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 /*
  38  * control flags for do_chunk_alloc's force field
  39  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
  40  * if we really need one.
  41  *
  42  * CHUNK_ALLOC_LIMITED means to only try and allocate one
  43  * if we have very few chunks already allocated.  This is
  44  * used as part of the clustering code to help make sure
  45  * we have a good pool of storage to cluster in, without
  46  * filling the FS with empty chunks
  47  *
  48  * CHUNK_ALLOC_FORCE means it must try to allocate one
  49  *
  50  */
  51 enum {
  52         CHUNK_ALLOC_NO_FORCE = 0,
  53         CHUNK_ALLOC_LIMITED = 1,
  54         CHUNK_ALLOC_FORCE = 2,
  55 };
  56
  57 /*
  58  * Control how reservations are dealt with.
  59  *
  60  * RESERVE_FREE - freeing a reservation.
  61  * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
  62  *   ENOSPC accounting
  63  * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
  64  *   bytes_may_use as the ENOSPC accounting is done elsewhere
  65  */
  66 enum {
  67         RESERVE_FREE = 0,
  68         RESERVE_ALLOC = 1,
  69         RESERVE_ALLOC_NO_ACCOUNT = 2,
  70 };
  71
  72 static int update_block_group(struct btrfs_trans_handle *trans,
  73                               struct btrfs_root *root,
  74                               u64 bytenr, u64 num_bytes, int alloc);
  75 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  76                                 struct btrfs_root *root,
  77                                 u64 bytenr, u64 num_bytes, u64 parent,
  78                                 u64 root_objectid, u64 owner_objectid,
  79                                 u64 owner_offset, int refs_to_drop,
  80                                 struct btrfs_delayed_extent_op *extra_op);
  81 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  82                                     struct extent_buffer *leaf,
  83                                     struct btrfs_extent_item *ei);
  84 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  85                                       struct btrfs_root *root,
  86                                       u64 parent, u64 root_objectid,
  87                                       u64 flags, u64 owner, u64 offset,
  88                                       struct btrfs_key *ins, int ref_mod);
  89 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  90                                      struct btrfs_root *root,
  91                                      u64 parent, u64 root_objectid,
  92                                      u64 flags, struct btrfs_disk_key *key,
  93                                      int level, struct btrfs_key *ins);
  94 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  95                           struct btrfs_root *extent_root, u64 alloc_bytes,
  96                           u64 flags, int force);
  97 static int find_next_key(struct btrfs_path *path, int level,
  98                          struct btrfs_key *key);
  99 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
 100                             int dump_block_groups);
 101 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
 102                                        u64 num_bytes, int reserve);
 103
 104 static noinline int
 105 block_group_cache_done(struct btrfs_block_group_cache *cache)
 106 {
 107         smp_mb();
 108         return cache->cached == BTRFS_CACHE_FINISHED;
 109 }
 110
 111 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 112 {
 113         return (cache->flags & bits) == bits;
 114 }
 115
 116 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
 117 {
 118         atomic_inc(&cache->count);
 119 }
 120
 121 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 122 {
 123         if (atomic_dec_and_test(&cache->count)) {
 124                 WARN_ON(cache->pinned > 0);
 125                 WARN_ON(cache->reserved > 0);
 126                 kfree(cache->free_space_ctl);
 127                 kfree(cache);
 128         }
 129 }
 130
 131 /*
 132  * this adds the block group to the fs_info rb tree for the block group
 133  * cache
 134  */
 135 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 136                                 struct btrfs_block_group_cache *block_group)
 137 {
 138         struct rb_node **p;
 139         struct rb_node *parent = NULL;
 140         struct btrfs_block_group_cache *cache;
 141
 142         spin_lock(&info->block_group_cache_lock);
 143         p = &info->block_group_cache_tree.rb_node;
 144
 145         while (*p) {
 146                 parent = *p;
 147                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 148                                  cache_node);
 149                 if (block_group->key.objectid < cache->key.objectid) {
 150                         p = &(*p)->rb_left;
 151                 } else if (block_group->key.objectid > cache->key.objectid) {
 152                         p = &(*p)->rb_right;
 153                 } else {
 154                         spin_unlock(&info->block_group_cache_lock);
 155                         return -EEXIST;
 156                 }
 157         }
 158
 159         rb_link_node(&block_group->cache_node, parent, p);
 160         rb_insert_color(&block_group->cache_node,
 161                         &info->block_group_cache_tree);
 162         spin_unlock(&info->block_group_cache_lock);
 163
 164         return 0;
 165 }
 166
 167 /*
 168  * This will return the block group at or after bytenr if contains is 0, else
 169  * it will return the block group that contains the bytenr
 170  */
 171 static struct btrfs_block_group_cache *
 172 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 173                               int contains)
 174 {
 175         struct btrfs_block_group_cache *cache, *ret = NULL;
 176         struct rb_node *n;
 177         u64 end, start;
 178
 179         spin_lock(&info->block_group_cache_lock);
 180         n = info->block_group_cache_tree.rb_node;
 181
 182         while (n) {
 183                 cache = rb_entry(n, struct btrfs_block_group_cache,
 184                                  cache_node);
 185                 end = cache->key.objectid + cache->key.offset - 1;
 186                 start = cache->key.objectid;
 187
 188                 if (bytenr < start) {
 189                         if (!contains && (!ret || start < ret->key.objectid))
 190                                 ret = cache;
 191                         n = n->rb_left;
 192                 } else if (bytenr > start) {
 193                         if (contains && bytenr <= end) {
 194                                 ret = cache;
 195                                 break;
 196                         }
 197                         n = n->rb_right;
 198                 } else {
 199                         ret = cache;
 200                         break;
 201                 }
 202         }
 203         if (ret)
 204                 btrfs_get_block_group(ret);
 205         spin_unlock(&info->block_group_cache_lock);
 206
 207         return ret;
 208 }
 209
 210 static int add_excluded_extent(struct btrfs_root *root,
 211                                u64 start, u64 num_bytes)
 212 {
 213         u64 end = start + num_bytes - 1;
 214         set_extent_bits(&root->fs_info->freed_extents[0],
 215                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 216         set_extent_bits(&root->fs_info->freed_extents[1],
 217                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 218         return 0;
 219 }
 220
 221 static void free_excluded_extents(struct btrfs_root *root,
 222                                   struct btrfs_block_group_cache *cache)
 223 {
 224         u64 start, end;
 225
 226         start = cache->key.objectid;
 227         end = start + cache->key.offset - 1;
 228
 229         clear_extent_bits(&root->fs_info->freed_extents[0],
 230                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 231         clear_extent_bits(&root->fs_info->freed_extents[1],
 232                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 233 }
 234
 235 static int exclude_super_stripes(struct btrfs_root *root,
 236                                  struct btrfs_block_group_cache *cache)
 237 {
 238         u64 bytenr;
 239         u64 *logical;
 240         int stripe_len;
 241         int i, nr, ret;
 242
 243         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 244                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 245                 cache->bytes_super += stripe_len;
 246                 ret = add_excluded_extent(root, cache->key.objectid,
 247                                           stripe_len);
 248                 BUG_ON(ret);
 249         }
 250
 251         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 252                 bytenr = btrfs_sb_offset(i);
 253                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 254                                        cache->key.objectid, bytenr,
 255                                        0, &logical, &nr, &stripe_len);
 256                 BUG_ON(ret);
 257
 258                 while (nr--) {
 259                         cache->bytes_super += stripe_len;
 260                         ret = add_excluded_extent(root, logical[nr],
 261                                                   stripe_len);
 262                         BUG_ON(ret);
 263                 }
 264
 265                 kfree(logical);
 266         }
 267         return 0;
 268 }
 269
 270 static struct btrfs_caching_control *
 271 get_caching_control(struct btrfs_block_group_cache *cache)
 272 {
 273         struct btrfs_caching_control *ctl;
 274
 275         spin_lock(&cache->lock);
 276         if (cache->cached != BTRFS_CACHE_STARTED) {
 277                 spin_unlock(&cache->lock);
 278                 return NULL;
 279         }
 280
 281         /* We're loading it the fast way, so we don't have a caching_ctl. */
 282         if (!cache->caching_ctl) {
 283                 spin_unlock(&cache->lock);
 284                 return NULL;
 285         }
 286
 287         ctl = cache->caching_ctl;
 288         atomic_inc(&ctl->count);
 289         spin_unlock(&cache->lock);
 290         return ctl;
 291 }
 292
 293 static void put_caching_control(struct btrfs_caching_control *ctl)
 294 {
 295         if (atomic_dec_and_test(&ctl->count))
 296                 kfree(ctl);
 297 }
 298
 299 /*
 300  * this is only called by cache_block_group, since we could have freed extents
 301  * we need to check the pinned_extents for any extents that can't be used yet
 302  * since their free space will be released as soon as the transaction commits.
