;
2175 again:
2176         /*
2177          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2178          * this prevents ref count from going down to zero when
2179          * there still are pending delayed ref.
2180          */
2181         node = rb_prev(&head->node.rb_node);
2182         while (1) {
2183                 if (!node)
2184                         break;
2185                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2186                                 rb_node);
2187                 if (ref->bytenr != head->node.bytenr)
2188                         break;
2189                 if (ref->action == action)
2190                         return ref;
2191                 node = rb_prev(node);
2192         }
2193         if (action == BTRFS_ADD_DELAYED_REF) {
2194                 action = BTRFS_DROP_DELAYED_REF;
2195                 goto again;
2196         }
2197         return NULL;
2198 }
2199
2200 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2201                                        struct btrfs_root *root,
2202                                        struct list_head *cluster)
2203 {
2204         struct btrfs_delayed_ref_root *delayed_refs;
2205         struct btrfs_delayed_ref_node *ref;
2206         struct btrfs_delayed_ref_head *locked_ref = NULL;
2207         struct btrfs_delayed_extent_op *extent_op;
2208         int ret;
2209         int count = 0;
2210         int must_insert_reserved = 0;
2211
2212         delayed_refs = &trans->transaction->delayed_refs;
2213         while (1) {
2214                 if (!locked_ref) {
2215                         /* pick a new head ref from the cluster list */
2216                         if (list_empty(cluster))
2217                                 break;
2218
2219                         locked_ref = list_entry(cluster->next,
2220                                      struct btrfs_delayed_ref_head, cluster);
2221
2222                         /* grab the lock that says we are going to process
2223                          * all the refs for this head */
2224                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2225
2226                         /*
2227                          * we may have dropped the spin lock to get the head
2228                          * mutex lock, and that might have given someone else
2229                          * time to free the head.  If that's true, it has been
2230                          * removed from our list and we can move on.
2231                          */
2232                         if (ret == -EAGAIN) {
2233                                 locked_ref = NULL;
2234                                 count++;
2235                                 continue;
2236                         }
2237                 }
2238
2239                 /*
2240                  * locked_ref is the head node, so we have to go one
2241                  * node back for any delayed ref updates
2242                  */
2243                 ref = select_delayed_ref(locked_ref);
2244
2245                 if (ref && ref->seq &&
2246                     btrfs_check_delayed_seq(delayed_refs, ref->seq)) {
2247                         /*
2248                          * there are still refs with lower seq numbers in the
2249                          * process of being added. Don't run this ref yet.
2250                          */
2251                         list_del_init(&locked_ref->cluster);
2252                         mutex_unlock(&locked_ref->mutex);
2253                         locked_ref = NULL;
2254                         delayed_refs->num_heads_ready++;
2255                         spin_unlock(&delayed_refs->lock);
2256                         cond_resched();
2257                         spin_lock(&delayed_refs->lock);
2258                         continue;
2259                 }
2260
2261                 /*
2262                  * record the must insert reserved flag before we
2263                  * drop the spin lock.
2264                  */
2265                 must_insert_reserved = locked_ref->must_insert_reserved;
2266                 locked_ref->must_insert_reserved = 0;
2267
2268                 extent_op = locked_ref->extent_op;
2269                 locked_ref->extent_op = NULL;
2270
2271                 if (!ref) {
2272                         /* All delayed refs have been processed, Go ahead
2273                          * and send the head node to run_one_delayed_ref,
2274                          * so that any accounting fixes can happen
2275                          */
2276                         ref = &locked_ref->node;
2277
2278                         if (extent_op && must_insert_reserved) {
2279                                 kfree(extent_op);
2280                                 extent_op = NULL;
2281                         }
2282
2283                         if (extent_op) {
2284                                 spin_unlock(&delayed_refs->lock);
2285
2286                                 ret = run_delayed_extent_op(trans, root,
2287                                                             ref, extent_op);
2288                                 BUG_ON(ret);
2289                                 kfree(extent_op);
2290
2291                                 cond_resched();
2292                                 spin_lock(&delayed_refs->lock);
2293                                 continue;
2294                         }
2295
2296                         list_del_init(&locked_ref->cluster);
2297                         locked_ref = NULL;
2298                 }
2299
2300                 ref->in_tree = 0;
2301                 rb_erase(&ref->rb_node, &delayed_refs->root);
2302                 delayed_refs->num_entries--;
2303
2304                 spin_unlock(&delayed_refs->lock);
2305
2306                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2307                                           must_insert_reserved);
2308                 BUG_ON(ret);
2309
2310                 btrfs_put_delayed_ref(ref);
2311                 kfree(extent_op);
2312                 count++;
2313
2314                 cond_resched();
2315                 spin_lock(&delayed_refs->lock);
2316         }
2317         return count;
2318 }
2319
2320 /*
2321  * this starts processing the delayed reference count updates and
2322  * extent insertions we have queued up so far.  count can be
2323  * 0, which means to process everything in the tree at the start
2324  * of the run (but not newly added entries), or it can be some target
2325  * number you'd like to process.
2326  */
2327 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2328                            struct btrfs_root *root, unsigned long count)
2329 {
2330         struct rb_node *node;
2331         struct btrfs_delayed_ref_root *delayed_refs;
2332         struct btrfs_delayed_ref_node *ref;
2333         struct list_head cluster;
2334         int ret;
2335         int run_all = count == (unsigned long)-1;
2336         int run_most = 0;
2337
2338         if (root == root->fs_info->extent_root)
2339                 root = root->fs_info->tree_root;
2340
2341         delayed_refs = &trans->transaction->delayed_refs;
2342         INIT_LIST_HEAD(&cluster);
2343 again:
2344         spin_lock(&delayed_refs->lock);
2345         if (count == 0) {
2346                 count = delayed_refs->num_entries * 2;
2347                 run_most = 1;
2348         }
2349         while (1) {
2350                 if (!(run_all || run_most) &&
2351                     delayed_refs->num_heads_ready < 64)
2352                         break;
2353
2354                 /*
2355                  * go find something we can process in the rbtree.  We start at
2356                  * the beginning of the tree, and then build a cluster
2357                  * of refs to process starting at the first one we are able to
2358                  * lock
2359                  */
2360                 ret = btrfs_find_ref_cluster(trans, &cluster,
2361                                              delayed_refs->run_delayed_start);
2362                 if (ret)
2363                         break;
2364
2365                 ret = run_clustered_refs(trans, root, &cluster);
2366                 BUG_ON(ret < 0);
2367
2368                 count -= min_t(unsigned long, ret, count);
2369
2370                 if (count == 0)
2371                         break;
2372         }
2373
2374         if (run_all) {
2375                 node = rb_first(&delayed_refs->root);
2376                 if (!node)
2377                         goto out;
2378                 count = (unsigned long)-1;
2379
2380                 while (node) {
2381                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2382                                        rb_node);
2383                         if (btrfs_delayed_ref_is_head(ref)) {
2384                                 struct btrfs_delayed_ref_head *head;
2385
2386                                 head = btrfs_delayed_node_to_head(ref);
2387                                 atomic_inc(&ref->refs);
2388
2389                                 spin_unlock(&delayed_refs->lock);
2390                                 /*
2391                                  * Mutex was contended, block until it's
2392                                  * released and try again
2393                                  */
2394                                 mutex_lock(&head->mutex);
2395                                 mutex_unlock(&head->mutex);
2396
2397                                 btrfs_put_delayed_ref(ref);
2398                                 cond_resched();
2399                                 goto again;
2400                         }
2401                         node = rb_next(node);
2402                 }
2403                 spin_unlock(&delayed_refs->lock);
2404                 schedule_timeout(1);
2405                 goto again;
2406         }
2407 out:
2408         spin_unlock(&delayed_refs->lock);
2409         return 0;
2410 }
2411
2412 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2413                                 struct btrfs_root *root,
2414                                 u64 bytenr, u64 num_bytes, u64 flags,
2415                                 int is_data)
2416 {
2417         struct btrfs_delayed_extent_op *extent_op;
2418         int ret;
2419
2420         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2421         if (!extent_op)
2422                 return -ENOMEM;
2423
2424         extent_op->flags_to_set = flags;
2425         extent_op->update_flags = 1;
2426         extent_op->update_key = 0;
2427         extent_op->is_data = is_data ? 1 : 0;
2428
2429         ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2430                                           num_bytes, extent_op);
2431         if (ret)
2432                 kfree(extent_op);
2433         return ret;
2434 }
2435
2436 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2437                                       struct btrfs_root *root,
2438                                       struct btrfs_path *path,
2439                                       u64 objectid, u64 offset, u64 bytenr)
2440 {
2441         struct btrfs_delayed_ref_head *head;
2442         struct btrfs_delayed_ref_node *ref;
2443         struct btrfs_delayed_data_ref *data_ref;
2444         struct btrfs_delayed_ref_root *delayed_refs;
2445         struct rb_node *node;
2446         int ret = 0;
2447
2448         ret = -ENOENT;
2449         delayed_refs = &trans->transaction->delayed_refs;
2450         spin_lock(&delayed_refs->lock);
2451         head = btrfs_find_delayed_ref_head(trans, bytenr);
2452         if (!head)
2453                 goto out;
2454
2455         if (!mutex_trylock(&head->mutex)) {
2456                 atomic_inc(&head->node.refs);
2457                 spin_unlock(&delayed_refs->lock);
2458
2459                 btrfs_release_path(path);
2460
2461                 /*
2462                  * Mutex was contended, block until it's released and let
2463                  * caller try again
2464                  */
2465                 mutex_lock(&head->mutex);
2466                 mutex_unlock(&head->mutex);
2467                 btrfs_put_delayed_ref(&head->node);
2468                 return -EAGAIN;
2469         }
2470
2471         node = rb_prev(&head->node.rb_node);
2472         if (!node)
2473                 goto out_unlock;
2474
2475         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2476
2477         if (ref->bytenr != bytenr)
2478                 goto out_unlock;
2479
2480         ret = 1;
2481         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2482                 goto out_unlock;
2483
2484         data_ref = btrfs_delayed_node_to_data_ref(ref);
2485
2486         node = rb_prev(node);
2487         if (node) {
2488                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2489                 if (ref->bytenr == bytenr)
2490                         goto out_unlock;
2491         }
2492
2493         if (data_ref->root != root->root_key.objectid ||
2494             data_ref->objectid != objectid || data_ref->offset != offset)
2495                 goto out_unlock;
2496
2497         ret = 0;
2498 out_unlock:
2499         mutex_unlock(&head->mutex);
2500 out:
2501         spin_unlock(&delayed_refs->lock);
2502         return ret;
2503 }
2504
2505 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2506                                         struct btrfs_root *root,
2507                                         struct btrfs_path *path,
2508                                         u64 objectid, u64 offset, u64 bytenr)
2509 {
2510         struct btrfs_root *extent_root = root->fs_info->extent_root;
2511         struct extent_buffer *leaf;
2512         struct btrfs_extent_data_ref *ref;
2513         struct btrfs_extent_inline_ref *iref;
2514         struct btrfs_extent_item *ei;
2515         struct btrfs_key key;
2516         u32 item_size;
2517         int ret;
2518
2519         key.objectid = bytenr;
2520         key.offset = (u64)-1;
2521         key.type = BTRFS_EXTENT_ITEM_KEY;
2522
2523         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2524         if (ret < 0)
2525                 goto out;
2526         BUG_ON(ret == 0);
2527
2528         ret = -ENOENT;
2529         if (path->slots[0] == 0)
2530                 goto out;
2531
2532         path->slots[0]--;
2533         leaf = path->nodes[0];
2534         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2535
2536         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2537                 goto out;
2538
2539         ret = 1;
2540         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2541 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2542         if (item_size < sizeof(*ei)) {
2543                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2544                 goto out;
2545         }
2546 #endif
2547         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2548
2549         if (item_size != sizeof(*ei) +
2550             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2551                 goto out;
2552
2553         if (btrfs_extent_generation(leaf, ei) <=
2554             btrfs_root_last_snapshot(&root->root_item))
2555                 goto out;
2556
2557         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2558         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2559             BTRFS_EXTENT_DATA_REF_KEY)
2560                 goto out;
2561
2562         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2563         if (btrfs_extent_refs(leaf, ei) !=
2564             btrfs_extent_data_ref_count(leaf, ref) ||
2565             btrfs_extent_data_ref_root(leaf, ref) !=
2566             root->root_key.objectid ||
2567             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2568             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2569                 goto out;
2570
2571         ret = 0;
2572 out:
2573         return ret;
2574 }
2575
2576 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2577                           struct btrfs_root *root,
2578                           u64 objectid, u64 offset, u64 bytenr)
2579 {
2580         struct btrfs_path *path;
2581         int ret;
2582         int ret2;
2583
2584         path = btrfs_alloc_path();
2585         if (!path)
2586                 return -ENOENT;
2587
2588         do {
2589                 ret = check_committed_ref(trans, root, path, objectid,
2590                                           offset, bytenr);
2591                 if (ret && ret != -ENOENT)
2592                         goto out;
2593
2594                 ret2 = check_delayed_ref(trans, root, path, objectid,
2595                                          offset, bytenr);
2596         } while (ret2 == -EAGAIN);
2597
2598         if (ret2 && ret2 != -ENOENT) {
2599                 ret = ret2;
2600                 goto out;
2601         }
2602
2603         if (ret != -ENOENT || ret2 != -ENOENT)
2604                 ret = 0;
2605 out:
2606         btrfs_free_path(path);
2607         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2608                 WARN_ON(ret > 0);
2609         return ret;
2610 }
2611
2612 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2613                            struct btrfs_root *root,
2614                            struct extent_buffer *buf,
2615                            int full_backref, int inc, int for_cow)
2616 {
2617         u64 bytenr;
2618         u64 num_bytes;
2619         u64 parent;
2620         u64 ref_root;
2621         u32 nritems;
2622         struct btrfs_key key;
2623         struct btrfs_file_extent_item *fi;
2624         int i;
2625         int level;
2626         int ret = 0;
2627         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2628                             u64, u64, u64, u64, u64, u64, int);
2629
2630         ref_root = btrfs_header_owner(buf);
2631         nritems = btrfs_header_nritems(buf);
2632         level = btrfs_header_level(buf);
2633
2634         if (!root->ref_cows && level == 0)
2635                 return 0;
2636
2637         if (inc)
2638                 process_func = btrfs_inc_extent_ref;
2639         else
2640                 process_func = btrfs_free_extent;
2641
2642         if (full_backref)
2643                 parent = buf->start;
2644         else
2645                 parent = 0;
2646
2647         for (i = 0; i < nritems; i++) {
2648                 if (level == 0) {
2649                         btrfs_item_key_to_cpu(buf, &key, i);
2650                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2651                                 continue;
2652                         fi = btrfs_item_ptr(buf, i,
2653                                             struct btrfs_file_extent_item);
2654                         if (btrfs_file_extent_type(buf, fi) ==
2655                             BTRFS_FILE_EXTENT_INLINE)
2656                                 continue;
2657                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2658                         if (bytenr == 0)
2659                                 continue;
2660
2661                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2662                         key.offset -= btrfs_file_extent_offset(buf, fi);
2663                         ret = process_func(trans, root, bytenr, num_bytes,
2664                                            parent, ref_root, key.objectid,
2665                                            key.offset, for_cow);
2666                         if (ret)
2667                                 goto fail;
2668                 } else {
2669                         bytenr = btrfs_node_blockptr(buf, i);
2670                         num_bytes = btrfs_level_size(root, level - 1);
2671                         ret = process_func(trans, root, bytenr, num_bytes,
2672                                            parent, ref_root, level - 1, 0,
2673                                            for_cow);
2674                         if (ret)
2675                                 goto fail;
2676                 }
2677         }
2678         return 0;
2679 fail:
2680         BUG();
2681         return ret;
2682 }
2683
2684 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2685                   struct extent_buffer *buf, int full_backref, int for_cow)
2686 {
2687         return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2688 }
2689
2690 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2691                   struct extent_buffer *buf, int full_backref, int for_cow)
2692 {
2693         return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2694 }
2695
2696 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2697                                  struct btrfs_root *root,
2698                                  struct btrfs_path *path,
2699                                  struct btrfs_block_group_cache *cache)
2700 {
2701         int ret;
2702         struct btrfs_root *extent_root = root->fs_info->extent_root;
2703         unsigned long bi;
2704         struct extent_buffer *leaf;
2705
2706         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2707         if (ret < 0)
2708                 goto fail;
2709         BUG_ON(ret);
2710
2711         leaf = path->nodes[0];
2712         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2713         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2714         btrfs_mark_buffer_dirty(leaf);
2715         btrfs_release_path(path);
2716 fail:
2717         if (ret)
2718                 return ret;
2719         return 0;
2720
2721 }
2722
2723 static struct btrfs_block_group_cache *
2724 next_block_group(struct btrfs_root *root,
2725                  struct btrfs_block_group_cache *cache)
2726 {
2727         struct rb_node *node;
2728         spin_lock(&root->fs_info->block_group_cache_lock);
2729         node = rb_next(&cache->cache_node);
2730         btrfs_put_block_group(cache);
2731         if (node) {
2732                 cache = rb_entry(node, struct btrfs_block_group_cache,
2733                                  cache_node);
2734                 btrfs_get_block_group(cache);
2735         } else
2736                 cache = NULL;
2737         spin_unlock(&root->fs_info->block_group_cache_lock);
2738         return cache;
2739 }
2740
2741 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2742                             struct btrfs_trans_handle *trans,
2743                             struct btrfs_path *path)
2744 {
2745         struct btrfs_root *root = block_group->fs_info->tree_root;
2746         struct inode *inode = NULL;
2747         u64 alloc_hint = 0;
2748         int dcs = BTRFS_DC_ERROR;
2749         int num_pages = 0;
2750         int retries = 0;
2751         int ret = 0;
2752
2753         /*
2754          * If this block group is smaller than 100 megs don't bother caching the
2755          * block group.
2756          */
2757         if (block_group->key.offset < (100 * 1024 * 1024)) {
2758                 spin_lock(&block_group->lock);
2759                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2760                 spin_unlock(&block_group->lock);
2761                 return 0;
2762         }
2763
2764 again:
2765         inode = lookup_free_space_inode(root, block_group, path);
2766         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2767                 ret = PTR_ERR(inode);
2768                 btrfs_release_path(path);
2769                 goto out;
2770         }
2771
2772         if (IS_ERR(inode)) {
2773                 BUG_ON(retries);
2774                 retries++;
2775
2776                 if (block_group->ro)
2777                         goto out_free;
2778
2779                 ret = create_free_space_inode(root, trans, block_group, path);
2780                 if (ret)
2781                         goto out_free;
2782                 goto again;
2783         }
2784
2785         /* We've already setup this transaction, go ahead and exit */
2786         if (block_group->cache_generation == trans->transid &&
2787             i_size_read(inode)) {
2788                 dcs = BTRFS_DC_SETUP;
2789                 goto out_put;
2790         }
2791
2792         /*
2793          * We want to set the generation to 0, that way if anything goes wrong
2794          * from here on out we know not to trust this cache when we load up next
2795          * time.
2796          */
2797         BTRFS_I(inode)->generation = 0;
2798         ret = btrfs_update_inode(trans, root, inode);
2799         WARN_ON(ret);
2800
2801         if (i_size_read(inode) > 0) {
2802                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2803                                                       inode);
2804                 if (ret)
2805                         goto out_put;
2806         }
2807
2808         spin_lock(&block_group->lock);
2809         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2810                 /* We're not cached, don't bother trying to write stuff out */
2811                 dcs = BTRFS_DC_WRITTEN;
2812                 spin_unlock(&block_group->lock);
2813                 goto out_put;
2814         }
2815         spin_unlock(&block_group->lock);
2816
2817         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2818         if (!num_pages)
2819                 num_pages = 1;
2820
2821         /*
2822          * Just to make absolutely sure we have enough space, we're going to
2823          * preallocate 12 pages worth of space for each block group.  In
2824          * practice we ought to use at most 8, but we need extra space so we can
2825          * add our header and have a terminator between the extents and the
2826          * bitmaps.
2827          */
2828         num_pages *= 16;
2829         num_pages *= PAGE_CACHE_SIZE;
2830
2831         ret = btrfs_check_data_free_space(inode, num_pages);
2832         if (ret)
2833                 goto out_put;
2834
2835         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2836                                               num_pages, num_pages,
2837                                               &alloc_hint);
2838         if (!ret)
2839                 dcs = BTRFS_DC_SETUP;
2840         btrfs_free_reserved_data_space(inode, num_pages);
2841
2842 out_put:
2843         iput(inode);
2844 out_free:
2845         btrfs_release_path(path);
2846 out:
2847         spin_lock(&block_group->lock);
2848         if (!ret)
2849                 block_group->cache_generation = trans->transid;
2850         block_group->disk_cache_state = dcs;
2851         spin_unlock(&block_group->lock);
2852
2853         return ret;
2854 }
2855
2856 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2857                                    struct btrfs_root *root)
2858 {
2859         struct btrfs_block_group_cache *cache;
2860         int err = 0;
2861         struct btrfs_path *path;
2862         u64 last = 0;
2863
2864         path = btrfs_alloc_path();
2865         if (!path)
2866                 return -ENOMEM;
2867
2868 again:
2869         while (1) {
2870                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2871                 while (cache) {
2872                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2873                                 break;
2874                         cache = next_block_group(root, cache);
2875                 }
2876                 if (!cache) {
2877                         if (last == 0)
2878                                 break;
2879                         last = 0;
2880                         continue;
2881                 }
2882                 err = cache_save_setup(cache, trans, path);
2883                 last = cache->key.objectid + cache->key.offset;
2884                 btrfs_put_block_group(cache);
2885         }
2886
2887         while (1) {
2888                 if (last == 0) {
2889                         err = btrfs_run_delayed_refs(trans, root,
2890                                                      (unsigned long)-1);
2891                         BUG_ON(err);
2892                 }
2893
2894                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2895                 while (cache) {
2896                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2897                                 btrfs_put_block_group(cache);
2898                                 goto again;
2899                         }
2900
2901                         if (cache->dirty)
2902                                 break;
2903                         cache = next_block_group(root, cache);
2904                 }
2905                 if (!cache) {
2906                         if (last == 0)
2907                                 break;
2908                         last = 0;
2909                         continue;
2910                 }
2911
2912                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2913                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2914                 cache->dirty = 0;
2915                 last = cache->key.objectid + cache->key.offset;
2916
2917                 err = write_one_cache_group(trans, root, path, cache);
2918                 BUG_ON(err);
2919                 btrfs_put_block_group(cache);
2920         }
2921
2922         while (1) {
2923                 /*
2924                  * I don't think this is needed since we're just marking our
2925                  * preallocated extent as written, but just in case it can't
2926                  * hurt.