 303  */
 304 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 305                               struct btrfs_fs_info *info, u64 start, u64 end)
 306 {
 307         u64 extent_start, extent_end, size, total_added = 0;
 308         int ret;
 309
 310         while (start < end) {
 311                 ret = find_first_extent_bit(info->pinned_extents, start,
 312                                             &extent_start, &extent_end,
 313                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 314                 if (ret)
 315                         break;
 316
 317                 if (extent_start <= start) {
 318                         start = extent_end + 1;
 319                 } else if (extent_start > start && extent_start < end) {
 320                         size = extent_start - start;
 321                         total_added += size;
 322                         ret = btrfs_add_free_space(block_group, start,
 323                                                    size);
 324                         BUG_ON(ret);
 325                         start = extent_end + 1;
 326                 } else {
 327                         break;
 328                 }
 329         }
 330
 331         if (start < end) {
 332                 size = end - start;
 333                 total_added += size;
 334                 ret = btrfs_add_free_space(block_group, start, size);
 335                 BUG_ON(ret);
 336         }
 337
 338         return total_added;
 339 }
 340
 341 static noinline void caching_thread(struct btrfs_work *work)
 342 {
 343         struct btrfs_block_group_cache *block_group;
 344         struct btrfs_fs_info *fs_info;
 345         struct btrfs_caching_control *caching_ctl;
 346         struct btrfs_root *extent_root;
 347         struct btrfs_path *path;
 348         struct extent_buffer *leaf;
 349         struct btrfs_key key;
 350         u64 total_found = 0;
 351         u64 last = 0;
 352         u32 nritems;
 353         int ret = 0;
 354
 355         caching_ctl = container_of(work, struct btrfs_caching_control, work);
 356         block_group = caching_ctl->block_group;
 357         fs_info = block_group->fs_info;
 358         extent_root = fs_info->extent_root;
 359
 360         path = btrfs_alloc_path();
 361         if (!path)
 362                 goto out;
 363
 364         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 365
 366         /*
 367          * We don't want to deadlock with somebody trying to allocate a new
 368          * extent for the extent root while also trying to search the extent
 369          * root to add free space.  So we skip locking and search the commit
 370          * root, since its read-only
 371          */
 372         path->skip_locking = 1;
 373         path->search_commit_root = 1;
 374         path->reada = 1;
 375
 376         key.objectid = last;
 377         key.offset = 0;
 378         key.type = BTRFS_EXTENT_ITEM_KEY;
 379 again:
 380         mutex_lock(&caching_ctl->mutex);
 381         /* need to make sure the commit_root doesn't disappear */
 382         down_read(&fs_info->extent_commit_sem);
 383
 384         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 385         if (ret < 0)
 386                 goto err;
 387
 388         leaf = path->nodes[0];
 389         nritems = btrfs_header_nritems(leaf);
 390
 391         while (1) {
 392                 if (btrfs_fs_closing(fs_info) > 1) {
 393                         last = (u64)-1;
 394                         break;
 395                 }
 396
 397                 if (path->slots[0] < nritems) {
 398                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 399                 } else {
 400                         ret = find_next_key(path, 0, &key);
 401                         if (ret)
 402                                 break;
 403
 404                         if (need_resched() ||
 405                             btrfs_next_leaf(extent_root, path)) {
 406                                 caching_ctl->progress = last;
 407                                 btrfs_release_path(path);
 408                                 up_read(&fs_info->extent_commit_sem);
 409                                 mutex_unlock(&caching_ctl->mutex);
 410                                 cond_resched();
 411                                 goto again;
 412                         }
 413                         leaf = path->nodes[0];
 414                         nritems = btrfs_header_nritems(leaf);
 415                         continue;
 416                 }
 417
 418                 if (key.objectid < block_group->key.objectid) {
 419                         path->slots[0]++;
 420                         continue;
 421                 }
 422
 423                 if (key.objectid >= block_group->key.objectid +
 424                     block_group->key.offset)
 425                         break;
 426
 427                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 428                         total_found += add_new_free_space(block_group,
 429                                                           fs_info, last,
 430                                                           key.objectid);
 431                         last = key.objectid + key.offset;
 432
 433                         if (total_found > (1024 * 1024 * 2)) {
 434                                 total_found = 0;
 435                                 wake_up(&caching_ctl->wait);
 436                         }
 437                 }
 438                 path->slots[0]++;
 439         }
 440         ret = 0;
 441
 442         total_found += add_new_free_space(block_group, fs_info, last,
 443                                           block_group->key.objectid +
 444                                           block_group->key.offset);
 445         caching_ctl->progress = (u64)-1;
 446
 447         spin_lock(&block_group->lock);
 448         block_group->caching_ctl = NULL;
 449         block_group->cached = BTRFS_CACHE_FINISHED;
 450         spin_unlock(&block_group->lock);
 451
 452 err:
 453         btrfs_free_path(path);
 454         up_read(&fs_info->extent_commit_sem);
 455
 456         free_excluded_extents(extent_root, block_group);
 457
 458         mutex_unlock(&caching_ctl->mutex);
 459 out:
 460         wake_up(&caching_ctl->wait);
 461
 462         put_caching_control(caching_ctl);
 463         btrfs_put_block_group(block_group);
 464 }
 465
 466 static int cache_block_group(struct btrfs_block_group_cache *cache,
 467                              struct btrfs_trans_handle *trans,
 468                              struct btrfs_root *root,
 469                              int load_cache_only)
 470 {
 471         DEFINE_WAIT(wait);
 472         struct btrfs_fs_info *fs_info = cache->fs_info;
 473         struct btrfs_caching_control *caching_ctl;
 474         int ret = 0;
 475
 476         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 477         BUG_ON(!caching_ctl);
 478
 479         INIT_LIST_HEAD(&caching_ctl->list);
 480         mutex_init(&caching_ctl->mutex);
 481         init_waitqueue_head(&caching_ctl->wait);
 482         caching_ctl->block_group = cache;
 483         caching_ctl->progress = cache->key.objectid;
 484         atomic_set(&caching_ctl->count, 1);
 485         caching_ctl->work.func = caching_thread;
 486
 487         spin_lock(&cache->lock);
 488         /*
 489          * This should be a rare occasion, but this could happen I think in the
 490          * case where one thread starts to load the space cache info, and then
 491          * some other thread starts a transaction commit which tries to do an
 492          * allocation while the other thread is still loading the space cache
 493          * info.  The previous loop should have kept us from choosing this block
 494          * group, but if we've moved to the state where we will wait on caching
 495          * block groups we need to first check if we're doing a fast load here,
 496          * so we can wait for it to finish, otherwise we could end up allocating
 497          * from a block group who's cache gets evicted for one reason or
 498          * another.
 499          */
 500         while (cache->cached == BTRFS_CACHE_FAST) {
 501                 struct btrfs_caching_control *ctl;
 502
 503                 ctl = cache->caching_ctl;
 504                 atomic_inc(&ctl->count);
 505                 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
 506                 spin_unlock(&cache->lock);
 507
 508                 schedule();
 509
 510                 finish_wait(&ctl->wait, &wait);
 511                 put_caching_control(ctl);
 512                 spin_lock(&cache->lock);
 513         }
 514
 515         if (cache->cached != BTRFS_CACHE_NO) {
 516                 spin_unlock(&cache->lock);
 517                 kfree(caching_ctl);
 518                 return 0;
 519         }
 520         WARN_ON(cache->caching_ctl);
 521         cache->caching_ctl = caching_ctl;
 522         cache->cached = BTRFS_CACHE_FAST;
 523         spin_unlock(&cache->lock);
 524
 525         /*
 526          * We can't do the read from on-disk cache during a commit since we need
 527          * to have the normal tree locking.  Also if we are currently trying to
 528          * allocate blocks for the tree root we can't do the fast caching since
 529          * we likely hold important locks.
 530          */
 531         if (trans && (!trans->transaction->in_commit) &&
 532             (root && root != root->fs_info->tree_root) &&
 533             btrfs_test_opt(root, SPACE_CACHE)) {
 534                 ret = load_free_space_cache(fs_info, cache);
 535
 536                 spin_lock(&cache->lock);
 537                 if (ret == 1) {
 538                         cache->caching_ctl = NULL;
 539                         cache->cached = BTRFS_CACHE_FINISHED;
 540                         cache->last_byte_to_unpin = (u64)-1;
 541                 } else {
 542                         if (load_cache_only) {
 543                                 cache->caching_ctl = NULL;
 544                                 cache->cached = BTRFS_CACHE_NO;
 545                         } else {
 546                                 cache->cached = BTRFS_CACHE_STARTED;
 547                         }
 548                 }
 549                 spin_unlock(&cache->lock);
 550                 wake_up(&caching_ctl->wait);
 551                 if (ret == 1) {
 552                         put_caching_control(caching_ctl);
 553                         free_excluded_extents(fs_info->extent_root, cache);
 554                         return 0;
 555                 }
 556         } else {
 557                 /*
 558                  * We are not going to do the fast caching, set cached to the
 559                  * appropriate value and wakeup any waiters.
 560                  */
 561                 spin_lock(&cache->lock);
 562                 if (load_cache_only) {
 563                         cache->caching_ctl = NULL;
 564                         cache->cached = BTRFS_CACHE_NO;
 565                 } else {
 566                         cache->cached = BTRFS_CACHE_STARTED;
 567                 }
 568                 spin_unlock(&cache->lock);
 569                 wake_up(&caching_ctl->wait);
 570         }
 571
 572         if (load_cache_only) {
 573                 put_caching_control(caching_ctl);
 574                 return 0;
 575         }
 576
 577         down_write(&fs_info->extent_commit_sem);
 578         atomic_inc(&caching_ctl->count);
 579         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 580         up_write(&fs_info->extent_commit_sem);
 581
 582         btrfs_get_block_group(cache);
 583
 584         btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
 585
 586         return ret;
 587 }
 588
 589 /*
 590  * return the block group that starts at or after bytenr
 591  */
 592 static struct btrfs_block_group_cache *
 593 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 594 {
 595         struct btrfs_block_group_cache *cache;
 596
 597         cache = block_group_cache_tree_search(info, bytenr, 0);
 598
 599         return cache;
 600 }
 601
 602 /*
 603  * return the block group that contains the given bytenr
 604  */
 605 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 606                                                  struct btrfs_fs_info *info,
 607                                                  u64 bytenr)
 608 {
 609         struct btrfs_block_group_cache *cache;
 610
 611         cache = block_group_cache_tree_search(info, bytenr, 1);
 612
 613         return cache;
 614 }
 615
 616 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 617                                                   u64 flags)
 618 {
 619         struct list_head *head = &info->space_info;
 620         struct btrfs_space_info *found;
 621
 622         flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
 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                                 goto next;
2292                         }
2293
2294                         list_del_init(&locked_ref->cluster);
2295                         locked_ref = NULL;
2296                 }
2297
2298                 ref->in_tree = 0;
2299                 rb_erase(&ref->rb_node, &delayed_refs->root);
2300                 delayed_refs->num_entries--;
2301                 /*
2302                  * we modified num_entries, but as we're currently running
2303                  * delayed refs, skip
2304                  *     wake_up(&delayed_refs->seq_wait);
2305                  * here.
2306                  */
2307                 spin_unlock(&delayed_refs->lock);
2308
2309                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2310                                           must_insert_reserved);
2311                 BUG_ON(ret);
2312
2313                 btrfs_put_delayed_ref(ref);
2314                 kfree(extent_op);
2315                 count++;
2316 next:
2317                 do_chunk_alloc(trans, root->fs_info->extent_root,
2318                                2 * 1024 * 1024,
2319                                btrfs_get_alloc_profile(root, 0),
2320                                CHUNK_ALLOC_NO_FORCE);
2321                 cond_resched();
2322                 spin_lock(&delayed_refs->lock);
2323         }
2324         return count;
2325 }
2326
2327
2328 static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs,
2329                         unsigned long num_refs)
2330 {
2331         struct list_head *first_seq = delayed_refs->seq_head.next;
2332
2333         spin_unlock(&delayed_refs->lock);
2334         pr_debug("waiting for more refs (num %ld, first %p)\n",
2335                  num_refs, first_seq);
2336         wait_event(delayed_refs->seq_wait,
2337                    num_refs != delayed_refs->num_entries ||
2338                    delayed_refs->seq_head.next != first_seq);
2339         pr_debug("done waiting for more refs (num %ld, first %p)\n",
2340                  delayed_refs->num_entries, delayed_refs->seq_head.next);
2341         spin_lock(&delayed_refs->lock);
2342 }
2343
2344 /*
2345  * this starts processing the delayed reference count updates and
2346  * extent insertions we have queued up so far.  count can be
2347  * 0, which means to process everything in the tree at the start
2348  * of the run (but not newly added entries), or it can be some target
2349  * number you'd like to process.
2350  */
2351 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2352                            struct btrfs_root *root, unsigned long count)
2353 {
2354         struct rb_node *node;
2355         struct btrfs_delayed_ref_root *delayed_refs;
2356         struct btrfs_delayed_ref_node *ref;
2357         struct list_head cluster;
2358         int ret;
2359         u64 delayed_start;
2360         int run_all = count == (unsigned long)-1;
2361         int run_most = 0;
2362         unsigned long num_refs = 0;
2363         int consider_waiting;
2364
2365         if (root == root->fs_info->extent_root)
2366                 root = root->fs_info->tree_root;
2367
2368         do_chunk_alloc(trans, root->fs_info->extent_root,
2369                        2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2370                        CHUNK_ALLOC_NO_FORCE);
2371
2372         delayed_refs = &trans->transaction->delayed_refs;
2373         INIT_LIST_HEAD(&cluster);
2374 again:
2375         consider_waiting = 0;
2376         spin_lock(&delayed_refs->lock);
2377         if (count == 0) {
2378                 count = delayed_refs->num_entries * 2;
2379                 run_most = 1;
2380         }
2381         while (1) {
2382                 if (!(run_all || run_most) &&
2383                     delayed_refs->num_heads_ready < 64)
2384                         break;
2385
2386                 /*
2387                  * go find something we can process in the rbtree.  We start at
2388                  * the beginning of the tree, and then build a cluster
2389                  * of refs to process starting at the first one we are able to
2390                  * lock
2391                  */
2392                 delayed_start = delayed_refs->run_delayed_start;
2393                 ret = btrfs_find_ref_cluster(trans, &cluster,
2394                                              delayed_refs->run_delayed_start);
2395                 if (ret)
2396                         break;
2397
2398                 if (delayed_start >= delayed_refs->run_delayed_start) {
2399                         if (consider_waiting == 0) {
2400                                 /*
2401                                  * btrfs_find_ref_cluster looped. let's do one
2402                                  * more cycle. if we don't run any delayed ref
2403                                  * during that cycle (because we can't because
2404                                  * all of them are blocked) and if the number of
2405                                  * refs doesn't change, we avoid busy waiting.
2406                                  */
2407                                 consider_waiting = 1;
2408                                 num_refs = delayed_refs->num_entries;
2409                         } else {
2410                                 wait_for_more_refs(delayed_refs, num_refs);
2411                                 /*
2412                                  * after waiting, things have changed. we
2413                                  * dropped the lock and someone else might have
2414                                  * run some refs, built new clusters and so on.
2415                                  * therefore, we restart staleness detection.