2927                  */
2928                 if (last == 0) {
2929                         err = btrfs_run_delayed_refs(trans, root,
2930                                                      (unsigned long)-1);
2931                         BUG_ON(err);
2932                 }
2933
2934                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2935                 while (cache) {
2936                         /*
2937                          * Really this shouldn't happen, but it could if we
2938                          * couldn't write the entire preallocated extent and
2939                          * splitting the extent resulted in a new block.
2940                          */
2941                         if (cache->dirty) {
2942                                 btrfs_put_block_group(cache);
2943                                 goto again;
2944                         }
2945                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2946                                 break;
2947                         cache = next_block_group(root, cache);
2948                 }
2949                 if (!cache) {
2950                         if (last == 0)
2951                                 break;
2952                         last = 0;
2953                         continue;
2954                 }
2955
2956                 btrfs_write_out_cache(root, trans, cache, path);
2957
2958                 /*
2959                  * If we didn't have an error then the cache state is still
2960                  * NEED_WRITE, so we can set it to WRITTEN.
2961                  */
2962                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2963                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2964                 last = cache->key.objectid + cache->key.offset;
2965                 btrfs_put_block_group(cache);
2966         }
2967
2968         btrfs_free_path(path);
2969         return 0;
2970 }
2971
2972 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2973 {
2974         struct btrfs_block_group_cache *block_group;
2975         int readonly = 0;
2976
2977         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2978         if (!block_group || block_group->ro)
2979                 readonly = 1;
2980         if (block_group)
2981                 btrfs_put_block_group(block_group);
2982         return readonly;
2983 }
2984
2985 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2986                              u64 total_bytes, u64 bytes_used,
2987                              struct btrfs_space_info **space_info)
2988 {
2989         struct btrfs_space_info *found;
2990         int i;
2991         int factor;
2992
2993         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2994                      BTRFS_BLOCK_GROUP_RAID10))
2995                 factor = 2;
2996         else
2997                 factor = 1;
2998
2999         found = __find_space_info(info, flags);
3000         if (found) {
3001                 spin_lock(&found->lock);
3002                 found->total_bytes += total_bytes;
3003                 found->disk_total += total_bytes * factor;
3004                 found->bytes_used += bytes_used;
3005                 found->disk_used += bytes_used * factor;
3006                 found->full = 0;
3007                 spin_unlock(&found->lock);
3008                 *space_info = found;
3009                 return 0;
3010         }
3011         found = kzalloc(sizeof(*found), GFP_NOFS);
3012         if (!found)
3013                 return -ENOMEM;
3014
3015         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3016                 INIT_LIST_HEAD(&found->block_groups[i]);
3017         init_rwsem(&found->groups_sem);
3018         spin_lock_init(&found->lock);
3019         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3020                                 BTRFS_BLOCK_GROUP_SYSTEM |
3021                                 BTRFS_BLOCK_GROUP_METADATA);
3022         found->total_bytes = total_bytes;
3023         found->disk_total = total_bytes * factor;
3024         found->bytes_used = bytes_used;
3025         found->disk_used = bytes_used * factor;
3026         found->bytes_pinned = 0;
3027         found->bytes_reserved = 0;
3028         found->bytes_readonly = 0;
3029         found->bytes_may_use = 0;
3030         found->full = 0;
3031         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3032         found->chunk_alloc = 0;
3033         found->flush = 0;
3034         init_waitqueue_head(&found->wait);
3035         *space_info = found;
3036         list_add_rcu(&found->list, &info->space_info);
3037         return 0;
3038 }
3039
3040 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3041 {
3042         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3043                                    BTRFS_BLOCK_GROUP_RAID1 |
3044                                    BTRFS_BLOCK_GROUP_RAID10 |
3045                                    BTRFS_BLOCK_GROUP_DUP);
3046         if (extra_flags) {
3047                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3048                         fs_info->avail_data_alloc_bits |= extra_flags;
3049                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3050                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3051                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3052                         fs_info->avail_system_alloc_bits |= extra_flags;
3053         }
3054 }
3055
3056 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3057 {
3058         /*
3059          * we add in the count of missing devices because we want
3060          * to make sure that any RAID levels on a degraded FS
3061          * continue to be honored.
3062          */
3063         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3064                 root->fs_info->fs_devices->missing_devices;
3065
3066         if (num_devices == 1)
3067                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3068         if (num_devices < 4)
3069                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3070
3071         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3072             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3073                       BTRFS_BLOCK_GROUP_RAID10))) {
3074                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3075         }
3076
3077         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3078             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3079                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3080         }
3081
3082         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3083             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3084              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3085              (flags & BTRFS_BLOCK_GROUP_DUP)))
3086                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3087         return flags;
3088 }
3089
3090 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3091 {
3092         if (flags & BTRFS_BLOCK_GROUP_DATA)
3093                 flags |= root->fs_info->avail_data_alloc_bits &
3094                          root->fs_info->data_alloc_profile;
3095         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3096                 flags |= root->fs_info->avail_system_alloc_bits &
3097                          root->fs_info->system_alloc_profile;
3098         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3099                 flags |= root->fs_info->avail_metadata_alloc_bits &
3100                          root->fs_info->metadata_alloc_profile;
3101         return btrfs_reduce_alloc_profile(root, flags);
3102 }
3103
3104 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3105 {
3106         u64 flags;
3107
3108         if (data)
3109                 flags = BTRFS_BLOCK_GROUP_DATA;
3110         else if (root == root->fs_info->chunk_root)
3111                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3112         else
3113                 flags = BTRFS_BLOCK_GROUP_METADATA;
3114
3115         return get_alloc_profile(root, flags);
3116 }
3117
3118 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3119 {
3120         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3121                                                        BTRFS_BLOCK_GROUP_DATA);
3122 }
3123
3124 /*
3125  * This will check the space that the inode allocates from to make sure we have
3126  * enough space for bytes.
3127  */
3128 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3129 {
3130         struct btrfs_space_info *data_sinfo;
3131         struct btrfs_root *root = BTRFS_I(inode)->root;
3132         u64 used;
3133         int ret = 0, committed = 0, alloc_chunk = 1;
3134
3135         /* make sure bytes are sectorsize aligned */
3136         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3137
3138         if (root == root->fs_info->tree_root ||
3139             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3140                 alloc_chunk = 0;
3141                 committed = 1;
3142         }
3143
3144         data_sinfo = BTRFS_I(inode)->space_info;
3145         if (!data_sinfo)
3146                 goto alloc;
3147
3148 again:
3149         /* make sure we have enough space to handle the data first */
3150         spin_lock(&data_sinfo->lock);
3151         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3152                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3153                 data_sinfo->bytes_may_use;
3154
3155         if (used + bytes > data_sinfo->total_bytes) {
3156                 struct btrfs_trans_handle *trans;
3157
3158                 /*
3159                  * if we don't have enough free bytes in this space then we need
3160                  * to alloc a new chunk.
3161                  */
3162                 if (!data_sinfo->full && alloc_chunk) {
3163                         u64 alloc_target;
3164
3165                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3166                         spin_unlock(&data_sinfo->lock);
3167 alloc:
3168                         alloc_target = btrfs_get_alloc_profile(root, 1);
3169                         trans = btrfs_join_transaction(root);
3170                         if (IS_ERR(trans))
3171                                 return PTR_ERR(trans);
3172
3173                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3174                                              bytes + 2 * 1024 * 1024,
3175                                              alloc_target,
3176                                              CHUNK_ALLOC_NO_FORCE);
3177                         btrfs_end_transaction(trans, root);
3178                         if (ret < 0) {
3179                                 if (ret != -ENOSPC)
3180                                         return ret;
3181                                 else
3182                                         goto commit_trans;
3183                         }
3184
3185                         if (!data_sinfo) {
3186                                 btrfs_set_inode_space_info(root, inode);
3187                                 data_sinfo = BTRFS_I(inode)->space_info;
3188                         }
3189                         goto again;
3190                 }
3191
3192                 /*
3193                  * If we have less pinned bytes than we want to allocate then
3194                  * don't bother committing the transaction, it won't help us.
3195                  */
3196                 if (data_sinfo->bytes_pinned < bytes)
3197                         committed = 1;
3198                 spin_unlock(&data_sinfo->lock);
3199
3200                 /* commit the current transaction and try again */
3201 commit_trans:
3202                 if (!committed &&
3203                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3204                         committed = 1;
3205                         trans = btrfs_join_transaction(root);
3206                         if (IS_ERR(trans))
3207                                 return PTR_ERR(trans);
3208                         ret = btrfs_commit_transaction(trans, root);
3209                         if (ret)
3210                                 return ret;
3211                         goto again;
3212                 }
3213
3214                 return -ENOSPC;
3215         }
3216         data_sinfo->bytes_may_use += bytes;
3217         spin_unlock(&data_sinfo->lock);
3218
3219         return 0;
3220 }
3221
3222 /*
3223  * Called if we need to clear a data reservation for this inode.
3224  */
3225 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3226 {
3227         struct btrfs_root *root = BTRFS_I(inode)->root;
3228         struct btrfs_space_info *data_sinfo;
3229
3230         /* make sure bytes are sectorsize aligned */
3231         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3232
3233         data_sinfo = BTRFS_I(inode)->space_info;
3234         spin_lock(&data_sinfo->lock);
3235         data_sinfo->bytes_may_use -= bytes;
3236         spin_unlock(&data_sinfo->lock);
3237 }
3238
3239 static void force_metadata_allocation(struct btrfs_fs_info *info)
3240 {
3241         struct list_head *head = &info->space_info;
3242         struct btrfs_space_info *found;
3243
3244         rcu_read_lock();
3245         list_for_each_entry_rcu(found, head, list) {
3246                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3247                         found->force_alloc = CHUNK_ALLOC_FORCE;
3248         }
3249         rcu_read_unlock();
3250 }
3251
3252 static int should_alloc_chunk(struct btrfs_root *root,
3253                               struct btrfs_space_info *sinfo, u64 alloc_bytes,
3254                               int force)
3255 {
3256         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3257         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3258         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3259         u64 thresh;
3260
3261         if (force == CHUNK_ALLOC_FORCE)
3262                 return 1;
3263
3264         /*
3265          * We need to take into account the global rsv because for all intents
3266          * and purposes it's used space.  Don't worry about locking the
3267          * global_rsv, it doesn't change except when the transaction commits.
3268          */
3269         num_allocated += global_rsv->size;
3270
3271         /*
3272          * in limited mode, we want to have some free space up to
3273          * about 1% of the FS size.
3274          */
3275         if (force == CHUNK_ALLOC_LIMITED) {
3276                 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3277                 thresh = max_t(u64, 64 * 1024 * 1024,
3278                                div_factor_fine(thresh, 1));
3279
3280                 if (num_bytes - num_allocated < thresh)
3281                         return 1;
3282         }
3283
3284         /*
3285          * we have two similar checks here, one based on percentage
3286          * and once based on a hard number of 256MB.  The idea
3287          * is that if we have a good amount of free
3288          * room, don't allocate a chunk.  A good mount is
3289          * less than 80% utilized of the chunks we have allocated,
3290          * or more than 256MB free
3291          */
3292         if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3293                 return 0;
3294
3295         if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3296                 return 0;
3297
3298         thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3299
3300         /* 256MB or 5% of the FS */
3301         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3302
3303         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3304                 return 0;
3305         return 1;
3306 }
3307
3308 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3309                           struct btrfs_root *extent_root, u64 alloc_bytes,
3310                           u64 flags, int force)
3311 {
3312         struct btrfs_space_info *space_info;
3313         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3314         int wait_for_alloc = 0;
3315         int ret = 0;
3316
3317         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3318
3319         space_info = __find_space_info(extent_root->fs_info, flags);
3320         if (!space_info) {
3321                 ret = update_space_info(extent_root->fs_info, flags,
3322                                         0, 0, &space_info);
3323                 BUG_ON(ret);
3324         }
3325         BUG_ON(!space_info);
3326
3327 again:
3328         spin_lock(&space_info->lock);
3329         if (space_info->force_alloc)
3330                 force = space_info->force_alloc;
3331         if (space_info->full) {
3332                 spin_unlock(&space_info->lock);
3333                 return 0;
3334         }
3335
3336         if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3337                 spin_unlock(&space_info->lock);
3338                 return 0;
3339         } else if (space_info->chunk_alloc) {
3340                 wait_for_alloc = 1;
3341         } else {
3342                 space_info->chunk_alloc = 1;
3343         }
3344
3345         spin_unlock(&space_info->lock);
3346
3347         mutex_lock(&fs_info->chunk_mutex);
3348
3349         /*
3350          * The chunk_mutex is held throughout the entirety of a chunk
3351          * allocation, so once we've acquired the chunk_mutex we know that the
3352          * other guy is done and we need to recheck and see if we should
3353          * allocate.
3354          */
3355         if (wait_for_alloc) {
3356                 mutex_unlock(&fs_info->chunk_mutex);
3357                 wait_for_alloc = 0;
3358                 goto again;
3359         }
3360
3361         /*
3362          * If we have mixed data/metadata chunks we want to make sure we keep
3363          * allocating mixed chunks instead of individual chunks.
3364          */
3365         if (btrfs_mixed_space_info(space_info))
3366                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3367
3368         /*
3369          * if we're doing a data chunk, go ahead and make sure that
3370          * we keep a reasonable number of metadata chunks allocated in the
3371          * FS as well.
3372          */
3373         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3374                 fs_info->data_chunk_allocations++;
3375                 if (!(fs_info->data_chunk_allocations %
3376                       fs_info->metadata_ratio))
3377                         force_metadata_allocation(fs_info);
3378         }
3379
3380         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3381         if (ret < 0 && ret != -ENOSPC)
3382                 goto out;
3383
3384         spin_lock(&space_info->lock);
3385         if (ret)
3386                 space_info->full = 1;
3387         else
3388                 ret = 1;
3389
3390         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3391         space_info->chunk_alloc = 0;
3392         spin_unlock(&space_info->lock);
3393 out:
3394         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3395         return ret;
3396 }
3397
3398 /*
3399  * shrink metadata reservation for delalloc
3400  */
3401 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3402                            bool wait_ordered)
3403 {
3404         struct btrfs_block_rsv *block_rsv;
3405         struct btrfs_space_info *space_info;
3406         struct btrfs_trans_handle *trans;
3407         u64 reserved;
3408         u64 max_reclaim;
3409         u64 reclaimed = 0;
3410         long time_left;
3411         unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3412         int loops = 0;
3413         unsigned long progress;
3414
3415         trans = (struct btrfs_trans_handle *)current->journal_info;
3416         block_rsv = &root->fs_info->delalloc_block_rsv;
3417         space_info = block_rsv->space_info;
3418
3419         smp_mb();
3420         reserved = space_info->bytes_may_use;
3421         progress = space_info->reservation_progress;
3422
3423         if (reserved == 0)
3424                 return 0;
3425
3426         smp_mb();
3427         if (root->fs_info->delalloc_bytes == 0) {
3428                 if (trans)
3429                         return 0;
3430                 btrfs_wait_ordered_extents(root, 0, 0);
3431                 return 0;
3432         }
3433
3434         max_reclaim = min(reserved, to_reclaim);
3435         nr_pages = max_t(unsigned long, nr_pages,
3436                          max_reclaim >> PAGE_CACHE_SHIFT);
3437         while (loops < 1024) {
3438                 /* have the flusher threads jump in and do some IO */
3439                 smp_mb();
3440                 nr_pages = min_t(unsigned long, nr_pages,
3441                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3442                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3443
3444                 spin_lock(&space_info->lock);
3445                 if (reserved > space_info->bytes_may_use)
3446                         reclaimed += reserved - space_info->bytes_may_use;
3447                 reserved = space_info->bytes_may_use;
3448                 spin_unlock(&space_info->lock);
3449
3450                 loops++;
3451
3452                 if (reserved == 0 || reclaimed >= max_reclaim)
3453                         break;
3454
3455                 if (trans && trans->transaction->blocked)
3456                         return -EAGAIN;
3457
3458                 if (wait_ordered && !trans) {
3459                         btrfs_wait_ordered_extents(root, 0, 0);
3460                 } else {
3461                         time_left = schedule_timeout_interruptible(1);
3462
3463                         /* We were interrupted, exit */
3464                         if (time_left)
3465                                 break;
3466                 }
3467
3468                 /* we've kicked the IO a few times, if anything has been freed,
3469                  * exit.  There is no sense in looping here for a long time
3470                  * when we really need to commit the transaction, or there are
3471                  * just too many writers without enough free space
3472                  */
3473
3474                 if (loops > 3) {
3475                         smp_mb();
3476                         if (progress != space_info->reservation_progress)
3477                                 break;
3478                 }
3479
3480         }
3481
3482         return reclaimed >= to_reclaim;
3483 }
3484
3485 /**
3486  * maybe_commit_transaction - possibly commit the transaction if its ok to
3487  * @root - the root we're allocating for
3488  * @bytes - the number of bytes we want to reserve
3489  * @force - force the commit
3490  *
3491  * This will check to make sure that committing the transaction will actually
3492  * get us somewhere and then commit the transaction if it does.  Otherwise it
3493  * will return -ENOSPC.
3494  */
3495 static int may_commit_transaction(struct btrfs_root *root,
3496                                   struct btrfs_space_info *space_info,
3497                                   u64 bytes, int force)
3498 {
3499         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3500         struct btrfs_trans_handle *trans;
3501
3502         trans = (struct btrfs_trans_handle *)current->journal_info;
3503         if (trans)
3504                 return -EAGAIN;
3505
3506         if (force)
3507                 goto commit;
3508
3509         /* See if there is enough pinned space to make this reservation */
3510         spin_lock(&space_info->lock);
3511         if (space_info->bytes_pinned >= bytes) {
3512                 spin_unlock(&space_info->lock);
3513                 goto commit;
3514         }
3515         spin_unlock(&space_info->lock);
3516
3517         /*
3518          * See if there is some space in the delayed insertion reservation for
3519          * this reservation.
3520          */
3521         if (space_info != delayed_rsv->space_info)
3522                 return -ENOSPC;
3523
3524         spin_lock(&delayed_rsv->lock);
3525         if (delayed_rsv->size < bytes) {
3526                 spin_unlock(&delayed_rsv->lock);
3527                 return -ENOSPC;
3528         }
3529         spin_unlock(&delayed_rsv->lock);
3530
3531 commit:
3532         trans = btrfs_join_transaction(root);
3533         if (IS_ERR(trans))
3534                 return -ENOSPC;
3535
3536         return btrfs_commit_transaction(trans, root);
3537 }
3538
3539 /**
3540  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3541  * @root - the root we're allocating for
3542  * @block_rsv - the block_rsv we're allocating for
3543  * @orig_bytes - the number of bytes we want
3544  * @flush - wether or not we can flush to make our reservation
3545  *
3546  * This will reserve orgi_bytes number of bytes from the space info associated
3547  * with the block_rsv.  If there is not enough space it will make an attempt to
3548  * flush out space to make room.  It will do this by flushing delalloc if
3549  * possible or committing the transaction.  If flush is 0 then no attempts to
3550  * regain reservations will be made and this will fail if there is not enough
3551  * space already.
3552  */
3553 static int reserve_metadata_bytes(struct btrfs_root *root,
3554                                   struct btrfs_block_rsv *block_rsv,
3555                                   u64 orig_bytes, int flush)
3556 {
3557         struct btrfs_space_info *space_info = block_rsv->space_info;
3558         u64 used;
3559         u64 num_bytes = orig_bytes;
3560         int retries = 0;
3561         int ret = 0;
3562         bool committed = false;
3563         bool flushing = false;
3564         bool wait_ordered = false;
3565
3566 again:
3567         ret = 0;
3568         spin_lock(&space_info->lock);
3569         /*
3570          * We only want to wait if somebody other than us is flushing and we are
3571          * actually alloed to flush.
3572          */
3573         while (flush && !flushing && space_info->flush) {
3574                 spin_unlock(&space_info->lock);
3575                 /*
3576                  * If we have a trans handle we can't wait because the flusher
3577                  * may have to commit the transaction, which would mean we would
3578                  * deadlock since we are waiting for the flusher to finish, but
3579                  * hold the current transaction open.
3580                  */
3581                 if (current->journal_info)
3582                         return -EAGAIN;
3583                 ret = wait_event_interruptible(space_info->wait,
3584                                                !space_info->flush);
3585                 /* Must have been interrupted, return */
3586                 if (ret)
3587                         return -EINTR;
3588
3589                 spin_lock(&space_info->lock);
3590         }
3591
3592         ret = -ENOSPC;
3593         used = space_info->bytes_used + space_info->bytes_reserved +
3594                 space_info->bytes_pinned + space_info->bytes_readonly +
3595                 space_info->bytes_may_use;
3596
3597         /*
3598          * The idea here is that we've not already over-reserved the block group
3599          * then we can go ahead and save our reservation first and then start
3600          * flushing if we need to.  Otherwise if we've already overcommitted
3601          * lets start flushing stuff first and then come back and try to make
3602          * our reservation.
3603          */
3604         if (used <= space_info->total_bytes) {
3605                 if (used + orig_bytes <= space_info->total_bytes) {
3606                         space_info->bytes_may_use += orig_bytes;
3607                         ret = 0;
3608                 } else {
3609                         /*
3610                          * Ok set num_bytes to orig_bytes since we aren't
3611                          * overocmmitted, this way we only try and reclaim what
3612                          * we need.
3613                          */
3614                         num_bytes = orig_bytes;
3615                 }
3616         } else {
3617                 /*
3618                  * Ok we're over committed, set num_bytes to the overcommitted
3619                  * amount plus the amount of bytes that we need for this
3620                  * reservation.
3621                  */
3622                 wait_ordered = true;
3623                 num_bytes = used - space_info->total_bytes +
3624                         (orig_bytes * (retries + 1));
3625         }
3626
3627         if (ret) {
3628                 u64 profile = btrfs_get_alloc_profile(root, 0);
3629                 u64 avail;
3630
3631                 /*
3632                  * If we have a lot of space that's pinned, don't bother doing
3633                  * the overcommit dance yet and just commit the transaction.
3634                  */
3635                 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3636                 do_div(avail, 10);
3637                 if (space_info->bytes_pinned >= avail && flush && !committed) {
3638                         space_info->flush = 1;
3639                         flushing = true;
3640                         spin_unlock(&space_info->lock);
3641                         ret = may_commit_transaction(root, space_info,
3642                                                      orig_bytes, 1);
3643                         if (ret)
3644                                 goto out;
3645                         committed = true;
3646                         goto again;
3647                 }
3648
3649                 spin_lock(&root->fs_info->free_chunk_lock);
3650                 avail = root->fs_info->free_chunk_space;
3651
3652                 /*
3653                  * If we have dup, raid1 or raid10 then only half of the free
3654                  * space is actually useable.
3655                  */
3656                 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3657                                BTRFS_BLOCK_GROUP_RAID1 |
3658                                BTRFS_BLOCK_GROUP_RAID10))
3659                         avail >>= 1;
3660
3661                 /*
3662                  * If we aren't flushing don't let us overcommit too much, say
3663                  * 1/8th of the space.  If we can flush, let it overcommit up to
3664                  * 1/2 of the space.
3665                  */
3666                 if (flush)
3667                         avail >>= 3;
3668                 else
3669                         avail >>= 1;
3670                  spin_unlock(&root->fs_info->free_chunk_lock);
3671
3672                 if (used + num_bytes < space_info->total_bytes + avail) {
3673                         space_info->bytes_may_use += orig_bytes;
3674                         ret = 0;
3675                 } else {
3676                         wait_ordered = true;
3677                 }
3678         }
3679
3680         /*
3681          * Couldn't make our reservation, save our place so while we're trying
3682          * to reclaim space we can actually use it instead of somebody else
3683          * stealing it from us.
3684          */
3685         if (ret && flush) {
3686                 flushing = true;
3687                 space_info->flush = 1;
3688         }
3689
3690         spin_unlock(&space_info->lock);
3691
3692         if (!ret || !flush)
3693                 goto out;
3694
3695         /*
3696          * We do synchronous shrinking since we don't actually unreserve
3697          * metadata until after the IO is completed.
3698          */
3699         ret = shrink_delalloc(root, num_bytes, wait_ordered);
3700         if (ret < 0)
3701                 goto out;
3702
3703         ret = 0;
3704
3705         /*
3706          * So if we were overcommitted it's possible that somebody else flushed
3707          * out enough space and we simply didn't have enough space to reclaim,
3708          * so go back around and try again.
3709          */
3710         if (retries < 2) {
3711                 wait_ordered = true;
3712                 retries++;
3713                 goto again;
3714         }
3715
3716         ret = -ENOSPC;
3717         if (committed)
3718                 goto out;
3719
3720         ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3721         if (!ret) {
3722                 committed = true;
3723                 goto again;
3724         }
3725
3726 out:
3727         if (flushing) {
3728                 spin_lock(&space_info->lock);
3729                 space_info->flush = 0;
3730                 wake_up_all(&space_info->wait);
3731                 spin_unlock(&space_info->lock);
3732         }
3733         return ret;
3734 }
3735
3736 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3737                                              struct btrfs_root *root)
3738 {
3739         struct btrfs_block_rsv *block_rsv = NULL;
3740
3741         if (root->ref_cows || root == root->fs_info->csum_root)
3742                 block_rsv = trans->block_rsv;
3743
3744         if (!block_rsv)
3745                 block_rsv = root->block_rsv;
3746
3747         if (!block_rsv)
3748                 block_rsv = &root->fs_info->empty_block_rsv;
3749
3750         return block_rsv;
3751 }
3752
3753 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3754                                u64 num_bytes)
3755 {
3756         int ret = -ENOSPC;
3757         spin_lock(&block_rsv->lock);
3758         if (block_rsv->reserved >= num_bytes) {
3759                 block_rsv->reserved -= num_bytes;
3760                 if (block_rsv->reserved < block_rsv->size)
3761                         block_rsv->full = 0;
3762                 ret = 0;
3763         }
3764         spin_unlock(&block_rsv->lock);
3765         return ret;
3766 }
3767
3768 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3769                                 u64 num_bytes, int update_size)
3770 {
3771         spin_lock(&block_rsv->lock);
3772         block_rsv->reserved += num_bytes;
3773         if (update_size)
3774                 block_rsv->size += num_bytes;
3775         else if (block_rsv->reserved >= block_rsv->size)
3776                 block_rsv->full = 1;
3777         spin_unlock(&block_rsv->lock);
3778 }
3779
3780 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3781                                     struct btrfs_block_rsv *dest, u64 num_bytes)
3782 {
3783         struct btrfs_space_info *space_info = block_rsv->space_info;
3784
3785         spin_lock(&block_rsv->lock);
3786         if (num_bytes == (u64)-1)
3787                 num_bytes = block_rsv->size;
3788         block_rsv->size -= num_bytes;
3789         if (block_rsv->reserved >= block_rsv->size) {
3790                 num_bytes = block_rsv->reserved - block_rsv->size;
3791                 block_rsv->reserved = block_rsv->size;
3792                 block_rsv->full = 1;
3793         } else {
3794                 num_bytes = 0;
3795         }
3796         spin_unlock(&block_rsv->lock);
3797
3798         if (num_bytes > 0) {
3799                 if (dest) {
3800                         spin_lock(&dest->lock);
3801                         if (!dest->full) {
3802                                 u64 bytes_to_add;
3803
3804                                 bytes_to_add = dest->size - dest->reserved;
3805                                 bytes_to_add = min(num_bytes, bytes_to_add);
3806                                 dest->reserved += bytes_to_add;
3807                                 if (dest->reserved >= dest->size)
3808                                         dest->full = 1;
3809                                 num_bytes -= bytes_to_add;
3810                         }
3811                         spin_unlock(&dest->lock);
3812                 }
3813                 if (num_bytes) {
3814                         spin_lock(&space_info->lock);
3815                         space_info->bytes_may_use -= num_bytes;
3816                         space_info->reservation_progress++;
3817                         spin_unlock(&space_info->lock);
3818                 }
3819         }
3820 }
3821
3822 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3823                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3824 {
3825         int ret;
3826
3827         ret = block_rsv_use_bytes(src, num_bytes);
3828         if (ret)
3829                 return ret;
3830
3831         block_rsv_add_bytes(dst, num_bytes, 1);
3832         return 0;
3833 }
3834
3835 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3836 {
3837         memset(rsv, 0, sizeof(*rsv));
3838         spin_lock_init(&rsv->lock);
3839 }
3840
3841 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3842 {
3843         struct btrfs_block_rsv *block_rsv;
3844         struct btrfs_fs_info *fs_info = root->fs_info;
3845
3846         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3847         if (!block_rsv)
3848                 return NULL;
3849
3850         btrfs_init_block_rsv(block_rsv);
3851         block_rsv->space_info = __find_space_info(fs_info,
3852                                                   BTRFS_BLOCK_GROUP_METADATA);
3853         return block_rsv;
3854 }
3855
3856 void btrfs_free_block_rsv(struct btrfs_root *root,
3857                           struct btrfs_block_rsv *rsv)
3858 {
3859         btrfs_block_rsv_release(root, rsv, (u64)-1);
3860         kfree(rsv);
3861 }
3862
3863 static inline int __block_rsv_add(struct btrfs_root *root,
3864                                   struct btrfs_block_rsv *block_rsv,
3865                                   u64 num_bytes, int flush)
3866 {
3867         int ret;
3868
3869         if (num_bytes == 0)
3870                 return 0;
3871
3872         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3873         if (!ret) {
3874                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3875                 return 0;
3876         }
3877
3878         return ret;
3879 }
3880
3881 int btrfs_block_rsv_add(struct btrfs_root *root,
3882                         struct btrfs_block_rsv *block_rsv,
3883                         u64 num_bytes)
3884 {
3885         return __block_rsv_add(root, block_rsv, num_bytes, 1);
3886 }
3887
3888 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3889                                 struct btrfs_block_rsv *block_rsv,
3890                                 u64 num_bytes)
3891 {
3892         return __block_rsv_add(root, block_rsv, num_bytes, 0);
3893 }
3894
3895 int btrfs_block_rsv_check(struct btrfs_root *root,
3896                           struct btrfs_block_rsv *block_rsv, int min_factor)
3897 {
3898         u64 num_bytes = 0;
3899         int ret = -ENOSPC;
3900
3901         if (!block_rsv)
3902                 return 0;
3903
3904         spin_lock(&block_rsv->lock);
3905         num_bytes = div_factor(block_rsv->size, min_factor);
3906         if (block_rsv->reserved >= num_bytes)
3907                 ret = 0;
3908         spin_unlock(&block_rsv->lock);
3909
3910         return ret;
3911 }
3912
3913 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
3914                                            struct btrfs_block_rsv *block_rsv,
3915                                            u64 min_reserved, int flush)
3916 {
3917         u64 num_bytes = 0;
3918         int ret = -ENOSPC;
3919
3920         if (!block_rsv)
3921                 return 0;
3922
3923         spin_lock(&block_rsv->lock);
3924         num_bytes = min_reserved;
3925         if (block_rsv->reserved >= num_bytes)
3926                 ret = 0;
3927         else
3928                 num_bytes -= block_rsv->reserved;
3929         spin_unlock(&block_rsv->lock);
3930
3931         if (!ret)
3932                 return 0;
3933
3934         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3935         if (!ret) {
3936                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3937                 return 0;
3938         }
3939
3940         return ret;
3941 }
3942
3943 int btrfs_block_rsv_refill(struct btrfs_root *root,
3944                            struct btrfs_block_rsv *block_rsv,
3945                            u64 min_reserved)
3946 {
3947         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
3948 }
3949
3950 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
3951                                    struct btrfs_block_rsv *block_rsv,
3952                                    u64 min_reserved)
3953 {
3954         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
3955 }
3956
3957 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3958                             struct btrfs_block_rsv *dst_rsv,
3959                             u64 num_bytes)
3960 {
3961         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3962 }
3963
3964 void btrfs_block_rsv_release(struct btrfs_root *root,
3965                              struct btrfs_block_rsv *block_rsv,
3966                              u64 num_bytes)
3967 {
3968         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3969         if (global_rsv->full || global_rsv == block_rsv ||
3970             block_rsv->space_info != global_rsv->space_info)
3971                 global_rsv = NULL;
3972         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3973 }
3974
3975 /*
3976  * helper to calculate size of global block reservation.
3977  * the desired value is sum of space used by extent tree,
3978  * checksum tree and root tree
3979  */
3980 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3981 {
3982         struct btrfs_space_info *sinfo;
3983         u64 num_bytes;
3984         u64 meta_used;
3985         u64 data_used;
3986         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
3987
3988         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3989         spin_lock(&sinfo->lock);
3990         data_used = sinfo->bytes_used;
3991         spin_unlock(&sinfo->lock);
3992
3993         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3994         spin_lock(&sinfo->lock);
3995         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3996                 data_used = 0;
3997         meta_used = sinfo->bytes_used;
3998         spin_unlock(&sinfo->lock);
3999
4000         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4001                     csum_size * 2;
4002         num_bytes += div64_u64(data_used + meta_used, 50);
4003
4004         if (num_bytes * 3 > meta_used)
4005                 num_bytes = div64_u64(meta_used, 3);
4006
4007         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4008 }
4009
4010 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4011 {
4012         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4013         struct btrfs_space_info *sinfo = block_rsv->space_info;
4014         u64 num_bytes;
4015
4016         num_bytes = calc_global_metadata_size(fs_info);
4017
4018         spin_lock(&block_rsv->lock);
4019         spin_lock(&sinfo->lock);
4020
4021         block_rsv->size = num_bytes;
4022
4023         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4024                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4025                     sinfo->bytes_may_use;
4026
4027         if (sinfo->total_bytes > num_bytes) {
4028                 num_bytes = sinfo->total_bytes - num_bytes;
4029                 block_rsv->reserved += num_bytes;
4030                 sinfo->bytes_may_use += num_bytes;
4031         }
4032
4033         if (block_rsv->reserved >= block_rsv->size) {
4034                 num_bytes = block_rsv->reserved - block_rsv->size;
4035                 sinfo->bytes_may_use -= num_bytes;
4036                 sinfo->reservation_progress++;
4037                 block_rsv->reserved = block_rsv->size;
4038                 block_rsv->full = 1;
4039         }
4040
4041         spin_unlock(&sinfo->lock);
4042         spin_unlock(&block_rsv->lock);
4043 }
4044
4045 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4046 {
4047         struct btrfs_space_info *space_info;
4048
4049         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4050         fs_info->chunk_block_rsv.space_info = space_info;
4051
4052         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4053         fs_info->global_block_rsv.space_info = space_info;
4054         fs_info->delalloc_block_rsv.space_info = space_info;
4055         fs_info->trans_block_rsv.space_info = space_info;
4056         fs_info->empty_block_rsv.space_info = space_info;
4057         fs_info->delayed_block_rsv.space_info = space_info;
4058
4059         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4060         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4061         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4062         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4063         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4064
4065         update_global_block_rsv(fs_info);
4066 }
4067
4068 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4069 {
4070         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
4071         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4072         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4073         WARN_ON(fs_info->trans_block_rsv.size > 0);
4074         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4075         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4076         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4077         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4078         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4079 }
4080
4081 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4082                                   struct btrfs_root *root)
4083 {
4084         if (!trans->bytes_reserved)
4085                 return;
4086
4087         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4088         trans->bytes_reserved = 0;
4089 }
4090
4091 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4092                                   struct inode *inode)
4093 {
4094         struct btrfs_root *root = BTRFS_I(inode)->root;
4095         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4096         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4097
4098         /*
4099          * We need to hold space in order to delete our orphan item once we've
4100          * added it, so this takes the reservation so we can release it later
4101          * when we are truly done with the orphan item.
4102          */
4103         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4104         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4105 }
4106
4107 void btrfs_orphan_release_metadata(struct inode *inode)
4108 {
4109         struct btrfs_root *root = BTRFS_I(inode)->root;
4110         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4111         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4112 }
4113
4114 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4115                                 struct btrfs_pending_snapshot *pending)
4116 {
4117         struct btrfs_root *root = pending->root;
4118         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4119         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4120         /*
4121          * two for root back/forward refs, two for directory entries
4122          * and one for root of the snapshot.
4123          */
4124         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4125         dst_rsv->space_info = src_rsv->space_info;
4126         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4127 }
4128
4129 /**
4130  * drop_outstanding_extent - drop an outstanding extent
4131  * @inode: the inode we're dropping the extent for
4132  *
4133  * This is called when we are freeing up an outstanding extent, either called
4134  * after an error or after an extent is written.  This will return the number of
4135  * reserved extents that need to be freed.  This must be called with
4136  * BTRFS_I(inode)->lock held.
4137  */
4138 static unsigned drop_outstanding_extent(struct inode *inode)
4139 {
4140         unsigned drop_inode_space = 0;
4141         unsigned dropped_extents = 0;
4142
4143         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4144         BTRFS_I(inode)->outstanding_extents--;
4145
4146         if (BTRFS_I(inode)->outstanding_extents == 0 &&
4147             BTRFS_I(inode)->delalloc_meta_reserved) {
4148                 drop_inode_space = 1;
4149                 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4150         }
4151
4152         /*
4153          * If we have more or the same amount of outsanding extents than we have
4154          * reserved then we need to leave the reserved extents count alone.
4155          */
4156         if (BTRFS_I(inode)->outstanding_extents >=
4157             BTRFS_I(inode)->reserved_extents)
4158                 return drop_inode_space;
4159
4160         dropped_extents = BTRFS_I(inode)->reserved_extents -
4161                 BTRFS_I(inode)->outstanding_extents;
4162         BTRFS_I(inode)->reserved_extents -= dropped_extents;
4163         return dropped_extents + drop_inode_space;
4164 }
4165
4166 /**
4167  * calc_csum_metadata_size - return the amount of metada space that must be
4168  *      reserved/free'd for the given bytes.
4169  * @inode: the inode we're manipulating
4170  * @num_bytes: the number of bytes in question
4171  * @reserve: 1 if we are reserving space, 0 if we are freeing space
4172  *
4173  * This adjusts the number of csum_bytes in the inode and then returns the
4174  * correct amount of metadata that must either be reserved or freed.  We
4175  * calculate how many checksums we can fit into one leaf and then divide the
4176  * number of bytes that will need to be checksumed by this value to figure out
4177  * how many checksums will be required.  If we are adding bytes then the number
4178  * may go up and we will return the number of additional bytes that must be
4179  * reserved.  If it is going down we will return the number of bytes that must
4180  * be freed.
4181  *
4182  * This must be called with BTRFS_I(inode)->lock held.
4183  */
4184 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4185                                    int reserve)
4186 {
4187         struct btrfs_root *root = BTRFS_I(inode)->root;
4188         u64 csum_size;
4189         int num_csums_per_leaf;
4190         int num_csums;
4191         int old_csums;
4192
4193         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4194             BTRFS_I(inode)->csum_bytes == 0)
4195                 return 0;
4196
4197         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4198         if (reserve)
4199                 BTRFS_I(inode)->csum_bytes += num_bytes;
4200         else
4201                 BTRFS_I(inode)->csum_bytes -= num_bytes;
4202         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4203         num_csums_per_leaf = (int)div64_u64(csum_size,
4204                                             sizeof(struct btrfs_csum_item) +
4205                                             sizeof(struct btrfs_disk_key));
4206         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4207         num_csums = num_csums + num_csums_per_leaf - 1;
4208         num_csums = num_csums / num_csums_per_leaf;
4209
4210         old_csums = old_csums + num_csums_per_leaf - 1;
4211         old_csums = old_csums / num_csums_per_leaf;
4212
4213         /* No change, no need to reserve more */
4214         if (old_csums == num_csums)
4215                 return 0;
4216
4217         if (reserve)
4218                 return btrfs_calc_trans_metadata_size(root,
4219                                                       num_csums - old_csums);
4220
4221         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4222 }
4223
4224 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4225 {
4226         struct btrfs_root *root = BTRFS_I(inode)->root;
4227         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4228         u64 to_reserve = 0;
4229         unsigned nr_extents = 0;
4230         int flush = 1;
4231         int ret;
4232
4233         if (btrfs_is_free_space_inode(root, inode))
4234                 flush = 0;
4235
4236         if (flush && btrfs_transaction_in_commit(root->fs_info))
4237                 schedule_timeout(1);
4238
4239         num_bytes = ALIGN(num_bytes, root->sectorsize);
4240
4241         spin_lock(&BTRFS_I(inode)->lock);
4242         BTRFS_I(inode)->outstanding_extents++;
4243
4244         if (BTRFS_I(inode)->outstanding_extents >
4245             BTRFS_I(inode)->reserved_extents) {
4246                 nr_extents = BTRFS_I(inode)->outstanding_extents -
4247                         BTRFS_I(inode)->reserved_extents;
4248                 BTRFS_I(inode)->reserved_extents += nr_extents;
4249         }
4250
4251         /*
4252          * Add an item to reserve for updating the inode when we complete the
4253          * delalloc io.
4254          */
4255         if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4256                 nr_extents++;
4257                 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4258         }
4259
4260         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4261         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4262         spin_unlock(&BTRFS_I(inode)->lock);
4263
4264         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4265         if (ret) {
4266                 u64 to_free = 0;
4267                 unsigned dropped;
4268
4269                 spin_lock(&BTRFS_I(inode)->lock);
4270                 dropped = drop_outstanding_extent(inode);
4271                 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4272                 spin_unlock(&BTRFS_I(inode)->lock);
4273                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4274
4275                 /*
4276                  * Somebody could have come in and twiddled with the
4277                  * reservation, so if we have to free more than we would have
4278                  * reserved from this reservation go ahead and release those
4279                  * bytes.
4280                  */
4281                 to_free -= to_reserve;
4282                 if (to_free)
4283                         btrfs_block_rsv_release(root, block_rsv, to_free);
4284                 return ret;
4285         }
4286
4287         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4288
4289         return 0;
4290 }
4291
4292 /**
4293  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4294  * @inode: the inode to release the reservation for
4295  * @num_bytes: the number of bytes we're releasing
4296  *
4297  * This will release the metadata reservation for an inode.  This can be called
4298  * once we complete IO for a given set of bytes to release their metadata
4299  * reservations.