2416                                  */
2417                                 consider_waiting = 0;
2418                         }
2419                 }
2420
2421                 ret = run_clustered_refs(trans, root, &cluster);
2422                 BUG_ON(ret < 0);
2423
2424                 count -= min_t(unsigned long, ret, count);
2425
2426                 if (count == 0)
2427                         break;
2428
2429                 if (ret || delayed_refs->run_delayed_start == 0) {
2430                         /* refs were run, let's reset staleness detection */
2431                         consider_waiting = 0;
2432                 }
2433         }
2434
2435         if (run_all) {
2436                 node = rb_first(&delayed_refs->root);
2437                 if (!node)
2438                         goto out;
2439                 count = (unsigned long)-1;
2440
2441                 while (node) {
2442                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2443                                        rb_node);
2444                         if (btrfs_delayed_ref_is_head(ref)) {
2445                                 struct btrfs_delayed_ref_head *head;
2446
2447                                 head = btrfs_delayed_node_to_head(ref);
2448                                 atomic_inc(&ref->refs);
2449
2450                                 spin_unlock(&delayed_refs->lock);
2451                                 /*
2452                                  * Mutex was contended, block until it's
2453                                  * released and try again
2454                                  */
2455                                 mutex_lock(&head->mutex);
2456                                 mutex_unlock(&head->mutex);
2457
2458                                 btrfs_put_delayed_ref(ref);
2459                                 cond_resched();
2460                                 goto again;
2461                         }
2462                         node = rb_next(node);
2463                 }
2464                 spin_unlock(&delayed_refs->lock);
2465                 schedule_timeout(1);
2466                 goto again;
2467         }
2468 out:
2469         spin_unlock(&delayed_refs->lock);
2470         return 0;
2471 }
2472
2473 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2474                                 struct btrfs_root *root,
2475                                 u64 bytenr, u64 num_bytes, u64 flags,
2476                                 int is_data)
2477 {
2478         struct btrfs_delayed_extent_op *extent_op;
2479         int ret;
2480
2481         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2482         if (!extent_op)
2483                 return -ENOMEM;
2484
2485         extent_op->flags_to_set = flags;
2486         extent_op->update_flags = 1;
2487         extent_op->update_key = 0;
2488         extent_op->is_data = is_data ? 1 : 0;
2489
2490         ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2491                                           num_bytes, extent_op);
2492         if (ret)
2493                 kfree(extent_op);
2494         return ret;
2495 }
2496
2497 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2498                                       struct btrfs_root *root,
2499                                       struct btrfs_path *path,
2500                                       u64 objectid, u64 offset, u64 bytenr)
2501 {
2502         struct btrfs_delayed_ref_head *head;
2503         struct btrfs_delayed_ref_node *ref;
2504         struct btrfs_delayed_data_ref *data_ref;
2505         struct btrfs_delayed_ref_root *delayed_refs;
2506         struct rb_node *node;
2507         int ret = 0;
2508
2509         ret = -ENOENT;
2510         delayed_refs = &trans->transaction->delayed_refs;
2511         spin_lock(&delayed_refs->lock);
2512         head = btrfs_find_delayed_ref_head(trans, bytenr);
2513         if (!head)
2514                 goto out;
2515
2516         if (!mutex_trylock(&head->mutex)) {
2517                 atomic_inc(&head->node.refs);
2518                 spin_unlock(&delayed_refs->lock);
2519
2520                 btrfs_release_path(path);
2521
2522                 /*
2523                  * Mutex was contended, block until it's released and let
2524                  * caller try again
2525                  */
2526                 mutex_lock(&head->mutex);
2527                 mutex_unlock(&head->mutex);
2528                 btrfs_put_delayed_ref(&head->node);
2529                 return -EAGAIN;
2530         }
2531
2532         node = rb_prev(&head->node.rb_node);
2533         if (!node)
2534                 goto out_unlock;
2535
2536         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2537
2538         if (ref->bytenr != bytenr)
2539                 goto out_unlock;
2540
2541         ret = 1;
2542         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2543                 goto out_unlock;
2544
2545         data_ref = btrfs_delayed_node_to_data_ref(ref);
2546
2547         node = rb_prev(node);
2548         if (node) {
2549                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2550                 if (ref->bytenr == bytenr)
2551                         goto out_unlock;
2552         }
2553
2554         if (data_ref->root != root->root_key.objectid ||
2555             data_ref->objectid != objectid || data_ref->offset != offset)
2556                 goto out_unlock;
2557
2558         ret = 0;
2559 out_unlock:
2560         mutex_unlock(&head->mutex);
2561 out:
2562         spin_unlock(&delayed_refs->lock);
2563         return ret;
2564 }
2565
2566 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2567                                         struct btrfs_root *root,
2568                                         struct btrfs_path *path,
2569                                         u64 objectid, u64 offset, u64 bytenr)
2570 {
2571         struct btrfs_root *extent_root = root->fs_info->extent_root;
2572         struct extent_buffer *leaf;
2573         struct btrfs_extent_data_ref *ref;
2574         struct btrfs_extent_inline_ref *iref;
2575         struct btrfs_extent_item *ei;
2576         struct btrfs_key key;
2577         u32 item_size;
2578         int ret;
2579
2580         key.objectid = bytenr;
2581         key.offset = (u64)-1;
2582         key.type = BTRFS_EXTENT_ITEM_KEY;
2583
2584         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2585         if (ret < 0)
2586                 goto out;
2587         BUG_ON(ret == 0);
2588
2589         ret = -ENOENT;
2590         if (path->slots[0] == 0)
2591                 goto out;
2592
2593         path->slots[0]--;
2594         leaf = path->nodes[0];
2595         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2596
2597         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2598                 goto out;
2599
2600         ret = 1;
2601         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2602 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2603         if (item_size < sizeof(*ei)) {
2604                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2605                 goto out;
2606         }
2607 #endif
2608         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2609
2610         if (item_size != sizeof(*ei) +
2611             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2612                 goto out;
2613
2614         if (btrfs_extent_generation(leaf, ei) <=
2615             btrfs_root_last_snapshot(&root->root_item))
2616                 goto out;
2617
2618         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2619         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2620             BTRFS_EXTENT_DATA_REF_KEY)
2621                 goto out;
2622
2623         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2624         if (btrfs_extent_refs(leaf, ei) !=
2625             btrfs_extent_data_ref_count(leaf, ref) ||
2626             btrfs_extent_data_ref_root(leaf, ref) !=
2627             root->root_key.objectid ||
2628             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2629             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2630                 goto out;
2631
2632         ret = 0;
2633 out:
2634         return ret;
2635 }
2636
2637 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2638                           struct btrfs_root *root,
2639                           u64 objectid, u64 offset, u64 bytenr)
2640 {
2641         struct btrfs_path *path;
2642         int ret;
2643         int ret2;
2644
2645         path = btrfs_alloc_path();
2646         if (!path)
2647                 return -ENOENT;
2648
2649         do {
2650                 ret = check_committed_ref(trans, root, path, objectid,
2651                                           offset, bytenr);
2652                 if (ret && ret != -ENOENT)
2653                         goto out;
2654
2655                 ret2 = check_delayed_ref(trans, root, path, objectid,
2656                                          offset, bytenr);
2657         } while (ret2 == -EAGAIN);
2658
2659         if (ret2 && ret2 != -ENOENT) {
2660                 ret = ret2;
2661                 goto out;
2662         }
2663
2664         if (ret != -ENOENT || ret2 != -ENOENT)
2665                 ret = 0;
2666 out:
2667         btrfs_free_path(path);
2668         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2669                 WARN_ON(ret > 0);
2670         return ret;
2671 }
2672
2673 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2674                            struct btrfs_root *root,
2675                            struct extent_buffer *buf,
2676                            int full_backref, int inc, int for_cow)
2677 {
2678         u64 bytenr;
2679         u64 num_bytes;
2680         u64 parent;
2681         u64 ref_root;
2682         u32 nritems;
2683         struct btrfs_key key;
2684         struct btrfs_file_extent_item *fi;
2685         int i;
2686         int level;
2687         int ret = 0;
2688         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2689                             u64, u64, u64, u64, u64, u64, int);
2690
2691         ref_root = btrfs_header_owner(buf);
2692         nritems = btrfs_header_nritems(buf);
2693         level = btrfs_header_level(buf);
2694
2695         if (!root->ref_cows && level == 0)
2696                 return 0;
2697
2698         if (inc)
2699                 process_func = btrfs_inc_extent_ref;
2700         else
2701                 process_func = btrfs_free_extent;
2702
2703         if (full_backref)
2704                 parent = buf->start;
2705         else
2706                 parent = 0;
2707
2708         for (i = 0; i < nritems; i++) {
2709                 if (level == 0) {
2710                         btrfs_item_key_to_cpu(buf, &key, i);
2711                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2712                                 continue;
2713                         fi = btrfs_item_ptr(buf, i,
2714                                             struct btrfs_file_extent_item);
2715                         if (btrfs_file_extent_type(buf, fi) ==
2716                             BTRFS_FILE_EXTENT_INLINE)
2717                                 continue;
2718                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2719                         if (bytenr == 0)
2720                                 continue;
2721
2722                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2723                         key.offset -= btrfs_file_extent_offset(buf, fi);
2724                         ret = process_func(trans, root, bytenr, num_bytes,
2725                                            parent, ref_root, key.objectid,
2726                                            key.offset, for_cow);
2727                         if (ret)
2728                                 goto fail;
2729                 } else {
2730                         bytenr = btrfs_node_blockptr(buf, i);
2731                         num_bytes = btrfs_level_size(root, level - 1);
2732                         ret = process_func(trans, root, bytenr, num_bytes,
2733                                            parent, ref_root, level - 1, 0,
2734                                            for_cow);
2735                         if (ret)
2736                                 goto fail;
2737                 }
2738         }
2739         return 0;
2740 fail:
2741         BUG();
2742         return ret;
2743 }
2744
2745 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2746                   struct extent_buffer *buf, int full_backref, int for_cow)
2747 {
2748         return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2749 }
2750
2751 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2752                   struct extent_buffer *buf, int full_backref, int for_cow)
2753 {
2754         return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2755 }
2756
2757 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2758                                  struct btrfs_root *root,
2759                                  struct btrfs_path *path,
2760                                  struct btrfs_block_group_cache *cache)
2761 {
2762         int ret;
2763         struct btrfs_root *extent_root = root->fs_info->extent_root;
2764         unsigned long bi;
2765         struct extent_buffer *leaf;
2766
2767         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2768         if (ret < 0)
2769                 goto fail;
2770         BUG_ON(ret);
2771
2772         leaf = path->nodes[0];
2773         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2774         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2775         btrfs_mark_buffer_dirty(leaf);
2776         btrfs_release_path(path);
2777 fail:
2778         if (ret)
2779                 return ret;
2780         return 0;
2781
2782 }
2783
2784 static struct btrfs_block_group_cache *
2785 next_block_group(struct btrfs_root *root,
2786                  struct btrfs_block_group_cache *cache)
2787 {
2788         struct rb_node *node;
2789         spin_lock(&root->fs_info->block_group_cache_lock);
2790         node = rb_next(&cache->cache_node);
2791         btrfs_put_block_group(cache);
2792         if (node) {
2793                 cache = rb_entry(node, struct btrfs_block_group_cache,
2794                                  cache_node);
2795                 btrfs_get_block_group(cache);
2796         } else
2797                 cache = NULL;
2798         spin_unlock(&root->fs_info->block_group_cache_lock);
2799         return cache;
2800 }
2801
2802 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2803                             struct btrfs_trans_handle *trans,
2804                             struct btrfs_path *path)
2805 {
2806         struct btrfs_root *root = block_group->fs_info->tree_root;
2807         struct inode *inode = NULL;
2808         u64 alloc_hint = 0;
2809         int dcs = BTRFS_DC_ERROR;
2810         int num_pages = 0;
2811         int retries = 0;
2812         int ret = 0;
2813
2814         /*
2815          * If this block group is smaller than 100 megs don't bother caching the
2816          * block group.