4300  */
4301 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4302 {
4303         struct btrfs_root *root = BTRFS_I(inode)->root;
4304         u64 to_free = 0;
4305         unsigned dropped;
4306
4307         num_bytes = ALIGN(num_bytes, root->sectorsize);
4308         spin_lock(&BTRFS_I(inode)->lock);
4309         dropped = drop_outstanding_extent(inode);
4310
4311         to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4312         spin_unlock(&BTRFS_I(inode)->lock);
4313         if (dropped > 0)
4314                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4315
4316         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4317                                 to_free);
4318 }
4319
4320 /**
4321  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4322  * @inode: inode we're writing to
4323  * @num_bytes: the number of bytes we want to allocate
4324  *
4325  * This will do the following things
4326  *
4327  * o reserve space in the data space info for num_bytes
4328  * o reserve space in the metadata space info based on number of outstanding
4329  *   extents and how much csums will be needed
4330  * o add to the inodes ->delalloc_bytes
4331  * o add it to the fs_info's delalloc inodes list.
4332  *
4333  * This will return 0 for success and -ENOSPC if there is no space left.
4334  */
4335 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4336 {
4337         int ret;
4338
4339         ret = btrfs_check_data_free_space(inode, num_bytes);
4340         if (ret)
4341                 return ret;
4342
4343         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4344         if (ret) {
4345                 btrfs_free_reserved_data_space(inode, num_bytes);
4346                 return ret;
4347         }
4348
4349         return 0;
4350 }
4351
4352 /**
4353  * btrfs_delalloc_release_space - release data and metadata space for delalloc
4354  * @inode: inode we're releasing space for
4355  * @num_bytes: the number of bytes we want to free up
4356  *
4357  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
4358  * called in the case that we don't need the metadata AND data reservations
4359  * anymore.  So if there is an error or we insert an inline extent.
4360  *
4361  * This function will release the metadata space that was not used and will
4362  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4363  * list if there are no delalloc bytes left.
4364  */
4365 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4366 {
4367         btrfs_delalloc_release_metadata(inode, num_bytes);
4368         btrfs_free_reserved_data_space(inode, num_bytes);
4369 }
4370
4371 static int update_block_group(struct btrfs_trans_handle *trans,
4372                               struct btrfs_root *root,
4373                               u64 bytenr, u64 num_bytes, int alloc)
4374 {
4375         struct btrfs_block_group_cache *cache = NULL;
4376         struct btrfs_fs_info *info = root->fs_info;
4377         u64 total = num_bytes;
4378         u64 old_val;
4379         u64 byte_in_group;
4380         int factor;
4381
4382         /* block accounting for super block */
4383         spin_lock(&info->delalloc_lock);
4384         old_val = btrfs_super_bytes_used(info->super_copy);
4385         if (alloc)
4386                 old_val += num_bytes;
4387         else
4388                 old_val -= num_bytes;
4389         btrfs_set_super_bytes_used(info->super_copy, old_val);
4390         spin_unlock(&info->delalloc_lock);
4391
4392         while (total) {
4393                 cache = btrfs_lookup_block_group(info, bytenr);
4394                 if (!cache)
4395                         return -1;
4396                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4397                                     BTRFS_BLOCK_GROUP_RAID1 |
4398                                     BTRFS_BLOCK_GROUP_RAID10))
4399                         factor = 2;
4400                 else
4401                         factor = 1;
4402                 /*
4403                  * If this block group has free space cache written out, we
4404                  * need to make sure to load it if we are removing space.  This
4405                  * is because we need the unpinning stage to actually add the
4406                  * space back to the block group, otherwise we will leak space.
4407                  */
4408                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4409                         cache_block_group(cache, trans, NULL, 1);
4410
4411                 byte_in_group = bytenr - cache->key.objectid;
4412                 WARN_ON(byte_in_group > cache->key.offset);
4413
4414                 spin_lock(&cache->space_info->lock);
4415                 spin_lock(&cache->lock);
4416
4417                 if (btrfs_test_opt(root, SPACE_CACHE) &&
4418                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4419                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4420
4421                 cache->dirty = 1;
4422                 old_val = btrfs_block_group_used(&cache->item);
4423                 num_bytes = min(total, cache->key.offset - byte_in_group);
4424                 if (alloc) {
4425                         old_val += num_bytes;
4426                         btrfs_set_block_group_used(&cache->item, old_val);
4427                         cache->reserved -= num_bytes;
4428                         cache->space_info->bytes_reserved -= num_bytes;
4429                         cache->space_info->bytes_used += num_bytes;
4430                         cache->space_info->disk_used += num_bytes * factor;
4431                         spin_unlock(&cache->lock);
4432                         spin_unlock(&cache->space_info->lock);
4433                 } else {
4434                         old_val -= num_bytes;
4435                         btrfs_set_block_group_used(&cache->item, old_val);
4436                         cache->pinned += num_bytes;
4437                         cache->space_info->bytes_pinned += num_bytes;
4438                         cache->space_info->bytes_used -= num_bytes;
4439                         cache->space_info->disk_used -= num_bytes * factor;
4440                         spin_unlock(&cache->lock);
4441                         spin_unlock(&cache->space_info->lock);
4442
4443                         set_extent_dirty(info->pinned_extents,
4444                                          bytenr, bytenr + num_bytes - 1,
4445                                          GFP_NOFS | __GFP_NOFAIL);
4446                 }
4447                 btrfs_put_block_group(cache);
4448                 total -= num_bytes;
4449                 bytenr += num_bytes;
4450         }
4451         return 0;
4452 }
4453
4454 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4455 {
4456         struct btrfs_block_group_cache *cache;
4457         u64 bytenr;
4458
4459         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4460         if (!cache)
4461                 return 0;
4462
4463         bytenr = cache->key.objectid;
4464         btrfs_put_block_group(cache);
4465
4466         return bytenr;
4467 }
4468
4469 static int pin_down_extent(struct btrfs_root *root,
4470                            struct btrfs_block_group_cache *cache,
4471                            u64 bytenr, u64 num_bytes, int reserved)
4472 {
4473         spin_lock(&cache->space_info->lock);
4474         spin_lock(&cache->lock);
4475         cache->pinned += num_bytes;
4476         cache->space_info->bytes_pinned += num_bytes;
4477         if (reserved) {
4478                 cache->reserved -= num_bytes;
4479                 cache->space_info->bytes_reserved -= num_bytes;
4480         }
4481         spin_unlock(&cache->lock);
4482         spin_unlock(&cache->space_info->lock);
4483
4484         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4485                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4486         return 0;
4487 }
4488
4489 /*
4490  * this function must be called within transaction
4491  */
4492 int btrfs_pin_extent(struct btrfs_root *root,
4493                      u64 bytenr, u64 num_bytes, int reserved)
4494 {
4495         struct btrfs_block_group_cache *cache;
4496
4497         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4498         BUG_ON(!cache);
4499
4500         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4501
4502         btrfs_put_block_group(cache);
4503         return 0;
4504 }
4505
4506 /*
4507  * this function must be called within transaction
4508  */
4509 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4510                                     struct btrfs_root *root,
4511                                     u64 bytenr, u64 num_bytes)
4512 {
4513         struct btrfs_block_group_cache *cache;
4514
4515         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4516         BUG_ON(!cache);
4517
4518         /*
4519          * pull in the free space cache (if any) so that our pin
4520          * removes the free space from the cache.  We have load_only set
4521          * to one because the slow code to read in the free extents does check
4522          * the pinned extents.
4523          */
4524         cache_block_group(cache, trans, root, 1);
4525
4526         pin_down_extent(root, cache, bytenr, num_bytes, 0);
4527
4528         /* remove us from the free space cache (if we're there at all) */
4529         btrfs_remove_free_space(cache, bytenr, num_bytes);
4530         btrfs_put_block_group(cache);
4531         return 0;
4532 }
4533
4534 /**
4535  * btrfs_update_reserved_bytes - update the block_group and space info counters
4536  * @cache:      The cache we are manipulating
4537  * @num_bytes:  The number of bytes in question
4538  * @reserve:    One of the reservation enums
4539  *
4540  * This is called by the allocator when it reserves space, or by somebody who is
4541  * freeing space that was never actually used on disk.  For example if you
4542  * reserve some space for a new leaf in transaction A and before transaction A
4543  * commits you free that leaf, you call this with reserve set to 0 in order to
4544  * clear the reservation.
4545  *
4546  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4547  * ENOSPC accounting.  For data we handle the reservation through clearing the
4548  * delalloc bits in the io_tree.  We have to do this since we could end up
4549  * allocating less disk space for the amount of data we have reserved in the
4550  * case of compression.
4551  *
4552  * If this is a reservation and the block group has become read only we cannot
4553  * make the reservation and return -EAGAIN, otherwise this function always
4554  * succeeds.
4555  */
4556 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4557                                        u64 num_bytes, int reserve)
4558 {
4559         struct btrfs_space_info *space_info = cache->space_info;
4560         int ret = 0;
4561         spin_lock(&space_info->lock);
4562         spin_lock(&cache->lock);
4563         if (reserve != RESERVE_FREE) {
4564                 if (cache->ro) {
4565                         ret = -EAGAIN;
4566                 } else {
4567                         cache->reserved += num_bytes;
4568                         space_info->bytes_reserved += num_bytes;
4569                         if (reserve == RESERVE_ALLOC) {
4570                                 BUG_ON(space_info->bytes_may_use < num_bytes);
4571                                 space_info->bytes_may_use -= num_bytes;
4572                         }
4573                 }
4574         } else {
4575                 if (cache->ro)
4576                         space_info->bytes_readonly += num_bytes;
4577                 cache->reserved -= num_bytes;
4578                 space_info->bytes_reserved -= num_bytes;
4579                 space_info->reservation_progress++;
4580         }
4581         spin_unlock(&cache->lock);
4582         spin_unlock(&space_info->lock);
4583         return ret;
4584 }
4585
4586 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4587                                 struct btrfs_root *root)
4588 {
4589         struct btrfs_fs_info *fs_info = root->fs_info;
4590         struct btrfs_caching_control *next;
4591         struct btrfs_caching_control *caching_ctl;
4592         struct btrfs_block_group_cache *cache;
4593
4594         down_write(&fs_info->extent_commit_sem);
4595
4596         list_for_each_entry_safe(caching_ctl, next,
4597                                  &fs_info->caching_block_groups, list) {
4598                 cache = caching_ctl->block_group;
4599                 if (block_group_cache_done(cache)) {
4600                         cache->last_byte_to_unpin = (u64)-1;
4601                         list_del_init(&caching_ctl->list);
4602                         put_caching_control(caching_ctl);
4603                 } else {
4604                         cache->last_byte_to_unpin = caching_ctl->progress;
4605                 }
4606         }
4607
4608         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4609                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4610         else
4611                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4612
4613         up_write(&fs_info->extent_commit_sem);
4614
4615         update_global_block_rsv(fs_info);
4616         return 0;
4617 }
4618
4619 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4620 {
4621         struct btrfs_fs_info *fs_info = root->fs_info;
4622         struct btrfs_block_group_cache *cache = NULL;
4623         u64 len;
4624
4625         while (start <= end) {
4626                 if (!cache ||
4627                     start >= cache->key.objectid + cache->key.offset) {
4628                         if (cache)
4629                                 btrfs_put_block_group(cache);
4630                         cache = btrfs_lookup_block_group(fs_info, start);
4631                         BUG_ON(!cache);
4632                 }
4633
4634                 len = cache->key.objectid + cache->key.offset - start;
4635                 len = min(len, end + 1 - start);
4636
4637                 if (start < cache->last_byte_to_unpin) {
4638                         len = min(len, cache->last_byte_to_unpin - start);
4639                         btrfs_add_free_space(cache, start, len);
4640                 }
4641
4642                 start += len;
4643
4644                 spin_lock(&cache->space_info->lock);
4645                 spin_lock(&cache->lock);
4646                 cache->pinned -= len;
4647                 cache->space_info->bytes_pinned -= len;
4648                 if (cache->ro)
4649                         cache->space_info->bytes_readonly += len;
4650                 spin_unlock(&cache->lock);
4651                 spin_unlock(&cache->space_info->lock);
4652         }
4653
4654         if (cache)
4655                 btrfs_put_block_group(cache);
4656         return 0;
4657 }
4658
4659 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4660                                struct btrfs_root *root)
4661 {
4662         struct btrfs_fs_info *fs_info = root->fs_info;
4663         struct extent_io_tree *unpin;
4664         u64 start;
4665         u64 end;
4666         int ret;
4667
4668         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4669                 unpin = &fs_info->freed_extents[1];
4670         else
4671                 unpin = &fs_info->freed_extents[0];
4672
4673         while (1) {
4674                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4675                                             EXTENT_DIRTY);
4676                 if (ret)
4677                         break;
4678
4679                 if (btrfs_test_opt(root, DISCARD))
4680                         ret = btrfs_discard_extent(root, start,
4681                                                    end + 1 - start, NULL);
4682
4683                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4684                 unpin_extent_range(root, start, end);
4685                 cond_resched();
4686         }
4687
4688         return 0;
4689 }
4690
4691 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4692                                 struct btrfs_root *root,
4693                                 u64 bytenr, u64 num_bytes, u64 parent,
4694                                 u64 root_objectid, u64 owner_objectid,
4695                                 u64 owner_offset, int refs_to_drop,
4696                                 struct btrfs_delayed_extent_op *extent_op)
4697 {
4698         struct btrfs_key key;
4699         struct btrfs_path *path;
4700         struct btrfs_fs_info *info = root->fs_info;
4701         struct btrfs_root *extent_root = info->extent_root;
4702         struct extent_buffer *leaf;
4703         struct btrfs_extent_item *ei;
4704         struct btrfs_extent_inline_ref *iref;
4705         int ret;
4706         int is_data;
4707         int extent_slot = 0;
4708         int found_extent = 0;
4709         int num_to_del = 1;
4710         u32 item_size;
4711         u64 refs;
4712
4713         path = btrfs_alloc_path();
4714         if (!path)
4715                 return -ENOMEM;
4716
4717         path->reada = 1;
4718         path->leave_spinning = 1;
4719
4720         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4721         BUG_ON(!is_data && refs_to_drop != 1);
4722
4723         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4724                                     bytenr, num_bytes, parent,
4725                                     root_objectid, owner_objectid,
4726                                     owner_offset);
4727         if (ret == 0) {
4728                 extent_slot = path->slots[0];
4729                 while (extent_slot >= 0) {
4730                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4731                                               extent_slot);
4732                         if (key.objectid != bytenr)
4733                                 break;
4734                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4735                             key.offset == num_bytes) {
4736                                 found_extent = 1;
4737                                 break;
4738                         }
4739                         if (path->slots[0] - extent_slot > 5)
4740                                 break;
4741                         extent_slot--;
4742                 }
4743 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4744                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4745                 if (found_extent && item_size < sizeof(*ei))
4746                         found_extent = 0;
4747 #endif
4748                 if (!found_extent) {
4749                         BUG_ON(iref);
4750                         ret = remove_extent_backref(trans, extent_root, path,
4751                                                     NULL, refs_to_drop,
4752                                                     is_data);
4753                         BUG_ON(ret);
4754                         btrfs_release_path(path);
4755                         path->leave_spinning = 1;
4756
4757                         key.objectid = bytenr;
4758                         key.type = BTRFS_EXTENT_ITEM_KEY;
4759                         key.offset = num_bytes;
4760
4761                         ret = btrfs_search_slot(trans, extent_root,
4762                                                 &key, path, -1, 1);
4763                         if (ret) {
4764                                 printk(KERN_ERR "umm, got %d back from search"
4765                                        ", was looking for %llu\n", ret,
4766                                        (unsigned long long)bytenr);
4767                                 if (ret > 0)
4768                                         btrfs_print_leaf(extent_root,
4769                                                          path->nodes[0]);
4770                         }
4771                         BUG_ON(ret);
4772                         extent_slot = path->slots[0];
4773                 }
4774         } else {
4775                 btrfs_print_leaf(extent_root, path->nodes[0]);
4776                 WARN_ON(1);
4777                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4778                        "parent %llu root %llu  owner %llu offset %llu\n",
4779                        (unsigned long long)bytenr,
4780                        (unsigned long long)parent,
4781                        (unsigned long long)root_objectid,
4782                        (unsigned long long)owner_objectid,
4783                        (unsigned long long)owner_offset);
4784         }
4785
4786         leaf = path->nodes[0];
4787         item_size = btrfs_item_size_nr(leaf, extent_slot);
4788 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4789         if (item_size < sizeof(*ei)) {
4790                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4791                 ret = convert_extent_item_v0(trans, extent_root, path,
4792                                              owner_objectid, 0);
4793                 BUG_ON(ret < 0);
4794
4795                 btrfs_release_path(path);
4796                 path->leave_spinning = 1;
4797
4798                 key.objectid = bytenr;
4799                 key.type = BTRFS_EXTENT_ITEM_KEY;
4800                 key.offset = num_bytes;
4801
4802                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4803                                         -1, 1);
4804                 if (ret) {
4805                         printk(KERN_ERR "umm, got %d back from search"
4806                                ", was looking for %llu\n", ret,
4807                                (unsigned long long)bytenr);
4808                         btrfs_print_leaf(extent_root, path->nodes[0]);
4809                 }
4810                 BUG_ON(ret);
4811                 extent_slot = path->slots[0];
4812                 leaf = path->nodes[0];
4813                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4814         }
4815 #endif
4816         BUG_ON(item_size < sizeof(*ei));
4817         ei = btrfs_item_ptr(leaf, extent_slot,
4818                             struct btrfs_extent_item);
4819         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4820                 struct btrfs_tree_block_info *bi;
4821                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4822                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4823                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4824         }
4825
4826         refs = btrfs_extent_refs(leaf, ei);
4827         BUG_ON(refs < refs_to_drop);
4828         refs -= refs_to_drop;
4829
4830         if (refs > 0) {
4831                 if (extent_op)
4832                         __run_delayed_extent_op(extent_op, leaf, ei);
4833                 /*
4834                  * In the case of inline back ref, reference count will
4835                  * be updated by remove_extent_backref
4836                  */
4837                 if (iref) {
4838                         BUG_ON(!found_extent);
4839                 } else {
4840                         btrfs_set_extent_refs(leaf, ei, refs);
4841                         btrfs_mark_buffer_dirty(leaf);
4842                 }
4843                 if (found_extent) {
4844                         ret = remove_extent_backref(trans, extent_root, path,
4845                                                     iref, refs_to_drop,
4846                                                     is_data);
4847                         BUG_ON(ret);
4848                 }
4849         } else {
4850                 if (found_extent) {
4851                         BUG_ON(is_data && refs_to_drop !=
4852                                extent_data_ref_count(root, path, iref));
4853                         if (iref) {
4854                                 BUG_ON(path->slots[0] != extent_slot);
4855                         } else {
4856                                 BUG_ON(path->slots[0] != extent_slot + 1);
4857                                 path->slots[0] = extent_slot;
4858                                 num_to_del = 2;
4859                         }
4860                 }
4861
4862                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4863                                       num_to_del);
4864                 BUG_ON(ret);
4865                 btrfs_release_path(path);
4866
4867                 if (is_data) {
4868                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4869                         BUG_ON(ret);
4870                 } else {
4871                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4872                              bytenr >> PAGE_CACHE_SHIFT,
4873                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4874                 }
4875
4876                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4877                 BUG_ON(ret);
4878         }
4879         btrfs_free_path(path);
4880         return ret;
4881 }
4882
4883 /*
4884  * when we free an block, it is possible (and likely) that we free the last
4885  * delayed ref for that extent as well.  This searches the delayed ref tree for
4886  * a given extent, and if there are no other delayed refs to be processed, it
4887  * removes it from the tree.
4888  */
4889 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4890                                       struct btrfs_root *root, u64 bytenr)
4891 {
4892         struct btrfs_delayed_ref_head *head;
4893         struct btrfs_delayed_ref_root *delayed_refs;
4894         struct btrfs_delayed_ref_node *ref;
4895         struct rb_node *node;
4896         int ret = 0;
4897
4898         delayed_refs = &trans->transaction->delayed_refs;
4899         spin_lock(&delayed_refs->lock);
4900         head = btrfs_find_delayed_ref_head(trans, bytenr);
4901         if (!head)
4902                 goto out;
4903
4904         node = rb_prev(&head->node.rb_node);
4905         if (!node)
4906                 goto out;
4907
4908         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4909
4910         /* there are still entries for this ref, we can't drop it */
4911         if (ref->bytenr == bytenr)
4912                 goto out;
4913
4914         if (head->extent_op) {
4915                 if (!head->must_insert_reserved)
4916                         goto out;
4917                 kfree(head->extent_op);
4918                 head->extent_op = NULL;
4919         }
4920
4921         /*
4922          * waiting for the lock here would deadlock.  If someone else has it
4923          * locked they are already in the process of dropping it anyway
4924          */
4925         if (!mutex_trylock(&head->mutex))
4926                 goto out;
4927
4928         /*
4929          * at this point we have a head with no other entries.  Go
4930          * ahead and process it.
4931          */
4932         head->node.in_tree = 0;
4933         rb_erase(&head->node.rb_node, &delayed_refs->root);
4934
4935         delayed_refs->num_entries--;
4936
4937         /*
4938          * we don't take a ref on the node because we're removing it from the
4939          * tree, so we just steal the ref the tree was holding.
4940          */
4941         delayed_refs->num_heads--;
4942         if (list_empty(&head->cluster))
4943                 delayed_refs->num_heads_ready--;
4944
4945         list_del_init(&head->cluster);
4946         spin_unlock(&delayed_refs->lock);
4947
4948         BUG_ON(head->extent_op);
4949         if (head->must_insert_reserved)
4950                 ret = 1;
4951
4952         mutex_unlock(&head->mutex);
4953         btrfs_put_delayed_ref(&head->node);
4954         return ret;
4955 out:
4956         spin_unlock(&delayed_refs->lock);
4957         return 0;
4958 }
4959
4960 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4961                            struct btrfs_root *root,
4962                            struct extent_buffer *buf,
4963                            u64 parent, int last_ref, int for_cow)
4964 {
4965         struct btrfs_block_group_cache *cache = NULL;
4966         int ret;
4967
4968         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4969                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
4970                                         buf->start, buf->len,
4971                                         parent, root->root_key.objectid,
4972                                         btrfs_header_level(buf),
4973                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
4974                 BUG_ON(ret);
4975         }
4976
4977         if (!last_ref)
4978                 return;
4979
4980         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4981
4982         if (btrfs_header_generation(buf) == trans->transid) {
4983                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4984                         ret = check_ref_cleanup(trans, root, buf->start);
4985                         if (!ret)
4986                                 goto out;
4987                 }
4988
4989                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4990                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4991                         goto out;
4992                 }
4993
4994                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4995
4996                 btrfs_add_free_space(cache, buf->start, buf->len);
4997                 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
4998         }
4999 out:
5000         /*
5001          * Deleting the buffer, clear the corrupt flag since it doesn't matter
5002          * anymore.
5003          */
5004         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5005         btrfs_put_block_group(cache);
5006 }
5007
5008 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5009                       u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5010                       u64 owner, u64 offset, int for_cow)
5011 {
5012         int ret;
5013         struct btrfs_fs_info *fs_info = root->fs_info;
5014
5015         /*
5016          * tree log blocks never actually go into the extent allocation
5017          * tree, just update pinning info and exit early.
5018          */
5019         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5020                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5021                 /* unlocks the pinned mutex */
5022                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5023                 ret = 0;
5024         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5025                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5026                                         num_bytes,
5027                                         parent, root_objectid, (int)owner,
5028                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5029                 BUG_ON(ret);
5030         } else {
5031                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5032                                                 num_bytes,
5033                                                 parent, root_objectid, owner,
5034                                                 offset, BTRFS_DROP_DELAYED_REF,
5035                                                 NULL, for_cow);
5036                 BUG_ON(ret);
5037         }
5038         return ret;
5039 }
5040
5041 static u64 stripe_align(struct btrfs_root *root, u64 val)
5042 {
5043         u64 mask = ((u64)root->stripesize - 1);
5044         u64 ret = (val + mask) & ~mask;
5045         return ret;
5046 }
5047
5048 /*
5049  * when we wait for progress in the block group caching, its because
5050  * our allocation attempt failed at least once.  So, we must sleep
5051  * and let some progress happen before we try again.
5052  *
5053  * This function will sleep at least once waiting for new free space to
5054  * show up, and then it will check the block group free space numbers
5055  * for our min num_bytes.  Another option is to have it go ahead
5056  * and look in the rbtree for a free extent of a given size, but this
5057  * is a good start.
5058  */
5059 static noinline int
5060 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5061                                 u64 num_bytes)
5062 {
5063         struct btrfs_caching_control *caching_ctl;
5064         DEFINE_WAIT(wait);
5065
5066         caching_ctl = get_caching_control(cache);
5067         if (!caching_ctl)
5068                 return 0;
5069
5070         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5071                    (cache->free_space_ctl->free_space >= num_bytes));
5072
5073         put_caching_control(caching_ctl);
5074         return 0;
5075 }
5076
5077 static noinline int
5078 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5079 {
5080         struct btrfs_caching_control *caching_ctl;
5081         DEFINE_WAIT(wait);
5082
5083         caching_ctl = get_caching_control(cache);
5084         if (!caching_ctl)
5085                 return 0;
5086
5087         wait_event(caching_ctl->wait, block_group_cache_done(cache));
5088
5089         put_caching_control(caching_ctl);
5090         return 0;
5091 }
5092
5093 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5094 {
5095         int index;
5096         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5097                 index = 0;
5098         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5099                 index = 1;
5100         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5101                 index = 2;
5102         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5103                 index = 3;
5104         else
5105                 index = 4;
5106         return index;
5107 }
5108
5109 enum btrfs_loop_type {
5110         LOOP_FIND_IDEAL = 0,
5111         LOOP_CACHING_NOWAIT = 1,
5112         LOOP_CACHING_WAIT = 2,
5113         LOOP_ALLOC_CHUNK = 3,
5114         LOOP_NO_EMPTY_SIZE = 4,
5115 };
5116
5117 /*
5118  * walks the btree of allocated extents and find a hole of a given size.
5119  * The key ins is changed to record the hole:
5120  * ins->objectid == block start
5121  * ins->flags = BTRFS_EXTENT_ITEM_KEY
5122  * ins->offset == number of blocks
5123  * Any available blocks before search_start are skipped.
5124  */
5125 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5126                                      struct btrfs_root *orig_root,
5127                                      u64 num_bytes, u64 empty_size,
5128                                      u64 search_start, u64 search_end,
5129                                      u64 hint_byte, struct btrfs_key *ins,
5130                                      u64 data)
5131 {
5132         int ret = 0;
5133         struct btrfs_root *root = orig_root->fs_info->extent_root;
5134         struct btrfs_free_cluster *last_ptr = NULL;
5135         struct btrfs_block_group_cache *block_group = NULL;
5136         int empty_cluster = 2 * 1024 * 1024;
5137         int allowed_chunk_alloc = 0;
5138         int done_chunk_alloc = 0;
5139         struct btrfs_space_info *space_info;
5140         int last_ptr_loop = 0;
5141         int loop = 0;
5142         int index = 0;
5143         int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5144                 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5145         bool found_uncached_bg = false;
5146         bool failed_cluster_refill = false;
5147         bool failed_alloc = false;
5148         bool use_cluster = true;
5149         bool have_caching_bg = false;
5150         u64 ideal_cache_percent = 0;
5151         u64 ideal_cache_offset = 0;
5152
5153         WARN_ON(num_bytes < root->sectorsize);
5154         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5155         ins->objectid = 0;
5156         ins->offset = 0;
5157
5158         space_info = __find_space_info(root->fs_info, data);
5159         if (!space_info) {
5160                 printk(KERN_ERR "No space info for %llu\n", data);
5161                 return -ENOSPC;
5162         }
5163
5164         /*
5165          * If the space info is for both data and metadata it means we have a
5166          * small filesystem and we can't use the clustering stuff.
5167          */
5168         if (btrfs_mixed_space_info(space_info))
5169                 use_cluster = false;
5170
5171         if (orig_root->ref_cows || empty_size)
5172                 allowed_chunk_alloc = 1;
5173
5174         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5175                 last_ptr = &root->fs_info->meta_alloc_cluster;
5176                 if (!btrfs_test_opt(root, SSD))
5177                         empty_cluster = 64 * 1024;
5178         }
5179
5180         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5181             btrfs_test_opt(root, SSD)) {
5182                 last_ptr = &root->fs_info->data_alloc_cluster;
5183         }
5184
5185         if (last_ptr) {
5186                 spin_lock(&last_ptr->lock);
5187                 if (last_ptr->block_group)
5188                         hint_byte = last_ptr->window_start;
5189                 spin_unlock(&last_ptr->lock);
5190         }
5191
5192         search_start = max(search_start, first_logical_byte(root, 0));
5193         search_start = max(search_start, hint_byte);
5194
5195         if (!last_ptr)
5196                 empty_cluster = 0;
5197
5198         if (search_start == hint_byte) {
5199 ideal_cache:
5200                 block_group = btrfs_lookup_block_group(root->fs_info,
5201                                                        search_start);
5202                 /*
5203                  * we don't want to use the block group if it doesn't match our
5204                  * allocation bits, or if its not cached.
5205                  *
5206                  * However if we are re-searching with an ideal block group
5207                  * picked out then we don't care that the block group is cached.
5208                  */
5209                 if (block_group && block_group_bits(block_group, data) &&
5210                     (block_group->cached != BTRFS_CACHE_NO ||
5211                      search_start == ideal_cache_offset)) {
5212                         down_read(&space_info->groups_sem);
5213                         if (list_empty(&block_group->list) ||
5214                             block_group->ro) {
5215                                 /*
5216                                  * someone is removing this block group,
5217                                  * we can't jump into the have_block_group
5218                                  * target because our list pointers are not
5219                                  * valid
5220                                  */
5221                                 btrfs_put_block_group(block_group);
5222                                 up_read(&space_info->groups_sem);
5223                         } else {
5224                                 index = get_block_group_index(block_group);
5225                                 goto have_block_group;
5226                         }
5227                 } else if (block_group) {
5228                         btrfs_put_block_group(block_group);
5229                 }
5230         }
5231 search:
5232         have_caching_bg = false;
5233         down_read(&space_info->groups_sem);
5234         list_for_each_entry(block_group, &space_info->block_groups[index],
5235                             list) {
5236                 u64 offset;
5237                 int cached;
5238
5239                 btrfs_get_block_group(block_group);
5240                 search_start = block_group->key.objectid;
5241
5242                 /*
5243                  * this can happen if we end up cycling through all the
5244                  * raid types, but we want to make sure we only allocate
5245                  * for the proper type.
5246                  */
5247                 if (!block_group_bits(block_group, data)) {
5248                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
5249                                 BTRFS_BLOCK_GROUP_RAID1 |
5250                                 BTRFS_BLOCK_GROUP_RAID10;
5251
5252                         /*
5253                          * if they asked for extra copies and this block group
5254                          * doesn't provide them, bail.  This does allow us to
5255                          * fill raid0 from raid1.
5256                          */
5257                         if ((data & extra) && !(block_group->flags & extra))
5258                                 goto loop;
5259                 }
5260
5261 have_block_group:
5262                 cached = block_group_cache_done(block_group);
5263                 if (unlikely(!cached)) {
5264                         u64 free_percent;
5265
5266                         found_uncached_bg = true;
5267                         ret = cache_block_group(block_group, trans,
5268                                                 orig_root, 1);
5269                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5270                                 goto alloc;
5271
5272                         free_percent = btrfs_block_group_used(&block_group->item);
5273                         free_percent *= 100;
5274                         free_percent = div64_u64(free_percent,
5275                                                  block_group->key.offset);
5276                         free_percent = 100 - free_percent;
5277                         if (free_percent > ideal_cache_percent &&
5278                             likely(!block_group->ro)) {
5279                                 ideal_cache_offset = block_group->key.objectid;
5280                                 ideal_cache_percent = free_percent;
5281                         }
5282
5283                         /*
5284                          * The caching workers are limited to 2 threads, so we
5285                          * can queue as much work as we care to.
5286                          */
5287                         if (loop > LOOP_FIND_IDEAL) {
5288                                 ret = cache_block_group(block_group, trans,
5289                                                         orig_root, 0);
5290                                 BUG_ON(ret);
5291                         }
5292
5293                         /*
5294                          * If loop is set for cached only, try the next block
5295                          * group.
5296                          */
5297                         if (loop == LOOP_FIND_IDEAL)
5298                                 goto loop;
5299                 }
5300
5301 alloc:
5302                 if (unlikely(block_group->ro))
5303                         goto loop;
5304
5305                 spin_lock(&block_group->free_space_ctl->tree_lock);
5306                 if (cached &&
5307                     block_group->free_space_ctl->free_space <
5308                     num_bytes + empty_cluster + empty_size) {
5309                         spin_unlock(&block_group->free_space_ctl->tree_lock);
5310                         goto loop;
5311                 }
5312                 spin_unlock(&block_group->free_space_ctl->tree_lock);
5313
5314                 /*
5315                  * Ok we want to try and use the cluster allocator, so lets look
5316                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5317                  * have tried the cluster allocator plenty of times at this
5318                  * point and not have found anything, so we are likely way too
5319                  * fragmented for the clustering stuff to find anything, so lets
5320                  * just skip it and let the allocator find whatever block it can
5321                  * find
5322                  */
5323                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5324                         /*
5325                          * the refill lock keeps out other
5326                          * people trying to start a new cluster
5327                          */
5328                         spin_lock(&last_ptr->refill_lock);
5329                         if (!last_ptr->block_group ||
5330                             last_ptr->block_group->ro ||
5331                             !block_group_bits(last_ptr->block_group, data))
5332                                 goto refill_cluster;
5333
5334                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5335                                                  num_bytes, search_start);
5336                         if (offset) {
5337                                 /* we have a block, we're done */
5338                                 spin_unlock(&last_ptr->refill_lock);
5339                                 goto checks;
5340                         }
5341
5342                         spin_lock(&last_ptr->lock);
5343                         /*
5344                          * whoops, this cluster doesn't actually point to
5345                          * this block group.  Get a ref on the block
5346                          * group is does point to and try again
5347                          */
5348                         if (!last_ptr_loop && last_ptr->block_group &&
5349                             last_ptr->block_group != block_group &&
5350                             index <=
5351                                  get_block_group_index(last_ptr->block_group)) {
5352
5353                                 btrfs_put_block_group(block_group);
5354                                 block_group = last_ptr->block_group;
5355                                 btrfs_get_block_group(block_group);
5356                                 spin_unlock(&last_ptr->lock);
5357                                 spin_unlock(&last_ptr->refill_lock);
5358
5359                                 last_ptr_loop = 1;
5360                                 search_start = block_group->key.objectid;
5361                                 /*
5362                                  * we know this block group is properly
5363                                  * in the list because
5364                                  * btrfs_remove_block_group, drops the
5365                                  * cluster before it removes the block
5366                                  * group from the list
5367                                  */
5368                                 goto have_block_group;
5369                         }
5370                         spin_unlock(&last_ptr->lock);
5371 refill_cluster:
5372                         /*
5373                          * this cluster didn't work out, free it and
5374                          * start over
5375                          */
5376                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5377
5378                         last_ptr_loop = 0;
5379
5380                         /* allocate a cluster in this block group */
5381                         ret = btrfs_find_space_cluster(trans, root,
5382                                                block_group, last_ptr,
5383                                                search_start, num_bytes,
5384                                                empty_cluster + empty_size);
5385                         if (ret == 0) {
5386                                 /*
5387                                  * now pull our allocation out of this
5388                                  * cluster
5389                                  */
5390                                 offset = btrfs_alloc_from_cluster(block_group,
5391                                                   last_ptr, num_bytes,
5392                                                   search_start);
5393                                 if (offset) {
5394                                         /* we found one, proceed */
5395                                         spin_unlock(&last_ptr->refill_lock);
5396                                         goto checks;
5397                                 }
5398                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5399                                    && !failed_cluster_refill) {
5400                                 spin_unlock(&last_ptr->refill_lock);
5401
5402                                 failed_cluster_refill = true;
5403                                 wait_block_group_cache_progress(block_group,
5404                                        num_bytes + empty_cluster + empty_size);
5405                                 goto have_block_group;
5406                         }
5407
5408                         /*
5409                          * at this point we either didn't find a cluster
5410                          * or we weren't able to allocate a block from our
5411                          * cluster.  Free the cluster we've been trying
5412                          * to use, and go to the next block group
5413                          */
5414                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5415                         spin_unlock(&last_ptr->refill_lock);
5416                         goto loop;
5417                 }
5418
5419                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5420                                                     num_bytes, empty_size);
5421                 /*
5422                  * If we didn't find a chunk, and we haven't failed on this
5423                  * block group before, and this block group is in the middle of
5424                  * caching and we are ok with waiting, then go ahead and wait
5425                  * for progress to be made, and set failed_alloc to true.
5426                  *
5427                  * If failed_alloc is true then we've already waited on this
5428                  * block group once and should move on to the next block group.
5429                  */
5430                 if (!offset && !failed_alloc && !cached &&
5431                     loop > LOOP_CACHING_NOWAIT) {
5432                         wait_block_group_cache_progress(block_group,
5433                                                 num_bytes + empty_size);
5434                         failed_alloc = true;
5435                         goto have_block_group;
5436                 } else if (!offset) {
5437                         if (!cached)
5438                                 have_caching_bg = true;
5439                         goto loop;
5440                 }
5441 checks:
5442                 search_start = stripe_align(root, offset);
5443                 /* move on to the next group */
5444                 if (search_start + num_bytes >= search_end) {
5445                         btrfs_add_free_space(block_group, offset, num_bytes);
5446                         goto loop;
5447                 }
5448
5449                 /* move on to the next group */
5450                 if (search_start + num_bytes >
5451                     block_group->key.objectid + block_group->key.offset) {
5452                         btrfs_add_free_space(block_group, offset, num_bytes);
5453                         goto loop;
5454                 }
5455
5456                 ins->objectid = search_start;
5457                 ins->offset = num_bytes;
5458
5459                 if (offset < search_start)
5460                         btrfs_add_free_space(block_group, offset,
5461                                              search_start - offset);
5462                 BUG_ON(offset > search_start);
5463
5464                 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
5465                                                   alloc_type);
5466                 if (ret == -EAGAIN) {
5467                         btrfs_add_free_space(block_group, offset, num_bytes);
5468                         goto loop;
5469                 }
5470
5471                 /* we are all good, lets return */
5472                 ins->objectid = search_start;
5473                 ins->offset = num_bytes;
5474
5475                 if (offset < search_start)
5476                         btrfs_add_free_space(block_group, offset,
5477                                              search_start - offset);
5478                 BUG_ON(offset > search_start);
5479                 btrfs_put_block_group(block_group);
5480                 break;
5481 loop:
5482                 failed_cluster_refill = false;
5483                 failed_alloc = false;
5484                 BUG_ON(index != get_block_group_index(block_group));
5485                 btrfs_put_block_group(block_group);
5486         }
5487         up_read(&space_info->groups_sem);
5488
5489         if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5490                 goto search;
5491
5492         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5493                 goto search;
5494
5495         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5496          *                      for them to make caching progress.  Also
5497          *                      determine the best possible bg to cache
5498          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5499          *                      caching kthreads as we move along
5500          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5501          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5502          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5503          *                      again
5504          */
5505         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5506                 index = 0;
5507                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5508                         found_uncached_bg = false;
5509                         loop++;
5510                         if (!ideal_cache_percent)
5511                                 goto search;
5512
5513                         /*
5514                          * 1 of the following 2 things have happened so far
5515                          *
5516                          * 1) We found an ideal block group for caching that
5517                          * is mostly full and will cache quickly, so we might
5518                          * as well wait for it.
5519                          *
5520                          * 2) We searched for cached only and we didn't find
5521                          * anything, and we didn't start any caching kthreads
5522                          * either, so chances are we will loop through and
5523                          * start a couple caching kthreads, and then come back
5524                          * around and just wait for them.  This will be slower
5525                          * because we will have 2 caching kthreads reading at
5526                          * the same time when we could have just started one
5527                          * and waited for it to get far enough to give us an
5528                          * allocation, so go ahead and go to the wait caching
5529                          * loop.
5530                          */
5531                         loop = LOOP_CACHING_WAIT;
5532                         search_start = ideal_cache_offset;
5533                         ideal_cache_percent = 0;
5534                         goto ideal_cache;
5535                 } else if (loop == LOOP_FIND_IDEAL) {
5536                         /*
5537                          * Didn't find a uncached bg, wait on anything we find
5538                          * next.
5539                          */
5540                         loop = LOOP_CACHING_WAIT;
5541                         goto search;
5542                 }
5543
5544                 loop++;
5545
5546                 if (loop == LOOP_ALLOC_CHUNK) {
5547                        if (allowed_chunk_alloc) {
5548                                 ret = do_chunk_alloc(trans, root, num_bytes +
5549                                                      2 * 1024 * 1024, data,
5550                                                      CHUNK_ALLOC_LIMITED);
5551                                 allowed_chunk_alloc = 0;
5552                                 if (ret == 1)
5553                                         done_chunk_alloc = 1;
5554                         } else if (!done_chunk_alloc &&
5555                                    space_info->force_alloc ==
5556                                    CHUNK_ALLOC_NO_FORCE) {
5557                                 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5558                         }
5559
5560                        /*
5561                         * We didn't allocate a chunk, go ahead and drop the
5562                         * empty size and loop again.