2817          */
2818         if (block_group->key.offset < (100 * 1024 * 1024)) {
2819                 spin_lock(&block_group->lock);
2820                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2821                 spin_unlock(&block_group->lock);
2822                 return 0;
2823         }
2824
2825 again:
2826         inode = lookup_free_space_inode(root, block_group, path);
2827         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2828                 ret = PTR_ERR(inode);
2829                 btrfs_release_path(path);
2830                 goto out;
2831         }
2832
2833         if (IS_ERR(inode)) {
2834                 BUG_ON(retries);
2835                 retries++;
2836
2837                 if (block_group->ro)
2838                         goto out_free;
2839
2840                 ret = create_free_space_inode(root, trans, block_group, path);
2841                 if (ret)
2842                         goto out_free;
2843                 goto again;
2844         }
2845
2846         /* We've already setup this transaction, go ahead and exit */
2847         if (block_group->cache_generation == trans->transid &&
2848             i_size_read(inode)) {
2849                 dcs = BTRFS_DC_SETUP;
2850                 goto out_put;
2851         }
2852
2853         /*
2854          * We want to set the generation to 0, that way if anything goes wrong
2855          * from here on out we know not to trust this cache when we load up next
2856          * time.
2857          */
2858         BTRFS_I(inode)->generation = 0;
2859         ret = btrfs_update_inode(trans, root, inode);
2860         WARN_ON(ret);
2861
2862         if (i_size_read(inode) > 0) {
2863                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2864                                                       inode);
2865                 if (ret)
2866                         goto out_put;
2867         }
2868
2869         spin_lock(&block_group->lock);
2870         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2871                 /* We're not cached, don't bother trying to write stuff out */
2872                 dcs = BTRFS_DC_WRITTEN;
2873                 spin_unlock(&block_group->lock);
2874                 goto out_put;
2875         }
2876         spin_unlock(&block_group->lock);
2877
2878         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2879         if (!num_pages)
2880                 num_pages = 1;
2881
2882         /*
2883          * Just to make absolutely sure we have enough space, we're going to
2884          * preallocate 12 pages worth of space for each block group.  In
2885          * practice we ought to use at most 8, but we need extra space so we can
2886          * add our header and have a terminator between the extents and the
2887          * bitmaps.
2888          */
2889         num_pages *= 16;
2890         num_pages *= PAGE_CACHE_SIZE;
2891
2892         ret = btrfs_check_data_free_space(inode, num_pages);
2893         if (ret)
2894                 goto out_put;
2895
2896         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2897                                               num_pages, num_pages,
2898                                               &alloc_hint);
2899         if (!ret)
2900                 dcs = BTRFS_DC_SETUP;
2901         btrfs_free_reserved_data_space(inode, num_pages);
2902
2903 out_put:
2904         iput(inode);
2905 out_free:
2906         btrfs_release_path(path);
2907 out:
2908         spin_lock(&block_group->lock);
2909         if (!ret && dcs == BTRFS_DC_SETUP)
2910                 block_group->cache_generation = trans->transid;
2911         block_group->disk_cache_state = dcs;
2912         spin_unlock(&block_group->lock);
2913
2914         return ret;
2915 }
2916
2917 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2918                                    struct btrfs_root *root)
2919 {
2920         struct btrfs_block_group_cache *cache;
2921         int err = 0;
2922         struct btrfs_path *path;
2923         u64 last = 0;
2924
2925         path = btrfs_alloc_path();
2926         if (!path)
2927                 return -ENOMEM;
2928
2929 again:
2930         while (1) {
2931                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2932                 while (cache) {
2933                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2934                                 break;
2935                         cache = next_block_group(root, cache);
2936                 }
2937                 if (!cache) {
2938                         if (last == 0)
2939                                 break;
2940                         last = 0;
2941                         continue;
2942                 }
2943                 err = cache_save_setup(cache, trans, path);
2944                 last = cache->key.objectid + cache->key.offset;
2945                 btrfs_put_block_group(cache);
2946         }
2947
2948         while (1) {
2949                 if (last == 0) {
2950                         err = btrfs_run_delayed_refs(trans, root,
2951                                                      (unsigned long)-1);
2952                         BUG_ON(err);
2953                 }
2954
2955                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2956                 while (cache) {
2957                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2958                                 btrfs_put_block_group(cache);
2959                                 goto again;
2960                         }
2961
2962                         if (cache->dirty)
2963                                 break;
2964                         cache = next_block_group(root, cache);
2965                 }
2966                 if (!cache) {
2967                         if (last == 0)
2968                                 break;
2969                         last = 0;
2970                         continue;
2971                 }
2972
2973                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2974                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2975                 cache->dirty = 0;
2976                 last = cache->key.objectid + cache->key.offset;
2977
2978                 err = write_one_cache_group(trans, root, path, cache);
2979                 BUG_ON(err);
2980                 btrfs_put_block_group(cache);
2981         }
2982
2983         while (1) {
2984                 /*
2985                  * I don't think this is needed since we're just marking our
2986                  * preallocated extent as written, but just in case it can't
2987                  * hurt.
2988                  */
2989                 if (last == 0) {
2990                         err = btrfs_run_delayed_refs(trans, root,
2991                                                      (unsigned long)-1);
2992                         BUG_ON(err);
2993                 }
2994
2995                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2996                 while (cache) {
2997                         /*
2998                          * Really this shouldn't happen, but it could if we
2999                          * couldn't write the entire preallocated extent and
3000                          * splitting the extent resulted in a new block.
3001                          */
3002                         if (cache->dirty) {
3003                                 btrfs_put_block_group(cache);
3004                                 goto again;
3005                         }
3006                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3007                                 break;
3008                         cache = next_block_group(root, cache);
3009                 }
3010                 if (!cache) {
3011                         if (last == 0)
3012                                 break;
3013                         last = 0;
3014                         continue;
3015                 }
3016
3017                 btrfs_write_out_cache(root, trans, cache, path);
3018
3019                 /*
3020                  * If we didn't have an error then the cache state is still
3021                  * NEED_WRITE, so we can set it to WRITTEN.
3022                  */
3023                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3024                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
3025                 last = cache->key.objectid + cache->key.offset;
3026                 btrfs_put_block_group(cache);
3027         }
3028
3029         btrfs_free_path(path);
3030         return 0;
3031 }
3032
3033 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3034 {
3035         struct btrfs_block_group_cache *block_group;
3036         int readonly = 0;
3037
3038         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3039         if (!block_group || block_group->ro)
3040                 readonly = 1;
3041         if (block_group)
3042                 btrfs_put_block_group(block_group);
3043         return readonly;
3044 }
3045
3046 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3047                              u64 total_bytes, u64 bytes_used,
3048                              struct btrfs_space_info **space_info)
3049 {
3050         struct btrfs_space_info *found;
3051         int i;
3052         int factor;
3053
3054         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3055                      BTRFS_BLOCK_GROUP_RAID10))
3056                 factor = 2;
3057         else
3058                 factor = 1;
3059
3060         found = __find_space_info(info, flags);
3061         if (found) {
3062                 spin_lock(&found->lock);
3063                 found->total_bytes += total_bytes;
3064                 found->disk_total += total_bytes * factor;
3065                 found->bytes_used += bytes_used;
3066                 found->disk_used += bytes_used * factor;
3067                 found->full = 0;
3068                 spin_unlock(&found->lock);
3069                 *space_info = found;
3070                 return 0;
3071         }
3072         found = kzalloc(sizeof(*found), GFP_NOFS);
3073         if (!found)
3074                 return -ENOMEM;
3075
3076         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3077                 INIT_LIST_HEAD(&found->block_groups[i]);
3078         init_rwsem(&found->groups_sem);
3079         spin_lock_init(&found->lock);
3080         found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3081         found->total_bytes = total_bytes;
3082         found->disk_total = total_bytes * factor;
3083         found->bytes_used = bytes_used;
3084         found->disk_used = bytes_used * factor;
3085         found->bytes_pinned = 0;
3086         found->bytes_reserved = 0;
3087         found->bytes_readonly = 0;
3088         found->bytes_may_use = 0;
3089         found->full = 0;
3090         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3091         found->chunk_alloc = 0;
3092         found->flush = 0;
3093         init_waitqueue_head(&found->wait);
3094         *space_info = found;
3095         list_add_rcu(&found->list, &info->space_info);
3096         return 0;
3097 }
3098
3099 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3100 {
3101         u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
3102
3103         /* chunk -> extended profile */
3104         if (extra_flags == 0)
3105                 extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3106
3107         if (flags & BTRFS_BLOCK_GROUP_DATA)
3108                 fs_info->avail_data_alloc_bits |= extra_flags;
3109         if (flags & BTRFS_BLOCK_GROUP_METADATA)
3110                 fs_info->avail_metadata_alloc_bits |= extra_flags;
3111         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3112                 fs_info->avail_system_alloc_bits |= extra_flags;
3113 }
3114
3115 /*
3116  * @flags: available profiles in extended format (see ctree.h)
3117  *
3118  * Returns reduced profile in chunk format.  If profile changing is in
3119  * progress (either running or paused) picks the target profile (if it's
3120  * already available), otherwise falls back to plain reducing.