5563                         */
5564                        if (!done_chunk_alloc)
5565                                loop = LOOP_NO_EMPTY_SIZE;
5566                 }
5567
5568                 if (loop == LOOP_NO_EMPTY_SIZE) {
5569                         empty_size = 0;
5570                         empty_cluster = 0;
5571                 }
5572
5573                 goto search;
5574         } else if (!ins->objectid) {
5575                 ret = -ENOSPC;
5576         } else if (ins->objectid) {
5577                 ret = 0;
5578         }
5579
5580         return ret;
5581 }
5582
5583 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5584                             int dump_block_groups)
5585 {
5586         struct btrfs_block_group_cache *cache;
5587         int index = 0;
5588
5589         spin_lock(&info->lock);
5590         printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5591                (unsigned long long)info->flags,
5592                (unsigned long long)(info->total_bytes - info->bytes_used -
5593                                     info->bytes_pinned - info->bytes_reserved -
5594                                     info->bytes_readonly),
5595                (info->full) ? "" : "not ");
5596         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5597                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5598                (unsigned long long)info->total_bytes,
5599                (unsigned long long)info->bytes_used,
5600                (unsigned long long)info->bytes_pinned,
5601                (unsigned long long)info->bytes_reserved,
5602                (unsigned long long)info->bytes_may_use,
5603                (unsigned long long)info->bytes_readonly);
5604         spin_unlock(&info->lock);
5605
5606         if (!dump_block_groups)
5607                 return;
5608
5609         down_read(&info->groups_sem);
5610 again:
5611         list_for_each_entry(cache, &info->block_groups[index], list) {
5612                 spin_lock(&cache->lock);
5613                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5614                        "%llu pinned %llu reserved\n",
5615                        (unsigned long long)cache->key.objectid,
5616                        (unsigned long long)cache->key.offset,
5617                        (unsigned long long)btrfs_block_group_used(&cache->item),
5618                        (unsigned long long)cache->pinned,
5619                        (unsigned long long)cache->reserved);
5620                 btrfs_dump_free_space(cache, bytes);
5621                 spin_unlock(&cache->lock);
5622         }
5623         if (++index < BTRFS_NR_RAID_TYPES)
5624                 goto again;
5625         up_read(&info->groups_sem);
5626 }
5627
5628 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5629                          struct btrfs_root *root,
5630                          u64 num_bytes, u64 min_alloc_size,
5631                          u64 empty_size, u64 hint_byte,
5632                          u64 search_end, struct btrfs_key *ins,
5633                          u64 data)
5634 {
5635         int ret;
5636         u64 search_start = 0;
5637
5638         data = btrfs_get_alloc_profile(root, data);
5639 again:
5640         /*
5641          * the only place that sets empty_size is btrfs_realloc_node, which
5642          * is not called recursively on allocations
5643          */
5644         if (empty_size || root->ref_cows)
5645                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5646                                      num_bytes + 2 * 1024 * 1024, data,
5647                                      CHUNK_ALLOC_NO_FORCE);
5648
5649         WARN_ON(num_bytes < root->sectorsize);
5650         ret = find_free_extent(trans, root, num_bytes, empty_size,
5651                                search_start, search_end, hint_byte,
5652                                ins, data);
5653
5654         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5655                 num_bytes = num_bytes >> 1;
5656                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5657                 num_bytes = max(num_bytes, min_alloc_size);
5658                 do_chunk_alloc(trans, root->fs_info->extent_root,
5659                                num_bytes, data, CHUNK_ALLOC_FORCE);
5660                 goto again;
5661         }
5662         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5663                 struct btrfs_space_info *sinfo;
5664
5665                 sinfo = __find_space_info(root->fs_info, data);
5666                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5667                        "wanted %llu\n", (unsigned long long)data,
5668                        (unsigned long long)num_bytes);
5669                 dump_space_info(sinfo, num_bytes, 1);
5670         }
5671
5672         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5673
5674         return ret;
5675 }
5676
5677 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5678                                         u64 start, u64 len, int pin)
5679 {
5680         struct btrfs_block_group_cache *cache;
5681         int ret = 0;
5682
5683         cache = btrfs_lookup_block_group(root->fs_info, start);
5684         if (!cache) {
5685                 printk(KERN_ERR "Unable to find block group for %llu\n",
5686                        (unsigned long long)start);
5687                 return -ENOSPC;
5688         }
5689
5690         if (btrfs_test_opt(root, DISCARD))
5691                 ret = btrfs_discard_extent(root, start, len, NULL);
5692
5693         if (pin)
5694                 pin_down_extent(root, cache, start, len, 1);
5695         else {
5696                 btrfs_add_free_space(cache, start, len);
5697                 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5698         }
5699         btrfs_put_block_group(cache);
5700
5701         trace_btrfs_reserved_extent_free(root, start, len);
5702
5703         return ret;
5704 }
5705
5706 int btrfs_free_reserved_extent(struct btrfs_root *root,
5707                                         u64 start, u64 len)
5708 {
5709         return __btrfs_free_reserved_extent(root, start, len, 0);
5710 }
5711
5712 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5713                                        u64 start, u64 len)
5714 {
5715         return __btrfs_free_reserved_extent(root, start, len, 1);
5716 }
5717
5718 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5719                                       struct btrfs_root *root,
5720                                       u64 parent, u64 root_objectid,
5721                                       u64 flags, u64 owner, u64 offset,
5722                                       struct btrfs_key *ins, int ref_mod)
5723 {
5724         int ret;
5725         struct btrfs_fs_info *fs_info = root->fs_info;
5726         struct btrfs_extent_item *extent_item;
5727         struct btrfs_extent_inline_ref *iref;
5728         struct btrfs_path *path;
5729         struct extent_buffer *leaf;
5730         int type;
5731         u32 size;
5732
5733         if (parent > 0)
5734                 type = BTRFS_SHARED_DATA_REF_KEY;
5735         else
5736                 type = BTRFS_EXTENT_DATA_REF_KEY;
5737
5738         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5739
5740         path = btrfs_alloc_path();
5741         if (!path)
5742                 return -ENOMEM;
5743
5744         path->leave_spinning = 1;
5745         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5746                                       ins, size);
5747         BUG_ON(ret);
5748
5749         leaf = path->nodes[0];
5750         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5751                                      struct btrfs_extent_item);
5752         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5753         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5754         btrfs_set_extent_flags(leaf, extent_item,
5755                                flags | BTRFS_EXTENT_FLAG_DATA);
5756
5757         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5758         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5759         if (parent > 0) {
5760                 struct btrfs_shared_data_ref *ref;
5761                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5762                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5763                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5764         } else {
5765                 struct btrfs_extent_data_ref *ref;
5766                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5767                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5768                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5769                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5770                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5771         }
5772
5773         btrfs_mark_buffer_dirty(path->nodes[0]);
5774         btrfs_free_path(path);
5775
5776         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5777         if (ret) {
5778                 printk(KERN_ERR "btrfs update block group failed for %llu "
5779                        "%llu\n", (unsigned long long)ins->objectid,
5780                        (unsigned long long)ins->offset);
5781                 BUG();
5782         }
5783         return ret;
5784 }
5785
5786 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5787                                      struct btrfs_root *root,
5788                                      u64 parent, u64 root_objectid,
5789                                      u64 flags, struct btrfs_disk_key *key,
5790                                      int level, struct btrfs_key *ins)
5791 {
5792         int ret;
5793         struct btrfs_fs_info *fs_info = root->fs_info;
5794         struct btrfs_extent_item *extent_item;
5795         struct btrfs_tree_block_info *block_info;
5796         struct btrfs_extent_inline_ref *iref;
5797         struct btrfs_path *path;
5798         struct extent_buffer *leaf;
5799         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5800
5801         path = btrfs_alloc_path();
5802         if (!path)
5803                 return -ENOMEM;
5804
5805         path->leave_spinning = 1;
5806         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5807                                       ins, size);
5808         BUG_ON(ret);
5809
5810         leaf = path->nodes[0];
5811         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5812                                      struct btrfs_extent_item);
5813         btrfs_set_extent_refs(leaf, extent_item, 1);
5814         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5815         btrfs_set_extent_flags(leaf, extent_item,
5816                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5817         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5818
5819         btrfs_set_tree_block_key(leaf, block_info, key);
5820         btrfs_set_tree_block_level(leaf, block_info, level);
5821
5822         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5823         if (parent > 0) {
5824                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5825                 btrfs_set_extent_inline_ref_type(leaf, iref,
5826                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5827                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5828         } else {
5829                 btrfs_set_extent_inline_ref_type(leaf, iref,
5830                                                  BTRFS_TREE_BLOCK_REF_KEY);
5831                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5832         }
5833
5834         btrfs_mark_buffer_dirty(leaf);
5835         btrfs_free_path(path);
5836
5837         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5838         if (ret) {
5839                 printk(KERN_ERR "btrfs update block group failed for %llu "
5840                        "%llu\n", (unsigned long long)ins->objectid,
5841                        (unsigned long long)ins->offset);
5842                 BUG();
5843         }
5844         return ret;
5845 }
5846
5847 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5848                                      struct btrfs_root *root,
5849                                      u64 root_objectid, u64 owner,
5850                                      u64 offset, struct btrfs_key *ins)
5851 {
5852         int ret;
5853
5854         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5855
5856         ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
5857                                          ins->offset, 0,
5858                                          root_objectid, owner, offset,
5859                                          BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
5860         return ret;
5861 }
5862
5863 /*
5864  * this is used by the tree logging recovery code.  It records that
5865  * an extent has been allocated and makes sure to clear the free
5866  * space cache bits as well
5867  */
5868 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5869                                    struct btrfs_root *root,
5870                                    u64 root_objectid, u64 owner, u64 offset,
5871                                    struct btrfs_key *ins)
5872 {
5873         int ret;
5874         struct btrfs_block_group_cache *block_group;
5875         struct btrfs_caching_control *caching_ctl;
5876         u64 start = ins->objectid;
5877         u64 num_bytes = ins->offset;
5878
5879         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5880         cache_block_group(block_group, trans, NULL, 0);
5881         caching_ctl = get_caching_control(block_group);
5882
5883         if (!caching_ctl) {
5884                 BUG_ON(!block_group_cache_done(block_group));
5885                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5886                 BUG_ON(ret);
5887         } else {
5888                 mutex_lock(&caching_ctl->mutex);
5889
5890                 if (start >= caching_ctl->progress) {
5891                         ret = add_excluded_extent(root, start, num_bytes);
5892                         BUG_ON(ret);
5893                 } else if (start + num_bytes <= caching_ctl->progress) {
5894                         ret = btrfs_remove_free_space(block_group,
5895                                                       start, num_bytes);
5896                         BUG_ON(ret);
5897                 } else {
5898                         num_bytes = caching_ctl->progress - start;
5899                         ret = btrfs_remove_free_space(block_group,
5900                                                       start, num_bytes);
5901                         BUG_ON(ret);
5902
5903                         start = caching_ctl->progress;
5904                         num_bytes = ins->objectid + ins->offset -
5905                                     caching_ctl->progress;
5906                         ret = add_excluded_extent(root, start, num_bytes);
5907                         BUG_ON(ret);
5908                 }
5909
5910                 mutex_unlock(&caching_ctl->mutex);
5911                 put_caching_control(caching_ctl);
5912         }
5913
5914         ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5915                                           RESERVE_ALLOC_NO_ACCOUNT);
5916         BUG_ON(ret);
5917         btrfs_put_block_group(block_group);
5918         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5919                                          0, owner, offset, ins, 1);
5920         return ret;
5921 }
5922
5923 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5924                                             struct btrfs_root *root,
5925                                             u64 bytenr, u32 blocksize,
5926                                             int level)
5927 {
5928         struct extent_buffer *buf;
5929
5930         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5931         if (!buf)
5932                 return ERR_PTR(-ENOMEM);
5933         btrfs_set_header_generation(buf, trans->transid);
5934         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5935         btrfs_tree_lock(buf);
5936         clean_tree_block(trans, root, buf);
5937
5938         btrfs_set_lock_blocking(buf);
5939         btrfs_set_buffer_uptodate(buf);
5940
5941         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5942                 /*
5943                  * we allow two log transactions at a time, use different
5944                  * EXENT bit to differentiate dirty pages.
5945                  */
5946                 if (root->log_transid % 2 == 0)
5947                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5948                                         buf->start + buf->len - 1, GFP_NOFS);
5949                 else
5950                         set_extent_new(&root->dirty_log_pages, buf->start,
5951                                         buf->start + buf->len - 1, GFP_NOFS);
5952         } else {
5953                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5954                          buf->start + buf->len - 1, GFP_NOFS);
5955         }
5956         trans->blocks_used++;
5957         /* this returns a buffer locked for blocking */
5958         return buf;
5959 }
5960
5961 static struct btrfs_block_rsv *
5962 use_block_rsv(struct btrfs_trans_handle *trans,
5963               struct btrfs_root *root, u32 blocksize)
5964 {
5965         struct btrfs_block_rsv *block_rsv;
5966         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5967         int ret;
5968
5969         block_rsv = get_block_rsv(trans, root);
5970
5971         if (block_rsv->size == 0) {
5972                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5973                 /*
5974                  * If we couldn't reserve metadata bytes try and use some from
5975                  * the global reserve.
5976                  */
5977                 if (ret && block_rsv != global_rsv) {
5978                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5979                         if (!ret)
5980                                 return global_rsv;
5981                         return ERR_PTR(ret);
5982                 } else if (ret) {
5983                         return ERR_PTR(ret);
5984                 }
5985                 return block_rsv;
5986         }
5987
5988         ret = block_rsv_use_bytes(block_rsv, blocksize);
5989         if (!ret)
5990                 return block_rsv;
5991         if (ret) {
5992                 static DEFINE_RATELIMIT_STATE(_rs,
5993                                 DEFAULT_RATELIMIT_INTERVAL,
5994                                 /*DEFAULT_RATELIMIT_BURST*/ 2);
5995                 if (__ratelimit(&_rs)) {
5996                         printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
5997                         WARN_ON(1);
5998                 }
5999                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6000                 if (!ret) {
6001                         return block_rsv;
6002                 } else if (ret && block_rsv != global_rsv) {
6003                         ret = block_rsv_use_bytes(global_rsv, blocksize);
6004                         if (!ret)
6005                                 return global_rsv;
6006                 }
6007         }
6008
6009         return ERR_PTR(-ENOSPC);
6010 }
6011
6012 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
6013 {
6014         block_rsv_add_bytes(block_rsv, blocksize, 0);
6015         block_rsv_release_bytes(block_rsv, NULL, 0);
6016 }
6017
6018 /*
6019  * finds a free extent and does all the dirty work required for allocation
6020  * returns the key for the extent through ins, and a tree buffer for
6021  * the first block of the extent through buf.
6022  *
6023  * returns the tree buffer or NULL.
6024  */
6025 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6026                                         struct btrfs_root *root, u32 blocksize,
6027                                         u64 parent, u64 root_objectid,
6028                                         struct btrfs_disk_key *key, int level,
6029                                         u64 hint, u64 empty_size, int for_cow)
6030 {
6031         struct btrfs_key ins;
6032         struct btrfs_block_rsv *block_rsv;
6033         struct extent_buffer *buf;
6034         u64 flags = 0;
6035         int ret;
6036
6037
6038         block_rsv = use_block_rsv(trans, root, blocksize);
6039         if (IS_ERR(block_rsv))
6040                 return ERR_CAST(block_rsv);
6041
6042         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6043                                    empty_size, hint, (u64)-1, &ins, 0);
6044         if (ret) {
6045                 unuse_block_rsv(block_rsv, blocksize);
6046                 return ERR_PTR(ret);
6047         }
6048
6049         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6050                                     blocksize, level);
6051         BUG_ON(IS_ERR(buf));
6052
6053         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6054                 if (parent == 0)
6055                         parent = ins.objectid;
6056                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6057         } else
6058                 BUG_ON(parent > 0);
6059
6060         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6061                 struct btrfs_delayed_extent_op *extent_op;
6062                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6063                 BUG_ON(!extent_op);
6064                 if (key)
6065                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
6066                 else
6067                         memset(&extent_op->key, 0, sizeof(extent_op->key));
6068                 extent_op->flags_to_set = flags;
6069                 extent_op->update_key = 1;
6070                 extent_op->update_flags = 1;
6071                 extent_op->is_data = 0;
6072
6073                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6074                                         ins.objectid,
6075                                         ins.offset, parent, root_objectid,
6076                                         level, BTRFS_ADD_DELAYED_EXTENT,
6077                                         extent_op, for_cow);
6078                 BUG_ON(ret);
6079         }
6080         return buf;
6081 }
6082
6083 struct walk_control {
6084         u64 refs[BTRFS_MAX_LEVEL];
6085         u64 flags[BTRFS_MAX_LEVEL];
6086         struct btrfs_key update_progress;
6087         int stage;
6088         int level;
6089         int shared_level;
6090         int update_ref;
6091         int keep_locks;
6092         int reada_slot;
6093         int reada_count;
6094         int for_reloc;
6095 };
6096
6097 #define DROP_REFERENCE  1
6098 #define UPDATE_BACKREF  2
6099
6100 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6101                                      struct btrfs_root *root,
6102                                      struct walk_control *wc,
6103                                      struct btrfs_path *path)
6104 {
6105         u64 bytenr;
6106         u64 generation;
6107         u64 refs;
6108         u64 flags;
6109         u32 nritems;
6110         u32 blocksize;
6111         struct btrfs_key key;
6112         struct extent_buffer *eb;
6113         int ret;
6114         int slot;
6115         int nread = 0;
6116
6117         if (path->slots[wc->level] < wc->reada_slot) {
6118                 wc->reada_count = wc->reada_count * 2 / 3;
6119                 wc->reada_count = max(wc->reada_count, 2);
6120         } else {
6121                 wc->reada_count = wc->reada_count * 3 / 2;
6122                 wc->reada_count = min_t(int, wc->reada_count,
6123                                         BTRFS_NODEPTRS_PER_BLOCK(root));
6124         }
6125
6126         eb = path->nodes[wc->level];
6127         nritems = btrfs_header_nritems(eb);
6128         blocksize = btrfs_level_size(root, wc->level - 1);
6129
6130         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6131                 if (nread >= wc->reada_count)
6132                         break;
6133
6134                 cond_resched();
6135                 bytenr = btrfs_node_blockptr(eb, slot);
6136                 generation = btrfs_node_ptr_generation(eb, slot);
6137
6138                 if (slot == path->slots[wc->level])
6139                         goto reada;
6140
6141                 if (wc->stage == UPDATE_BACKREF &&
6142                     generation <= root->root_key.offset)
6143                         continue;
6144
6145                 /* We don't lock the tree block, it's OK to be racy here */
6146                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6147                                                &refs, &flags);
6148                 BUG_ON(ret);
6149                 BUG_ON(refs == 0);
6150
6151                 if (wc->stage == DROP_REFERENCE) {
6152                         if (refs == 1)
6153                                 goto reada;
6154
6155                         if (wc->level == 1 &&
6156                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6157                                 continue;
6158                         if (!wc->update_ref ||
6159                             generation <= root->root_key.offset)
6160                                 continue;
6161                         btrfs_node_key_to_cpu(eb, &key, slot);
6162                         ret = btrfs_comp_cpu_keys(&key,
6163                                                   &wc->update_progress);
6164                         if (ret < 0)
6165                                 continue;
6166                 } else {
6167                         if (wc->level == 1 &&
6168                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6169                                 continue;
6170                 }
6171 reada:
6172                 ret = readahead_tree_block(root, bytenr, blocksize,
6173                                            generation);
6174                 if (ret)
6175                         break;
6176                 nread++;
6177         }
6178         wc->reada_slot = slot;
6179 }
6180
6181 /*
6182  * hepler to process tree block while walking down the tree.
6183  *
6184  * when wc->stage == UPDATE_BACKREF, this function updates
6185  * back refs for pointers in the block.
6186  *
6187  * NOTE: return value 1 means we should stop walking down.
6188  */
6189 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6190                                    struct btrfs_root *root,
6191                                    struct btrfs_path *path,
6192                                    struct walk_control *wc, int lookup_info)
6193 {
6194         int level = wc->level;
6195         struct extent_buffer *eb = path->nodes[level];
6196         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6197         int ret;
6198
6199         if (wc->stage == UPDATE_BACKREF &&
6200             btrfs_header_owner(eb) != root->root_key.objectid)
6201                 return 1;
6202
6203         /*
6204          * when reference count of tree block is 1, it won't increase
6205          * again. once full backref flag is set, we never clear it.
6206          */
6207         if (lookup_info &&
6208             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6209              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6210                 BUG_ON(!path->locks[level]);
6211                 ret = btrfs_lookup_extent_info(trans, root,
6212                                                eb->start, eb->len,
6213                                                &wc->refs[level],
6214                                                &wc->flags[level]);
6215                 BUG_ON(ret);
6216                 BUG_ON(wc->refs[level] == 0);
6217         }
6218
6219         if (wc->stage == DROP_REFERENCE) {
6220                 if (wc->refs[level] > 1)
6221                         return 1;
6222
6223                 if (path->locks[level] && !wc->keep_locks) {
6224                         btrfs_tree_unlock_rw(eb, path->locks[level]);
6225                         path->locks[level] = 0;
6226                 }
6227                 return 0;
6228         }
6229
6230         /* wc->stage == UPDATE_BACKREF */
6231         if (!(wc->flags[level] & flag)) {
6232                 BUG_ON(!path->locks[level]);
6233                 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6234                 BUG_ON(ret);
6235                 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6236                 BUG_ON(ret);
6237                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6238                                                   eb->len, flag, 0);
6239                 BUG_ON(ret);
6240                 wc->flags[level] |= flag;
6241         }
6242
6243         /*
6244          * the block is shared by multiple trees, so it's not good to
6245          * keep the tree lock
6246          */
6247         if (path->locks[level] && level > 0) {
6248                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6249                 path->locks[level] = 0;
6250         }
6251         return 0;
6252 }
6253
6254 /*
6255  * hepler to process tree block pointer.
6256  *
6257  * when wc->stage == DROP_REFERENCE, this function checks
6258  * reference count of the block pointed to. if the block
6259  * is shared and we need update back refs for the subtree
6260  * rooted at the block, this function changes wc->stage to
6261  * UPDATE_BACKREF. if the block is shared and there is no
6262  * need to update back, this function drops the reference
6263  * to the block.
6264  *
6265  * NOTE: return value 1 means we should stop walking down.
6266  */
6267 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6268                                  struct btrfs_root *root,
6269                                  struct btrfs_path *path,
6270                                  struct walk_control *wc, int *lookup_info)
6271 {
6272         u64 bytenr;
6273         u64 generation;
6274         u64 parent;
6275         u32 blocksize;
6276         struct btrfs_key key;
6277         struct extent_buffer *next;
6278         int level = wc->level;
6279         int reada = 0;
6280         int ret = 0;
6281
6282         generation = btrfs_node_ptr_generation(path->nodes[level],
6283                                                path->slots[level]);
6284         /*
6285          * if the lower level block was created before the snapshot
6286          * was created, we know there is no need to update back refs
6287          * for the subtree
6288          */
6289         if (wc->stage == UPDATE_BACKREF &&
6290             generation <= root->root_key.offset) {
6291                 *lookup_info = 1;
6292                 return 1;
6293         }
6294
6295         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6296         blocksize = btrfs_level_size(root, level - 1);
6297
6298         next = btrfs_find_tree_block(root, bytenr, blocksize);
6299         if (!next) {
6300                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6301                 if (!next)
6302                         return -ENOMEM;
6303                 reada = 1;
6304         }
6305         btrfs_tree_lock(next);
6306         btrfs_set_lock_blocking(next);
6307
6308         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6309                                        &wc->refs[level - 1],
6310                                        &wc->flags[level - 1]);
6311         BUG_ON(ret);
6312         BUG_ON(wc->refs[level - 1] == 0);
6313         *lookup_info = 0;
6314
6315         if (wc->stage == DROP_REFERENCE) {
6316                 if (wc->refs[level - 1] > 1) {
6317                         if (level == 1 &&
6318                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6319                                 goto skip;
6320
6321                         if (!wc->update_ref ||
6322                             generation <= root->root_key.offset)
6323                                 goto skip;
6324
6325                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6326                                               path->slots[level]);
6327                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6328                         if (ret < 0)
6329                                 goto skip;
6330
6331                         wc->stage = UPDATE_BACKREF;
6332                         wc->shared_level = level - 1;
6333                 }
6334         } else {
6335                 if (level == 1 &&
6336                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6337                         goto skip;
6338         }
6339
6340         if (!btrfs_buffer_uptodate(next, generation)) {
6341                 btrfs_tree_unlock(next);
6342                 free_extent_buffer(next);
6343                 next = NULL;
6344                 *lookup_info = 1;
6345         }
6346
6347         if (!next) {
6348                 if (reada && level == 1)
6349                         reada_walk_down(trans, root, wc, path);
6350                 next = read_tree_block(root, bytenr, blocksize, generation);
6351                 if (!next)
6352                         return -EIO;
6353                 btrfs_tree_lock(next);
6354                 btrfs_set_lock_blocking(next);
6355         }
6356
6357         level--;
6358         BUG_ON(level != btrfs_header_level(next));
6359         path->nodes[level] = next;
6360         path->slots[level] = 0;
6361         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6362         wc->level = level;
6363         if (wc->level == 1)
6364                 wc->reada_slot = 0;
6365         return 0;
6366 skip:
6367         wc->refs[level - 1] = 0;
6368         wc->flags[level - 1] = 0;
6369         if (wc->stage == DROP_REFERENCE) {
6370                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6371                         parent = path->nodes[level]->start;
6372                 } else {
6373                         BUG_ON(root->root_key.objectid !=
6374                                btrfs_header_owner(path->nodes[level]));
6375                         parent = 0;
6376                 }
6377
6378                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6379                                 root->root_key.objectid, level - 1, 0, 0);
6380                 BUG_ON(ret);
6381         }
6382         btrfs_tree_unlock(next);
6383         free_extent_buffer(next);
6384         *lookup_info = 1;
6385         return 1;
6386 }
6387
6388 /*
6389  * hepler to process tree block while walking up the tree.
6390  *
6391  * when wc->stage == DROP_REFERENCE, this function drops
6392  * reference count on the block.
6393  *
6394  * when wc->stage == UPDATE_BACKREF, this function changes
6395  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6396  * to UPDATE_BACKREF previously while processing the block.
6397  *
6398  * NOTE: return value 1 means we should stop walking up.
6399  */
6400 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6401                                  struct btrfs_root *root,
6402                                  struct btrfs_path *path,
6403                                  struct walk_control *wc)
6404 {
6405         int ret;
6406         int level = wc->level;
6407         struct extent_buffer *eb = path->nodes[level];
6408         u64 parent = 0;
6409
6410         if (wc->stage == UPDATE_BACKREF) {
6411                 BUG_ON(wc->shared_level < level);
6412                 if (level < wc->shared_level)
6413                         goto out;
6414
6415                 ret = find_next_key(path, level + 1, &wc->update_progress);
6416                 if (ret > 0)
6417                         wc->update_ref = 0;
6418
6419                 wc->stage = DROP_REFERENCE;
6420                 wc->shared_level = -1;
6421                 path->slots[level] = 0;
6422
6423                 /*
6424                  * check reference count again if the block isn't locked.
6425                  * we should start walking down the tree again if reference
6426                  * count is one.
6427                  */
6428                 if (!path->locks[level]) {
6429                         BUG_ON(level == 0);
6430                         btrfs_tree_lock(eb);
6431                         btrfs_set_lock_blocking(eb);
6432                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6433
6434                         ret = btrfs_lookup_extent_info(trans, root,
6435                                                        eb->start, eb->len,
6436                                                        &wc->refs[level],
6437                                                        &wc->flags[level]);
6438                         BUG_ON(ret);
6439                         BUG_ON(wc->refs[level] == 0);
6440                         if (wc->refs[level] == 1) {
6441                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6442                                 return 1;
6443                         }
6444                 }
6445         }
6446
6447         /* wc->stage == DROP_REFERENCE */
6448         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6449
6450         if (wc->refs[level] == 1) {
6451                 if (level == 0) {
6452                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6453                                 ret = btrfs_dec_ref(trans, root, eb, 1,
6454                                                     wc->for_reloc);
6455                         else
6456                                 ret = btrfs_dec_ref(trans, root, eb, 0,
6457                                                     wc->for_reloc);
6458                         BUG_ON(ret);
6459                 }
6460                 /* make block locked assertion in clean_tree_block happy */
6461                 if (!path->locks[level] &&
6462                     btrfs_header_generation(eb) == trans->transid) {
6463                         btrfs_tree_lock(eb);
6464                         btrfs_set_lock_blocking(eb);
6465                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6466                 }
6467                 clean_tree_block(trans, root, eb);
6468         }
6469
6470         if (eb == root->node) {
6471                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6472                         parent = eb->start;
6473                 else
6474                         BUG_ON(root->root_key.objectid !=
6475                                btrfs_header_owner(eb));
6476         } else {
6477                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6478                         parent = path->nodes[level + 1]->start;
6479                 else
6480                         BUG_ON(root->root_key.objectid !=
6481                                btrfs_header_owner(path->nodes[level + 1]));
6482         }
6483
6484         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0);
6485 out:
6486         wc->refs[level] = 0;
6487         wc->flags[level] = 0;
6488         return 0;
6489 }
6490
6491 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6492                                    struct btrfs_root *root,
6493                                    struct btrfs_path *path,
6494                                    struct walk_control *wc)
6495 {
6496         int level = wc->level;
6497         int lookup_info = 1;
6498         int ret;
6499
6500         while (level >= 0) {
6501                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6502                 if (ret > 0)
6503                         break;
6504
6505                 if (level == 0)
6506                         break;
6507
6508                 if (path->slots[level] >=
6509                     btrfs_header_nritems(path->nodes[level]))
6510                         break;
6511
6512                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6513                 if (ret > 0) {
6514                         path->slots[level]++;
6515                         continue;
6516                 } else if (ret < 0)
6517                         return ret;
6518                 level = wc->level;
6519         }
6520         return 0;
6521 }
6522
6523 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6524                                  struct btrfs_root *root,
6525                                  struct btrfs_path *path,
6526                                  struct walk_control *wc, int max_level)
6527 {
6528         int level = wc->level;
6529         int ret;
6530
6531         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6532         while (level < max_level && path->nodes[level]) {
6533                 wc->level = level;
6534                 if (path->slots[level] + 1 <
6535                     btrfs_header_nritems(path->nodes[level])) {
6536                         path->slots[level]++;
6537                         return 0;
6538                 } else {
6539                         ret = walk_up_proc(trans, root, path, wc);
6540                         if (ret > 0)
6541                                 return 0;
6542
6543                         if (path->locks[level]) {
6544                                 btrfs_tree_unlock_rw(path->nodes[level],
6545                                                      path->locks[level]);
6546                                 path->locks[level] = 0;
6547                         }
6548                         free_extent_buffer(path->nodes[level]);
6549                         path->nodes[level] = NULL;
6550                         level++;
6551                 }
6552         }
6553         return 1;
6554 }
6555
6556 /*
6557  * drop a subvolume tree.
6558  *
6559  * this function traverses the tree freeing any blocks that only
6560  * referenced by the tree.
6561  *
6562  * when a shared tree block is found. this function decreases its
6563  * reference count by one. if update_ref is true, this function
6564  * also make sure backrefs for the shared block and all lower level
6565  * blocks are properly updated.
6566  */
6567 void btrfs_drop_snapshot(struct btrfs_root *root,
6568                          struct btrfs_block_rsv *block_rsv, int update_ref,
6569                          int for_reloc)
6570 {
6571         struct btrfs_path *path;
6572         struct btrfs_trans_handle *trans;
6573         struct btrfs_root *tree_root = root->fs_info->tree_root;
6574         struct btrfs_root_item *root_item = &root->root_item;
6575         struct walk_control *wc;
6576         struct btrfs_key key;
6577         int err = 0;
6578         int ret;
6579         int level;
6580
6581         path = btrfs_alloc_path();
6582         if (!path) {
6583                 err = -ENOMEM;
6584                 goto out;
6585         }
6586
6587         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6588         if (!wc) {
6589                 btrfs_free_path(path);
6590                 err = -ENOMEM;
6591                 goto out;
6592         }
6593
6594         trans = btrfs_start_transaction(tree_root, 0);
6595         BUG_ON(IS_ERR(trans));
6596
6597         if (block_rsv)
6598                 trans->block_rsv = block_rsv;
6599
6600         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6601                 level = btrfs_header_level(root->node);
6602                 path->nodes[level] = btrfs_lock_root_node(root);
6603                 btrfs_set_lock_blocking(path->nodes[level]);
6604                 path->slots[level] = 0;
6605                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6606                 memset(&wc->update_progress, 0,
6607                        sizeof(wc->update_progress));
6608         } else {
6609                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6610                 memcpy(&wc->update_progress, &key,
6611                        sizeof(wc->update_progress));
6612
6613                 level = root_item->drop_level;
6614                 BUG_ON(level == 0);
6615                 path->lowest_level = level;
6616                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6617                 path->lowest_level = 0;
6618                 if (ret < 0) {
6619                         err = ret;
6620                         goto out_free;
6621                 }
6622                 WARN_ON(ret > 0);
6623
6624                 /*
6625                  * unlock our path, this is safe because only this
6626                  * function is allowed to delete this snapshot
6627                  */
6628                 btrfs_unlock_up_safe(path, 0);
6629
6630                 level = btrfs_header_level(root->node);
6631                 while (1) {
6632                         btrfs_tree_lock(path->nodes[level]);
6633                         btrfs_set_lock_blocking(path->nodes[level]);
6634
6635                         ret = btrfs_lookup_extent_info(trans, root,
6636                                                 path->nodes[level]->start,
6637                                                 path->nodes[level]->len,
6638                                                 &wc->refs[level],
6639                                                 &wc->flags[level]);
6640                         BUG_ON(ret);
6641                         BUG_ON(wc->refs[level] == 0);
6642
6643                         if (level == root_item->drop_level)
6644                                 break;
6645
6646                         btrfs_tree_unlock(path->nodes[level]);
6647                         WARN_ON(wc->refs[level] != 1);
6648                         level--;
6649                 }
6650         }
6651
6652         wc->level = level;
6653         wc->shared_level = -1;
6654         wc->stage = DROP_REFERENCE;
6655         wc->update_ref = update_ref;
6656         wc->keep_locks = 0;
6657         wc->for_reloc = for_reloc;
6658         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6659
6660         while (1) {
6661                 ret = walk_down_tree(trans, root, path, wc);
6662                 if (ret < 0) {
6663                         err = ret;
6664                         break;
6665                 }
6666
6667                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6668                 if (ret < 0) {
6669                         err = ret;
6670                         break;
6671                 }
6672
6673                 if (ret > 0) {
6674                         BUG_ON(wc->stage != DROP_REFERENCE);
6675                         break;
6676                 }
6677
6678                 if (wc->stage == DROP_REFERENCE) {
6679                         level = wc->level;
6680                         btrfs_node_key(path->nodes[level],
6681                                        &root_item->drop_progress,
6682                                        path->slots[level]);
6683                         root_item->drop_level = level;
6684                 }
6685
6686                 BUG_ON(wc->level == 0);
6687                 if (btrfs_should_end_transaction(trans, tree_root)) {
6688                         ret = btrfs_update_root(trans, tree_root,
6689                                                 &root->root_key,
6690                                                 root_item);
6691                         BUG_ON(ret);
6692
6693                         btrfs_end_transaction_throttle(trans, tree_root);
6694                         trans = btrfs_start_transaction(tree_root, 0);
6695                         BUG_ON(IS_ERR(trans));
6696                         if (block_rsv)
6697                                 trans->block_rsv = block_rsv;
6698                 }
6699         }
6700         btrfs_release_path(path);
6701         BUG_ON(err);
6702
6703         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6704         BUG_ON(ret);
6705
6706         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6707                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6708                                            NULL, NULL);
6709                 BUG_ON(ret < 0);
6710                 if (ret > 0) {
6711                         /* if we fail to delete the orphan item this time
6712                          * around, it'll get picked up the next time.
6713                          *
6714                          * The most common failure here is just -ENOENT.
6715                          */
6716                         btrfs_del_orphan_item(trans, tree_root,
6717                                               root->root_key.objectid);
6718                 }
6719         }
6720
6721         if (root->in_radix) {
6722                 btrfs_free_fs_root(tree_root->fs_info, root);
6723         } else {
6724                 free_extent_buffer(root->node);
6725                 free_extent_buffer(root->commit_root);
6726                 kfree(root);
6727         }
6728 out_free:
6729         btrfs_end_transaction_throttle(trans, tree_root);
6730         kfree(wc);
6731         btrfs_free_path(path);
6732 out:
6733         if (err)
6734                 btrfs_std_error(root->fs_info, err);
6735         return;
6736 }
6737
6738 /*
6739  * drop subtree rooted at tree block 'node'.
6740  *
6741  * NOTE: this function will unlock and release tree block 'node'
6742  * only used by relocation code
6743  */
6744 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6745                         struct btrfs_root *root,
6746                         struct extent_buffer *node,
6747                         struct extent_buffer *parent)
6748 {
6749         struct btrfs_path *path;
6750         struct walk_control *wc;
6751         int level;
6752         int parent_level;
6753         int ret = 0;
6754         int wret;
6755
6756         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6757
6758         path = btrfs_alloc_path();
6759         if (!path)
6760                 return -ENOMEM;
6761
6762         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6763         if (!wc) {
6764                 btrfs_free_path(path);
6765                 return -ENOMEM;
6766         }
6767
6768         btrfs_assert_tree_locked(parent);
6769         parent_level = btrfs_header_level(parent);
6770         extent_buffer_get(parent);
6771         path->nodes[parent_level] = parent;
6772         path->slots[parent_level] = btrfs_header_nritems(parent);
6773
6774         btrfs_assert_tree_locked(node);
6775         level = btrfs_header_level(node);
6776         path->nodes[level] = node;
6777         path->slots[level] = 0;
6778         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6779
6780         wc->refs[parent_level] = 1;
6781         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6782         wc->level = level;
6783         wc->shared_level = -1;
6784         wc->stage = DROP_REFERENCE;
6785         wc->update_ref = 0;
6786         wc->keep_locks = 1;
6787         wc->for_reloc = 1;
6788         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6789
6790         while (1) {
6791                 wret = walk_down_tree(trans, root, path, wc);
6792                 if (wret < 0) {
6793                         ret = wret;
6794                         break;
6795                 }
6796
6797                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6798                 if (wret < 0)
6799                         ret = wret;
6800                 if (wret != 0)
6801                         break;
6802         }
6803
6804         kfree(wc);
6805         btrfs_free_path(path);
6806         return ret;
6807 }
6808
6809 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6810 {
6811         u64 num_devices;
6812         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6813                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6814
6815         /*
6816          * we add in the count of missing devices because we want
6817          * to make sure that any RAID levels on a degraded FS
6818          * continue to be honored.
6819          */
6820         num_devices = root->fs_info->fs_devices->rw_devices +
6821                 root->fs_info->fs_devices->missing_devices;
6822
6823         if (num_devices == 1) {
6824                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6825                 stripped = flags & ~stripped;
6826
6827                 /* turn raid0 into single device chunks */
6828                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6829                         return stripped;
6830
6831                 /* turn mirroring into duplication */
6832                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6833                              BTRFS_BLOCK_GROUP_RAID10))
6834                         return stripped | BTRFS_BLOCK_GROUP_DUP;
6835                 return flags;
6836         } else {
6837                 /* they already had raid on here, just return */
6838                 if (flags & stripped)
6839                         return flags;
6840
6841                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6842                 stripped = flags & ~stripped;
6843
6844                 /* switch duplicated blocks with raid1 */
6845                 if (flags & BTRFS_BLOCK_GROUP_DUP)
6846                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
6847
6848                 /* turn single device chunks into raid0 */
6849                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6850         }
6851         return flags;
6852 }
6853
6854 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6855 {
6856         struct btrfs_space_info *sinfo = cache->space_info;
6857         u64 num_bytes;
6858         u64 min_allocable_bytes;
6859         int ret = -ENOSPC;
6860
6861
6862         /*
6863          * We need some metadata space and system metadata space for
6864          * allocating chunks in some corner cases until we force to set
6865          * it to be readonly.
6866          */
6867         if ((sinfo->flags &
6868              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6869             !force)
6870                 min_allocable_bytes = 1 * 1024 * 1024;
6871         else
6872                 min_allocable_bytes = 0;
6873
6874         spin_lock(&sinfo->lock);
6875         spin_lock(&cache->lock);
6876
6877         if (cache->ro) {
6878                 ret = 0;
6879                 goto out;
6880         }
6881
6882         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6883                     cache->bytes_super - btrfs_block_group_used(&cache->item);
6884
6885         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6886             sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6887             min_allocable_bytes <= sinfo->total_bytes) {
6888                 sinfo->bytes_readonly += num_bytes;
6889                 cache->ro = 1;
6890                 ret = 0;
6891         }
6892 out:
6893         spin_unlock(&cache->lock);
6894         spin_unlock(&sinfo->lock);
6895         return ret;
6896 }
6897
6898 int btrfs_set_block_group_ro(struct btrfs_root *root,
6899                              struct btrfs_block_group_cache *cache)
6900
6901 {
6902         struct btrfs_trans_handle *trans;
6903         u64 alloc_flags;
6904         int ret;
6905
6906         BUG_ON(cache->ro);
6907
6908         trans = btrfs_join_transaction(root);
6909         BUG_ON(IS_ERR(trans));
6910
6911         alloc_flags = update_block_group_flags(root, cache->flags);
6912         if (alloc_flags != cache->flags)
6913                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6914                                CHUNK_ALLOC_FORCE);
6915
6916         ret = set_block_group_ro(cache, 0);
6917         if (!ret)
6918                 goto out;
6919         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6920         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6921                              CHUNK_ALLOC_FORCE);
6922         if (ret < 0)
6923                 goto out;
6924         ret = set_block_group_ro(cache, 0);
6925 out:
6926         btrfs_end_transaction(trans, root);
6927         return ret;
6928 }
6929
6930 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6931                             struct btrfs_root *root, u64 type)
6932 {
6933         u64 alloc_flags = get_alloc_profile(root, type);
6934         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6935                               CHUNK_ALLOC_FORCE);
6936 }
6937
6938 /*
6939  * helper to account the unused space of all the readonly block group in the
6940  * list. takes mirrors into account.
6941  */
6942 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6943 {
6944         struct btrfs_block_group_cache *block_group;
6945         u64 free_bytes = 0;
6946         int factor;
6947
6948         list_for_each_entry(block_group, groups_list, list) {
6949                 spin_lock(&block_group->lock);
6950
6951                 if (!block_group->ro) {
6952                         spin_unlock(&block_group->lock);
6953                         continue;
6954                 }
6955
6956                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6957                                           BTRFS_BLOCK_GROUP_RAID10 |
6958                                           BTRFS_BLOCK_GROUP_DUP))
6959                         factor = 2;
6960                 else
6961                         factor = 1;
6962
6963                 free_bytes += (block_group->key.offset -
6964                                btrfs_block_group_used(&block_group->item)) *
6965                                factor;
6966
6967                 spin_unlock(&block_group->lock);
6968         }
6969
6970         return free_bytes;
6971 }
6972
6973 /*
6974  * helper to account the unused space of all the readonly block group in the
6975  * space_info. takes mirrors into account.
6976  */
6977 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6978 {
6979         int i;
6980         u64 free_bytes = 0;
6981
6982         spin_lock(&sinfo->lock);
6983
6984         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6985                 if (!list_empty(&sinfo->block_groups[i]))
6986                         free_bytes += __btrfs_get_ro_block_group_free_space(
6987                                                 &sinfo->block_groups[i]);
6988
6989         spin_unlock(&sinfo->lock);
6990
6991         return free_bytes;
6992 }
6993
6994 int btrfs_set_block_group_rw(struct btrfs_root *root,
6995                               struct btrfs_block_group_cache *cache)
6996 {
6997         struct btrfs_space_info *sinfo = cache->space_info;
6998         u64 num_bytes;
6999
7000         BUG_ON(!cache->ro);
7001
7002         spin_lock(&sinfo->lock);
7003         spin_lock(&cache->lock);
7004         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7005                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7006         sinfo->bytes_readonly -= num_bytes;
7007         cache->ro = 0;
7008         spin_unlock(&cache->lock);
7009         spin_unlock(&sinfo->lock);
7010         return 0;
7011 }
7012
7013 /*
7014  * checks to see if its even possible to relocate this block group.
7015  *
7016  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7017  * ok to go ahead and try.
7018  */
7019 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7020 {
7021         struct btrfs_block_group_cache *block_group;
7022         struct btrfs_space_info *space_info;
7023         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7024         struct btrfs_device *device;
7025         u64 min_free;
7026         u64 dev_min = 1;
7027         u64 dev_nr = 0;
7028         int index;
7029         int full = 0;
7030         int ret = 0;
7031
7032         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7033
7034         /* odd, couldn't find the block group, leave it alone */
7035         if (!block_group)
7036                 return -1;
7037
7038         min_free = btrfs_block_group_used(&block_group->item);
7039
7040         /* no bytes used, we're good */
7041         if (!min_free)
7042                 goto out;
7043
7044         space_info = block_group->space_info;
7045         spin_lock(&space_info->lock);
7046
7047         full = space_info->full;
7048
7049         /*
7050          * if this is the last block group we have in this space, we can't
7051          * relocate it unless we're able to allocate a new chunk below.