3121  */
3122 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3123 {
3124         /*
3125          * we add in the count of missing devices because we want
3126          * to make sure that any RAID levels on a degraded FS
3127          * continue to be honored.
3128          */
3129         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3130                 root->fs_info->fs_devices->missing_devices;
3131
3132         /* pick restriper's target profile if it's available */
3133         spin_lock(&root->fs_info->balance_lock);
3134         if (root->fs_info->balance_ctl) {
3135                 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3136                 u64 tgt = 0;
3137
3138                 if ((flags & BTRFS_BLOCK_GROUP_DATA) &&
3139                     (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3140                     (flags & bctl->data.target)) {
3141                         tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3142                 } else if ((flags & BTRFS_BLOCK_GROUP_SYSTEM) &&
3143                            (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3144                            (flags & bctl->sys.target)) {
3145                         tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3146                 } else if ((flags & BTRFS_BLOCK_GROUP_METADATA) &&
3147                            (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3148                            (flags & bctl->meta.target)) {
3149                         tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3150                 }
3151
3152                 if (tgt) {
3153                         spin_unlock(&root->fs_info->balance_lock);
3154                         flags = tgt;
3155                         goto out;
3156                 }
3157         }
3158         spin_unlock(&root->fs_info->balance_lock);
3159
3160         if (num_devices == 1)
3161                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3162         if (num_devices < 4)
3163                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3164
3165         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3166             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3167                       BTRFS_BLOCK_GROUP_RAID10))) {
3168                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3169         }
3170
3171         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3172             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3173                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3174         }
3175
3176         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3177             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3178              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3179              (flags & BTRFS_BLOCK_GROUP_DUP))) {
3180                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3181         }
3182
3183 out:
3184         /* extended -> chunk profile */
3185         flags &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3186         return flags;
3187 }
3188
3189 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3190 {
3191         if (flags & BTRFS_BLOCK_GROUP_DATA)
3192                 flags |= root->fs_info->avail_data_alloc_bits;
3193         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3194                 flags |= root->fs_info->avail_system_alloc_bits;
3195         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3196                 flags |= root->fs_info->avail_metadata_alloc_bits;
3197
3198         return btrfs_reduce_alloc_profile(root, flags);
3199 }
3200
3201 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3202 {
3203         u64 flags;
3204
3205         if (data)
3206                 flags = BTRFS_BLOCK_GROUP_DATA;
3207         else if (root == root->fs_info->chunk_root)
3208                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3209         else
3210                 flags = BTRFS_BLOCK_GROUP_METADATA;
3211
3212         return get_alloc_profile(root, flags);
3213 }
3214
3215 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3216 {
3217         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3218                                                        BTRFS_BLOCK_GROUP_DATA);
3219 }
3220
3221 /*
3222  * This will check the space that the inode allocates from to make sure we have
3223  * enough space for bytes.
3224  */
3225 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3226 {
3227         struct btrfs_space_info *data_sinfo;
3228         struct btrfs_root *root = BTRFS_I(inode)->root;
3229         u64 used;
3230         int ret = 0, committed = 0, alloc_chunk = 1;
3231
3232         /* make sure bytes are sectorsize aligned */
3233         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3234
3235         if (root == root->fs_info->tree_root ||
3236             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3237                 alloc_chunk = 0;
3238                 committed = 1;
3239         }
3240
3241         data_sinfo = BTRFS_I(inode)->space_info;
3242         if (!data_sinfo)
3243                 goto alloc;
3244
3245 again:
3246         /* make sure we have enough space to handle the data first */
3247         spin_lock(&data_sinfo->lock);
3248         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3249                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3250                 data_sinfo->bytes_may_use;
3251
3252         if (used + bytes > data_sinfo->total_bytes) {
3253                 struct btrfs_trans_handle *trans;
3254
3255                 /*
3256                  * if we don't have enough free bytes in this space then we need
3257                  * to alloc a new chunk.
3258                  */
3259                 if (!data_sinfo->full && alloc_chunk) {
3260                         u64 alloc_target;
3261
3262                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3263                         spin_unlock(&data_sinfo->lock);
3264 alloc:
3265                         alloc_target = btrfs_get_alloc_profile(root, 1);
3266                         trans = btrfs_join_transaction(root);
3267                         if (IS_ERR(trans))
3268                                 return PTR_ERR(trans);
3269
3270                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3271                                              bytes + 2 * 1024 * 1024,
3272                                              alloc_target,
3273                                              CHUNK_ALLOC_NO_FORCE);
3274                         btrfs_end_transaction(trans, root);
3275                         if (ret < 0) {
3276                                 if (ret != -ENOSPC)
3277                                         return ret;
3278                                 else
3279                                         goto commit_trans;
3280                         }
3281
3282                         if (!data_sinfo) {
3283                                 btrfs_set_inode_space_info(root, inode);
3284                                 data_sinfo = BTRFS_I(inode)->space_info;
3285                         }
3286                         goto again;
3287                 }
3288
3289                 /*
3290                  * If we have less pinned bytes than we want to allocate then
3291                  * don't bother committing the transaction, it won't help us.
3292                  */
3293                 if (data_sinfo->bytes_pinned < bytes)
3294                         committed = 1;
3295                 spin_unlock(&data_sinfo->lock);
3296
3297                 /* commit the current transaction and try again */
3298 commit_trans:
3299                 if (!committed &&
3300                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3301                         committed = 1;
3302                         trans = btrfs_join_transaction(root);
3303                         if (IS_ERR(trans))
3304                                 return PTR_ERR(trans);
3305                         ret = btrfs_commit_transaction(trans, root);
3306                         if (ret)
3307                                 return ret;
3308                         goto again;
3309                 }
3310
3311                 return -ENOSPC;
3312         }
3313         data_sinfo->bytes_may_use += bytes;
3314         trace_btrfs_space_reservation(root->fs_info, "space_info",
3315                                       (u64)data_sinfo, bytes, 1);
3316         spin_unlock(&data_sinfo->lock);
3317
3318         return 0;
3319 }
3320
3321 /*
3322  * Called if we need to clear a data reservation for this inode.
3323  */
3324 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3325 {
3326         struct btrfs_root *root = BTRFS_I(inode)->root;
3327         struct btrfs_space_info *data_sinfo;
3328
3329         /* make sure bytes are sectorsize aligned */
3330         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3331
3332         data_sinfo = BTRFS_I(inode)->space_info;
3333         spin_lock(&data_sinfo->lock);
3334         data_sinfo->bytes_may_use -= bytes;
3335         trace_btrfs_space_reservation(root->fs_info, "space_info",
3336                                       (u64)data_sinfo, bytes, 0);
3337         spin_unlock(&data_sinfo->lock);
3338 }
3339
3340 static void force_metadata_allocation(struct btrfs_fs_info *info)
3341 {
3342         struct list_head *head = &info->space_info;
3343         struct btrfs_space_info *found;
3344
3345         rcu_read_lock();
3346         list_for_each_entry_rcu(found, head, list) {
3347                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3348                         found->force_alloc = CHUNK_ALLOC_FORCE;
3349         }
3350         rcu_read_unlock();
3351 }
3352
3353 static int should_alloc_chunk(struct btrfs_root *root,
3354                               struct btrfs_space_info *sinfo, u64 alloc_bytes,
3355                               int force)
3356 {
3357         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3358         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3359         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3360         u64 thresh;
3361
3362         if (force == CHUNK_ALLOC_FORCE)
3363                 return 1;
3364
3365         /*
3366          * We need to take into account the global rsv because for all intents
3367          * and purposes it's used space.  Don't worry about locking the
3368          * global_rsv, it doesn't change except when the transaction commits.
3369          */
3370         num_allocated += global_rsv->size;
3371
3372         /*
3373          * in limited mode, we want to have some free space up to
3374          * about 1% of the FS size.