7052          *
7053          * Otherwise, we need to make sure we have room in the space to handle
7054          * all of the extents from this block group.  If we can, we're good
7055          */
7056         if ((space_info->total_bytes != block_group->key.offset) &&
7057             (space_info->bytes_used + space_info->bytes_reserved +
7058              space_info->bytes_pinned + space_info->bytes_readonly +
7059              min_free < space_info->total_bytes)) {
7060                 spin_unlock(&space_info->lock);
7061                 goto out;
7062         }
7063         spin_unlock(&space_info->lock);
7064
7065         /*
7066          * ok we don't have enough space, but maybe we have free space on our
7067          * devices to allocate new chunks for relocation, so loop through our
7068          * alloc devices and guess if we have enough space.  However, if we
7069          * were marked as full, then we know there aren't enough chunks, and we
7070          * can just return.
7071          */
7072         ret = -1;
7073         if (full)
7074                 goto out;
7075
7076         /*
7077          * index:
7078          *      0: raid10
7079          *      1: raid1
7080          *      2: dup
7081          *      3: raid0
7082          *      4: single
7083          */
7084         index = get_block_group_index(block_group);
7085         if (index == 0) {
7086                 dev_min = 4;
7087                 /* Divide by 2 */
7088                 min_free >>= 1;
7089         } else if (index == 1) {
7090                 dev_min = 2;
7091         } else if (index == 2) {
7092                 /* Multiply by 2 */
7093                 min_free <<= 1;
7094         } else if (index == 3) {
7095                 dev_min = fs_devices->rw_devices;
7096                 do_div(min_free, dev_min);
7097         }
7098
7099         mutex_lock(&root->fs_info->chunk_mutex);
7100         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7101                 u64 dev_offset;
7102
7103                 /*
7104                  * check to make sure we can actually find a chunk with enough
7105                  * space to fit our block group in.
7106                  */
7107                 if (device->total_bytes > device->bytes_used + min_free) {
7108                         ret = find_free_dev_extent(NULL, device, min_free,
7109                                                    &dev_offset, NULL);
7110                         if (!ret)
7111                                 dev_nr++;
7112
7113                         if (dev_nr >= dev_min)
7114                                 break;
7115
7116                         ret = -1;
7117                 }
7118         }
7119         mutex_unlock(&root->fs_info->chunk_mutex);
7120 out:
7121         btrfs_put_block_group(block_group);
7122         return ret;
7123 }
7124
7125 static int find_first_block_group(struct btrfs_root *root,
7126                 struct btrfs_path *path, struct btrfs_key *key)
7127 {
7128         int ret = 0;
7129         struct btrfs_key found_key;
7130         struct extent_buffer *leaf;
7131         int slot;
7132
7133         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7134         if (ret < 0)
7135                 goto out;
7136
7137         while (1) {
7138                 slot = path->slots[0];
7139                 leaf = path->nodes[0];
7140                 if (slot >= btrfs_header_nritems(leaf)) {
7141                         ret = btrfs_next_leaf(root, path);
7142                         if (ret == 0)
7143                                 continue;
7144                         if (ret < 0)
7145                                 goto out;
7146                         break;
7147                 }
7148                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7149
7150                 if (found_key.objectid >= key->objectid &&
7151                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7152                         ret = 0;
7153                         goto out;
7154                 }
7155                 path->slots[0]++;
7156         }
7157 out:
7158         return ret;
7159 }
7160
7161 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7162 {
7163         struct btrfs_block_group_cache *block_group;
7164         u64 last = 0;
7165
7166         while (1) {
7167                 struct inode *inode;
7168
7169                 block_group = btrfs_lookup_first_block_group(info, last);
7170                 while (block_group) {
7171                         spin_lock(&block_group->lock);
7172                         if (block_group->iref)
7173                                 break;
7174                         spin_unlock(&block_group->lock);
7175                         block_group = next_block_group(info->tree_root,
7176                                                        block_group);
7177                 }
7178                 if (!block_group) {
7179                         if (last == 0)
7180                                 break;
7181                         last = 0;
7182                         continue;
7183                 }
7184
7185                 inode = block_group->inode;
7186                 block_group->iref = 0;
7187                 block_group->inode = NULL;
7188                 spin_unlock(&block_group->lock);
7189                 iput(inode);
7190                 last = block_group->key.objectid + block_group->key.offset;
7191                 btrfs_put_block_group(block_group);
7192         }
7193 }
7194
7195 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7196 {
7197         struct btrfs_block_group_cache *block_group;
7198         struct btrfs_space_info *space_info;
7199         struct btrfs_caching_control *caching_ctl;
7200         struct rb_node *n;
7201
7202         down_write(&info->extent_commit_sem);
7203         while (!list_empty(&info->caching_block_groups)) {
7204                 caching_ctl = list_entry(info->caching_block_groups.next,
7205                                          struct btrfs_caching_control, list);
7206                 list_del(&caching_ctl->list);
7207                 put_caching_control(caching_ctl);
7208         }
7209         up_write(&info->extent_commit_sem);
7210
7211         spin_lock(&info->block_group_cache_lock);
7212         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7213                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7214                                        cache_node);
7215                 rb_erase(&block_group->cache_node,
7216                          &info->block_group_cache_tree);
7217                 spin_unlock(&info->block_group_cache_lock);
7218
7219                 down_write(&block_group->space_info->groups_sem);
7220                 list_del(&block_group->list);
7221                 up_write(&block_group->space_info->groups_sem);
7222
7223                 if (block_group->cached == BTRFS_CACHE_STARTED)
7224                         wait_block_group_cache_done(block_group);
7225
7226                 /*
7227                  * We haven't cached this block group, which means we could
7228                  * possibly have excluded extents on this block group.
7229                  */
7230                 if (block_group->cached == BTRFS_CACHE_NO)
7231                         free_excluded_extents(info->extent_root, block_group);
7232
7233                 btrfs_remove_free_space_cache(block_group);
7234                 btrfs_put_block_group(block_group);
7235
7236                 spin_lock(&info->block_group_cache_lock);
7237         }
7238         spin_unlock(&info->block_group_cache_lock);
7239
7240         /* now that all the block groups are freed, go through and
7241          * free all the space_info structs.  This is only called during
7242          * the final stages of unmount, and so we know nobody is
7243          * using them.  We call synchronize_rcu() once before we start,
7244          * just to be on the safe side.
7245          */
7246         synchronize_rcu();
7247
7248         release_global_block_rsv(info);
7249
7250         while(!list_empty(&info->space_info)) {
7251                 space_info = list_entry(info->space_info.next,
7252                                         struct btrfs_space_info,
7253                                         list);
7254                 if (space_info->bytes_pinned > 0 ||
7255                     space_info->bytes_reserved > 0 ||
7256                     space_info->bytes_may_use > 0) {
7257                         WARN_ON(1);
7258                         dump_space_info(space_info, 0, 0);
7259                 }
7260                 list_del(&space_info->list);
7261                 kfree(space_info);
7262         }
7263         return 0;
7264 }
7265
7266 static void __link_block_group(struct btrfs_space_info *space_info,
7267                                struct btrfs_block_group_cache *cache)
7268 {
7269         int index = get_block_group_index(cache);
7270
7271         down_write(&space_info->groups_sem);
7272         list_add_tail(&cache->list, &space_info->block_groups[index]);
7273         up_write(&space_info->groups_sem);
7274 }
7275
7276 int btrfs_read_block_groups(struct btrfs_root *root)
7277 {
7278         struct btrfs_path *path;
7279         int ret;
7280         struct btrfs_block_group_cache *cache;
7281         struct btrfs_fs_info *info = root->fs_info;
7282         struct btrfs_space_info *space_info;
7283         struct btrfs_key key;
7284         struct btrfs_key found_key;
7285         struct extent_buffer *leaf;
7286         int need_clear = 0;
7287         u64 cache_gen;
7288
7289         root = info->extent_root;
7290         key.objectid = 0;
7291         key.offset = 0;
7292         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7293         path = btrfs_alloc_path();
7294         if (!path)
7295                 return -ENOMEM;
7296         path->reada = 1;
7297
7298         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7299         if (btrfs_test_opt(root, SPACE_CACHE) &&
7300             btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7301                 need_clear = 1;
7302         if (btrfs_test_opt(root, CLEAR_CACHE))
7303                 need_clear = 1;
7304
7305         while (1) {
7306                 ret = find_first_block_group(root, path, &key);
7307                 if (ret > 0)
7308                         break;
7309                 if (ret != 0)
7310                         goto error;
7311                 leaf = path->nodes[0];
7312                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7313                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7314                 if (!cache) {
7315                         ret = -ENOMEM;
7316                         goto error;
7317                 }
7318                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7319                                                 GFP_NOFS);
7320                 if (!cache->free_space_ctl) {
7321                         kfree(cache);
7322                         ret = -ENOMEM;
7323                         goto error;
7324                 }
7325
7326                 atomic_set(&cache->count, 1);
7327                 spin_lock_init(&cache->lock);
7328                 cache->fs_info = info;
7329                 INIT_LIST_HEAD(&cache->list);
7330                 INIT_LIST_HEAD(&cache->cluster_list);
7331
7332                 if (need_clear)
7333                         cache->disk_cache_state = BTRFS_DC_CLEAR;
7334
7335                 read_extent_buffer(leaf, &cache->item,
7336                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7337                                    sizeof(cache->item));
7338                 memcpy(&cache->key, &found_key, sizeof(found_key));
7339
7340                 key.objectid = found_key.objectid + found_key.offset;
7341                 btrfs_release_path(path);
7342                 cache->flags = btrfs_block_group_flags(&cache->item);
7343                 cache->sectorsize = root->sectorsize;
7344
7345                 btrfs_init_free_space_ctl(cache);
7346
7347                 /*
7348                  * We need to exclude the super stripes now so that the space
7349                  * info has super bytes accounted for, otherwise we'll think
7350                  * we have more space than we actually do.
7351                  */
7352                 exclude_super_stripes(root, cache);
7353
7354                 /*
7355                  * check for two cases, either we are full, and therefore
7356                  * don't need to bother with the caching work since we won't
7357                  * find any space, or we are empty, and we can just add all
7358                  * the space in and be done with it.  This saves us _alot_ of
7359                  * time, particularly in the full case.
7360                  */
7361                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7362                         cache->last_byte_to_unpin = (u64)-1;
7363                         cache->cached = BTRFS_CACHE_FINISHED;
7364                         free_excluded_extents(root, cache);
7365                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7366                         cache->last_byte_to_unpin = (u64)-1;
7367                         cache->cached = BTRFS_CACHE_FINISHED;
7368                         add_new_free_space(cache, root->fs_info,
7369                                            found_key.objectid,
7370                                            found_key.objectid +
7371                                            found_key.offset);
7372                         free_excluded_extents(root, cache);
7373                 }
7374
7375                 ret = update_space_info(info, cache->flags, found_key.offset,
7376                                         btrfs_block_group_used(&cache->item),
7377                                         &space_info);
7378                 BUG_ON(ret);
7379                 cache->space_info = space_info;
7380                 spin_lock(&cache->space_info->lock);
7381                 cache->space_info->bytes_readonly += cache->bytes_super;
7382                 spin_unlock(&cache->space_info->lock);
7383
7384                 __link_block_group(space_info, cache);
7385
7386                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7387                 BUG_ON(ret);
7388
7389                 set_avail_alloc_bits(root->fs_info, cache->flags);
7390                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7391                         set_block_group_ro(cache, 1);
7392         }
7393
7394         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7395                 if (!(get_alloc_profile(root, space_info->flags) &
7396                       (BTRFS_BLOCK_GROUP_RAID10 |
7397                        BTRFS_BLOCK_GROUP_RAID1 |
7398                        BTRFS_BLOCK_GROUP_DUP)))
7399                         continue;
7400                 /*
7401                  * avoid allocating from un-mirrored block group if there are
7402                  * mirrored block groups.
7403                  */
7404                 list_for_each_entry(cache, &space_info->block_groups[3], list)
7405                         set_block_group_ro(cache, 1);
7406                 list_for_each_entry(cache, &space_info->block_groups[4], list)
7407                         set_block_group_ro(cache, 1);
7408         }
7409
7410         init_global_block_rsv(info);
7411         ret = 0;
7412 error:
7413         btrfs_free_path(path);
7414         return ret;
7415 }
7416
7417 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7418                            struct btrfs_root *root, u64 bytes_used,
7419                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7420                            u64 size)
7421 {
7422         int ret;
7423         struct btrfs_root *extent_root;
7424         struct btrfs_block_group_cache *cache;
7425
7426         extent_root = root->fs_info->extent_root;
7427
7428         root->fs_info->last_trans_log_full_commit = trans->transid;
7429
7430         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7431         if (!cache)
7432                 return -ENOMEM;
7433         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7434                                         GFP_NOFS);
7435         if (!cache->free_space_ctl) {
7436                 kfree(cache);
7437                 return -ENOMEM;
7438         }
7439
7440         cache->key.objectid = chunk_offset;
7441         cache->key.offset = size;
7442         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7443         cache->sectorsize = root->sectorsize;
7444         cache->fs_info = root->fs_info;
7445
7446         atomic_set(&cache->count, 1);
7447         spin_lock_init(&cache->lock);
7448         INIT_LIST_HEAD(&cache->list);
7449         INIT_LIST_HEAD(&cache->cluster_list);
7450
7451         btrfs_init_free_space_ctl(cache);
7452
7453         btrfs_set_block_group_used(&cache->item, bytes_used);
7454         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7455         cache->flags = type;
7456         btrfs_set_block_group_flags(&cache->item, type);
7457
7458         cache->last_byte_to_unpin = (u64)-1;
7459         cache->cached = BTRFS_CACHE_FINISHED;
7460         exclude_super_stripes(root, cache);
7461
7462         add_new_free_space(cache, root->fs_info, chunk_offset,
7463                            chunk_offset + size);
7464
7465         free_excluded_extents(root, cache);
7466
7467         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7468                                 &cache->space_info);
7469         BUG_ON(ret);
7470
7471         spin_lock(&cache->space_info->lock);
7472         cache->space_info->bytes_readonly += cache->bytes_super;
7473         spin_unlock(&cache->space_info->lock);
7474
7475         __link_block_group(cache->space_info, cache);
7476
7477         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7478         BUG_ON(ret);
7479
7480         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7481                                 sizeof(cache->item));
7482         BUG_ON(ret);
7483
7484         set_avail_alloc_bits(extent_root->fs_info, type);
7485
7486         return 0;
7487 }
7488
7489 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7490                              struct btrfs_root *root, u64 group_start)
7491 {
7492         struct btrfs_path *path;
7493         struct btrfs_block_group_cache *block_group;
7494         struct btrfs_free_cluster *cluster;
7495         struct btrfs_root *tree_root = root->fs_info->tree_root;
7496         struct btrfs_key key;
7497         struct inode *inode;
7498         int ret;
7499         int factor;
7500
7501         root = root->fs_info->extent_root;
7502
7503         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7504         BUG_ON(!block_group);
7505         BUG_ON(!block_group->ro);
7506
7507         /*
7508          * Free the reserved super bytes from this block group before
7509          * remove it.
7510          */
7511         free_excluded_extents(root, block_group);
7512
7513         memcpy(&key, &block_group->key, sizeof(key));
7514         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7515                                   BTRFS_BLOCK_GROUP_RAID1 |
7516                                   BTRFS_BLOCK_GROUP_RAID10))
7517                 factor = 2;
7518         else
7519                 factor = 1;
7520
7521         /* make sure this block group isn't part of an allocation cluster */
7522         cluster = &root->fs_info->data_alloc_cluster;
7523         spin_lock(&cluster->refill_lock);
7524         btrfs_return_cluster_to_free_space(block_group, cluster);
7525         spin_unlock(&cluster->refill_lock);
7526
7527         /*
7528          * make sure this block group isn't part of a metadata
7529          * allocation cluster
7530          */
7531         cluster = &root->fs_info->meta_alloc_cluster;
7532         spin_lock(&cluster->refill_lock);
7533         btrfs_return_cluster_to_free_space(block_group, cluster);
7534         spin_unlock(&cluster->refill_lock);
7535
7536         path = btrfs_alloc_path();
7537         if (!path) {
7538                 ret = -ENOMEM;
7539                 goto out;
7540         }
7541
7542         inode = lookup_free_space_inode(tree_root, block_group, path);
7543         if (!IS_ERR(inode)) {
7544                 ret = btrfs_orphan_add(trans, inode);
7545                 BUG_ON(ret);
7546                 clear_nlink(inode);
7547                 /* One for the block groups ref */
7548                 spin_lock(&block_group->lock);
7549                 if (block_group->iref) {
7550                         block_group->iref = 0;
7551                         block_group->inode = NULL;
7552                         spin_unlock(&block_group->lock);
7553                         iput(inode);
7554                 } else {
7555                         spin_unlock(&block_group->lock);
7556                 }
7557                 /* One for our lookup ref */
7558                 btrfs_add_delayed_iput(inode);
7559         }
7560
7561         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7562         key.offset = block_group->key.objectid;
7563         key.type = 0;
7564
7565         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7566         if (ret < 0)
7567                 goto out;
7568         if (ret > 0)
7569                 btrfs_release_path(path);
7570         if (ret == 0) {
7571                 ret = btrfs_del_item(trans, tree_root, path);
7572                 if (ret)
7573                         goto out;
7574                 btrfs_release_path(path);
7575         }
7576
7577         spin_lock(&root->fs_info->block_group_cache_lock);
7578         rb_erase(&block_group->cache_node,
7579                  &root->fs_info->block_group_cache_tree);
7580         spin_unlock(&root->fs_info->block_group_cache_lock);
7581
7582         down_write(&block_group->space_info->groups_sem);
7583         /*
7584          * we must use list_del_init so people can check to see if they
7585          * are still on the list after taking the semaphore
7586          */
7587         list_del_init(&block_group->list);
7588         up_write(&block_group->space_info->groups_sem);
7589
7590         if (block_group->cached == BTRFS_CACHE_STARTED)
7591                 wait_block_group_cache_done(block_group);
7592
7593         btrfs_remove_free_space_cache(block_group);
7594
7595         spin_lock(&block_group->space_info->lock);
7596         block_group->space_info->total_bytes -= block_group->key.offset;
7597         block_group->space_info->bytes_readonly -= block_group->key.offset;
7598         block_group->space_info->disk_total -= block_group->key.offset * factor;
7599         spin_unlock(&block_group->space_info->lock);
7600
7601         memcpy(&key, &block_group->key, sizeof(key));
7602
7603         btrfs_clear_space_info_full(root->fs_info);
7604
7605         btrfs_put_block_group(block_group);
7606         btrfs_put_block_group(block_group);
7607
7608         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7609         if (ret > 0)
7610                 ret = -EIO;
7611         if (ret < 0)
7612                 goto out;
7613
7614         ret = btrfs_del_item(trans, root, path);
7615 out:
7616         btrfs_free_path(path);
7617         return ret;
7618 }
7619
7620 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7621 {
7622         struct btrfs_space_info *space_info;
7623         struct btrfs_super_block *disk_super;
7624         u64 features;
7625         u64 flags;
7626         int mixed = 0;
7627         int ret;
7628
7629         disk_super = fs_info->super_copy;
7630         if (!btrfs_super_root(disk_super))
7631                 return 1;
7632
7633         features = btrfs_super_incompat_flags(disk_super);
7634         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7635                 mixed = 1;
7636
7637         flags = BTRFS_BLOCK_GROUP_SYSTEM;
7638         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7639         if (ret)
7640                 goto out;
7641
7642         if (mixed) {
7643                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7644                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7645         } else {
7646                 flags = BTRFS_BLOCK_GROUP_METADATA;
7647                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7648                 if (ret)
7649                         goto out;
7650
7651                 flags = BTRFS_BLOCK_GROUP_DATA;
7652                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7653         }
7654 out:
7655         return ret;
7656 }
7657
7658 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7659 {
7660         return unpin_extent_range(root, start, end);
7661 }
7662
7663 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7664                                u64 num_bytes, u64 *actual_bytes)
7665 {
7666         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7667 }
7668
7669 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7670 {
7671         struct btrfs_fs_info *fs_info = root->fs_info;
7672         struct btrfs_block_group_cache *cache = NULL;
7673         u64 group_trimmed;
7674         u64 start;
7675         u64 end;
7676         u64 trimmed = 0;
7677         int ret = 0;
7678
7679         cache = btrfs_lookup_block_group(fs_info, range->start);
7680
7681         while (cache) {
7682                 if (cache->key.objectid >= (range->start + range->len)) {
7683                         btrfs_put_block_group(cache);
7684                         break;
7685                 }
7686
7687                 start = max(range->start, cache->key.objectid);
7688                 end = min(range->start + range->len,
7689                                 cache->key.objectid + cache->key.offset);
7690
7691                 if (end - start >= range->minlen) {
7692                         if (!block_group_cache_done(cache)) {
7693                                 ret = cache_block_group(cache, NULL, root, 0);
7694                                 if (!ret)
7695                                         wait_block_group_cache_done(cache);
7696                         }
7697                         ret = btrfs_trim_block_group(cache,
7698                                                      &group_trimmed,
7699                                                      start,
7700                                                      end,
7701                                                      range->minlen);
7702
7703                         trimmed += group_trimmed;
7704                         if (ret) {
7705                                 btrfs_put_block_group(cache);
7706                                 break;
7707                         }
7708                 }
7709
7710                 cache = next_block_group(fs_info->tree_root, cache);
7711         }
7712
7713         range->len = trimmed;
7714         return ret;
7715 }