3375          */
3376         if (force == CHUNK_ALLOC_LIMITED) {
3377                 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3378                 thresh = max_t(u64, 64 * 1024 * 1024,
3379                                div_factor_fine(thresh, 1));
3380
3381                 if (num_bytes - num_allocated < thresh)
3382                         return 1;
3383         }
3384         thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3385
3386         /* 256MB or 2% of the FS */
3387         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3388         /* system chunks need a much small threshold */
3389         if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM)
3390                 thresh = 32 * 1024 * 1024;
3391
3392         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3393                 return 0;
3394         return 1;
3395 }
3396
3397 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3398                           struct btrfs_root *extent_root, u64 alloc_bytes,
3399                           u64 flags, int force)
3400 {
3401         struct btrfs_space_info *space_info;
3402         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3403         int wait_for_alloc = 0;
3404         int ret = 0;
3405
3406         BUG_ON(!profile_is_valid(flags, 0));
3407
3408         space_info = __find_space_info(extent_root->fs_info, flags);
3409         if (!space_info) {
3410                 ret = update_space_info(extent_root->fs_info, flags,
3411                                         0, 0, &space_info);
3412                 BUG_ON(ret);
3413         }
3414         BUG_ON(!space_info);
3415
3416 again:
3417         spin_lock(&space_info->lock);
3418         if (force < space_info->force_alloc)
3419                 force = space_info->force_alloc;
3420         if (space_info->full) {
3421                 spin_unlock(&space_info->lock);
3422                 return 0;
3423         }
3424
3425         if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3426                 spin_unlock(&space_info->lock);
3427                 return 0;
3428         } else if (space_info->chunk_alloc) {
3429                 wait_for_alloc = 1;
3430         } else {
3431                 space_info->chunk_alloc = 1;
3432         }
3433
3434         spin_unlock(&space_info->lock);
3435
3436         mutex_lock(&fs_info->chunk_mutex);
3437
3438         /*
3439          * The chunk_mutex is held throughout the entirety of a chunk
3440          * allocation, so once we've acquired the chunk_mutex we know that the
3441          * other guy is done and we need to recheck and see if we should
3442          * allocate.
3443          */
3444         if (wait_for_alloc) {
3445                 mutex_unlock(&fs_info->chunk_mutex);
3446                 wait_for_alloc = 0;
3447                 goto again;
3448         }
3449
3450         /*
3451          * If we have mixed data/metadata chunks we want to make sure we keep
3452          * allocating mixed chunks instead of individual chunks.
3453          */
3454         if (btrfs_mixed_space_info(space_info))
3455                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3456
3457         /*
3458          * if we're doing a data chunk, go ahead and make sure that
3459          * we keep a reasonable number of metadata chunks allocated in the
3460          * FS as well.
3461          */
3462         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3463                 fs_info->data_chunk_allocations++;
3464                 if (!(fs_info->data_chunk_allocations %
3465                       fs_info->metadata_ratio))
3466                         force_metadata_allocation(fs_info);
3467         }
3468
3469         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3470         if (ret < 0 && ret != -ENOSPC)
3471                 goto out;
3472
3473         spin_lock(&space_info->lock);
3474         if (ret)
3475                 space_info->full = 1;
3476         else
3477                 ret = 1;
3478
3479         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3480         space_info->chunk_alloc = 0;
3481         spin_unlock(&space_info->lock);
3482 out:
3483         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3484         return ret;
3485 }
3486
3487 /*
3488  * shrink metadata reservation for delalloc
3489  */
3490 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3491                            bool wait_ordered)
3492 {
3493         struct btrfs_block_rsv *block_rsv;
3494         struct btrfs_space_info *space_info;
3495         struct btrfs_trans_handle *trans;
3496         u64 reserved;
3497         u64 max_reclaim;
3498         u64 reclaimed = 0;
3499         long time_left;
3500         unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3501         int loops = 0;
3502         unsigned long progress;
3503
3504         trans = (struct btrfs_trans_handle *)current->journal_info;
3505         block_rsv = &root->fs_info->delalloc_block_rsv;
3506         space_info = block_rsv->space_info;
3507
3508         smp_mb();
3509         reserved = space_info->bytes_may_use;
3510         progress = space_info->reservation_progress;
3511
3512         if (reserved == 0)
3513                 return 0;
3514
3515         smp_mb();
3516         if (root->fs_info->delalloc_bytes == 0) {
3517                 if (trans)
3518                         return 0;
3519                 btrfs_wait_ordered_extents(root, 0, 0);
3520                 return 0;
3521         }
3522
3523         max_reclaim = min(reserved, to_reclaim);
3524         nr_pages = max_t(unsigned long, nr_pages,
3525                          max_reclaim >> PAGE_CACHE_SHIFT);
3526         while (loops < 1024) {
3527                 /* have the flusher threads jump in and do some IO */
3528                 smp_mb();
3529                 nr_pages = min_t(unsigned long, nr_pages,
3530                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3531                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3532                                                 WB_REASON_FS_FREE_SPACE);
3533
3534                 spin_lock(&space_info->lock);
3535                 if (reserved > space_info->bytes_may_use)
3536                         reclaimed += reserved - space_info->bytes_may_use;
3537                 reserved = space_info->bytes_may_use;
3538                 spin_unlock(&space_info->lock);
3539
3540                 loops++;
3541
3542                 if (reserved == 0 || reclaimed >= max_reclaim)
3543                         break;
3544
3545                 if (trans && trans->transaction->blocked)
3546                         return -EAGAIN;
3547
3548                 if (wait_ordered && !trans) {
3549                         btrfs_wait_ordered_extents(root, 0, 0);
3550                 } else {
3551                         time_left = schedule_timeout_interruptible(1);
3552
3553                         /* We were interrupted, exit */
3554                         if (time_left)
3555                                 break;
3556                 }
3557
3558                 /* we've kicked the IO a few times, if anything has been freed,
3559                  * exit.  There is no sense in looping here for a long time
3560                  * when we really need to commit the transaction, or there are
3561                  * just too many writers without enough free space
3562                  */
3563
3564                 if (loops > 3) {
3565                         smp_mb();
3566                         if (progress != space_info->reservation_progress)
3567                                 break;
3568                 }
3569
3570         }
3571
3572         return reclaimed >= to_reclaim;
3573 }
3574
3575 /**
3576  * maybe_commit_transaction - possibly commit the transaction if its ok to
3577  * @root - the root we're allocating for
3578  * @bytes - the number of bytes we want to reserve
3579  * @force - force the commit
3580  *
3581  * This will check to make sure that committing the transaction will actually
3582  * get us somewhere and then commit the transaction if it does.  Otherwise it
3583  * will return -ENOSPC.
3584  */
3585 static int may_commit_transaction(struct btrfs_root *root,
3586                                   struct btrfs_space_info *space_info,
3587                                   u64 bytes, int force)
3588 {
3589         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3590         struct btrfs_trans_handle *trans;
3591
3592         trans = (struct btrfs_trans_handle *)current->journal_info;
3593         if (trans)
3594                 return -EAGAIN;
3595
3596         if (force)
3597                 goto commit;
3598
3599         /* See if there is enough pinned space to make this reservation */
3600         spin_lock(&space_info->lock);
3601         if (space_info->bytes_pinned >= bytes) {
3602                 spin_unlock(&space_info->lock);
3603                 goto commit;
3604         }
3605         spin_unlock(&space_info->lock);
3606
3607         /*
3608          * See if there is some space in the delayed insertion reservation for
3609          * this reservation.
3610          */
3611         if (space_info != delayed_rsv->space_info)
3612                 return -ENOSPC;
3613
3614         spin_lock(&space_info->lock);
3615         spin_lock(&delayed_rsv->lock);
3616         if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3617                 spin_unlock(&delayed_rsv->lock);
3618                 spin_unlock(&space_info->lock);
3619                 return -ENOSPC;
3620         }
3621         spin_unlock(&delayed_rsv->lock);
3622         spin_unlock(&space_info->lock);
3623
3624 commit:
3625         trans = btrfs_join_transaction(root);
3626         if (IS_ERR(trans))
3627                 return -ENOSPC;
3628
3629         return btrfs_commit_transaction(trans, root);
3630 }
3631
3632 /**
3633  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3634  * @root - the root we're allocating for
3635  * @block_rsv - the block_rsv we're allocating for
3636  * @orig_bytes - the number of bytes we want
3637  * @flush - wether or not we can flush to make our reservation
3638  *
3639  * This will reserve orgi_bytes number of bytes from the space info associated
3640  * with the block_rsv.  If there is not enough space it will make an attempt to
3641  * flush out space to make room.  It will do this by flushing delalloc if
3642  * possible or committing the transaction.  If flush is 0 then no attempts to
3643  * regain reservations will be made and this will fail if there is not enough
3644  * space already.
3645  */
3646 static int reserve_metadata_bytes(struct btrfs_root *root,
3647                                   struct btrfs_block_rsv *block_rsv,
3648                                   u64 orig_bytes, int flush)
3649 {
3650         struct btrfs_space_info *space_info = block_rsv->space_info;
3651         u64 used;
3652         u64 num_bytes = orig_bytes;
3653         int retries = 0;
3654         int ret = 0;
3655         bool committed = false;
3656         bool flushing = false;
3657         bool wait_ordered = false;
3658
3659 again:
3660         ret = 0;
3661         spin_lock(&space_info->lock);
3662         /*
3663          * We only want to wait if somebody other than us is flushing and we are
3664          * actually alloed to flush.
3665          */
3666         while (flush && !flushing && space_info->flush) {
3667                 spin_unlock(&space_info->lock);
3668                 /*
3669                  * If we have a trans handle we can't wait because the flusher
3670                  * may have to commit the transaction, which would mean we would
3671                  * deadlock since we are waiting for the flusher to finish, but
3672                  * hold the current transaction open.
3673                  */
3674                 if (current->journal_info)
3675                         return -EAGAIN;
3676                 ret = wait_event_interruptible(space_info->wait,
3677                                                !space_info->flush);
3678                 /* Must have been interrupted, return */
3679                 if (ret)
3680                         return -EINTR;
3681
3682                 spin_lock(&space_info->lock);
3683         }
3684
3685         ret = -ENOSPC;
3686         used = space_info->bytes_used + space_info->bytes_reserved +
3687                 space_info->bytes_pinned + space_info->bytes_readonly +
3688                 space_info->bytes_may_use;
3689
3690         /*
3691          * The idea here is that we've not already over-reserved the block group
3692          * then we can go ahead and save our reservation first and then start
3693          * flushing if we need to.  Otherwise if we've already overcommitted
3694          * lets start flushing stuff first and then come back and try to make
3695          * our reservation.
3696          */
3697         if (used <= space_info->total_bytes) {
3698                 if (used + orig_bytes <= space_info->total_bytes) {
3699                         space_info->bytes_may_use += orig_bytes;
3700                         trace_btrfs_space_reservation(root->fs_info,
3701                                                       "space_info",
3702                                                       (u64)space_info,
3703                                                       orig_bytes, 1);
3704                         ret = 0;
3705                 } else {
3706                         /*
3707                          * Ok set num_bytes to orig_bytes since we aren't
3708                          * overocmmitted, this way we only try and reclaim what
3709                          * we need.
3710                          */
3711                         num_bytes = orig_bytes;
3712                 }
3713         } else {
3714                 /*
3715                  * Ok we're over committed, set num_bytes to the overcommitted
3716                  * amount plus the amount of bytes that we need for this
3717                  * reservation.
3718                  */
3719                 wait_ordered = true;
3720                 num_bytes = used - space_info->total_bytes +
3721                         (orig_bytes * (retries + 1));
3722         }
3723
3724         if (ret) {
3725                 u64 profile = btrfs_get_alloc_profile(root, 0);
3726                 u64 avail;
3727
3728                 /*
3729                  * If we have a lot of space that's pinned, don't bother doing
3730                  * the overcommit dance yet and just commit the transaction.
3731                  */
3732                 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3733                 do_div(avail, 10);
3734                 if (space_info->bytes_pinned >= avail && flush && !committed) {
3735                         space_info->flush = 1;
3736                         flushing = true;
3737                         spin_unlock(&space_info->lock);
3738                         ret = may_commit_transaction(root, space_info,
3739                                                      orig_bytes, 1);
3740                         if (ret)
3741                                 goto out;
3742                         committed = true;
3743                         goto again;
3744                 }
3745
3746                 spin_lock(&root->fs_info->free_chunk_lock);
3747                 avail = root->fs_info->free_chunk_space;
3748
3749                 /*
3750                  * If we have dup, raid1 or raid10 then only half of the free
3751                  * space is actually useable.
3752                  */
3753                 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3754                                BTRFS_BLOCK_GROUP_RAID1 |
3755                                BTRFS_BLOCK_GROUP_RAID10))
3756                         avail >>= 1;
3757
3758                 /*
3759                  * If we aren't flushing don't let us overcommit too much, say
3760                  * 1/8th of the space.  If we can flush, let it overcommit up to
3761                  * 1/2 of the space.
3762                  */
3763                 if (flush)
3764                         avail >>= 3;
3765                 else
3766                         avail >>= 1;
3767                  spin_unlock(&root->fs_info->free_chunk_lock);
3768
3769                 if (used + num_bytes < space_info->total_bytes + avail) {
3770                         space_info->bytes_may_use += orig_bytes;
3771                         trace_btrfs_space_reservation(root->fs_info,
3772                                                       "space_info",
3773                                                       (u64)space_info,
3774                                                       orig_bytes, 1);
3775                         ret = 0;
3776                 } else {
3777                         wait_ordered = true;
3778                 }
3779         }
3780
3781         /*
3782          * Couldn't make our reservation, save our place so while we're trying
3783          * to reclaim space we can actually use it instead of somebody else
3784          * stealing it from us.
3785          */
3786         if (ret && flush) {
3787                 flushing = true;
3788                 space_info->flush = 1;
3789         }
3790
3791         spin_unlock(&space_info->lock);
3792
3793         if (!ret || !flush)
3794                 goto out;
3795
3796         /*
3797          * We do synchronous shrinking since we don't actually unreserve
3798          * metadata until after the IO is completed.
3799          */
3800         ret = shrink_delalloc(root, num_bytes, wait_ordered);
3801         if (ret < 0)
3802                 goto out;
3803
3804         ret = 0;
3805
3806         /*
3807          * So if we were overcommitted it's possible that somebody else flushed
3808          * out enough space and we simply didn't have enough space to reclaim,
3809          * so go back around and try again.
3810          */
3811         if (retries < 2) {
3812                 wait_ordered = true;
3813                 retries++;
3814                 goto again;
3815         }
3816
3817         ret = -ENOSPC;
3818         if (committed)
3819                 goto out;
3820
3821         ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3822         if (!ret) {
3823                 committed = true;
3824                 goto again;
3825         }
3826
3827 out:
3828         if (flushing) {
3829                 spin_lock(&space_info->lock);
3830                 space_info->flush = 0;
3831                 wake_up_all(&space_info->wait);
3832                 spin_unlock(&space_info->lock);
3833         }
3834         return ret;
3835 }
3836
3837 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3838                                              struct btrfs_root *root)
3839 {
3840         struct btrfs_block_rsv *block_rsv = NULL;
3841
3842         if (root->ref_cows || root == root->fs_info->csum_root)
3843                 block_rsv = trans->block_rsv;
3844
3845         if (!block_rsv)
3846                 block_rsv = root->block_rsv;
3847
3848         if (!block_rsv)
3849                 block_rsv = &root->fs_info->empty_block_rsv;
3850
3851         return block_rsv;
3852 }
3853
3854 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3855                                u64 num_bytes)
3856 {
3857         int ret = -ENOSPC;
3858         spin_lock(&block_rsv->lock);
3859         if (block_rsv->reserved >= num_bytes) {
3860                 block_rsv->reserved -= num_bytes;
3861                 if (block_rsv->reserved < block_rsv->size)
3862                         block_rsv->full = 0;
3863                 ret = 0;
3864         }
3865         spin_unlock(&block_rsv->lock);
3866         return ret;
3867 }
3868
3869 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3870                                 u64 num_bytes, int update_size)
3871 {
3872         spin_lock(&block_rsv->lock);
3873         block_rsv->reserved += num_bytes;
3874         if (update_size)
3875                 block_rsv->size += num_bytes;
3876         else if (block_rsv->reserved >= block_rsv->size)
3877                 block_rsv->full = 1;
3878         spin_unlock(&block_rsv->lock);
3879 }
3880
3881 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
3882                                     struct btrfs_block_rsv *block_rsv,
3883                                     struct btrfs_block_rsv *dest, u64 num_bytes)
3884 {
3885         struct btrfs_space_info *space_info = block_rsv->space_info;
3886
3887         spin_lock(&block_rsv->lock);
3888         if (num_bytes == (u64)-1)
3889                 num_bytes = block_rsv->size;
3890         block_rsv->size -= num_bytes;
3891         if (block_rsv->reserved >= block_rsv->size) {
3892                 num_bytes = block_rsv->reserved - block_rsv->size;
3893                 block_rsv->reserved = block_rsv->size;
3894                 block_rsv->full = 1;
3895         } else {
3896                 num_bytes = 0;
3897         }
3898         spin_unlock(&block_rsv->lock);
3899
3900         if (num_bytes > 0) {
3901                 if (dest) {
3902                         spin_lock(&dest->lock);
3903                         if (!dest->full) {
3904                                 u64 bytes_to_add;
3905
3906                                 bytes_to_add = dest->size - dest->reserved;
3907                                 bytes_to_add = min(num_bytes, bytes_to_add);
3908                                 dest->reserved += bytes_to_add;
3909                                 if (dest->reserved >= dest->size)
3910                                         dest->full = 1;
3911                                 num_bytes -= bytes_to_add;
3912                         }
3913                         spin_unlock(&dest->lock);
3914                 }
3915                 if (num_bytes) {
3916                         spin_lock(&space_info->lock);
3917                         space_info->bytes_may_use -= num_bytes;
3918                         trace_btrfs_space_reservation(fs_info, "space_info",
3919                                                       (u64)space_info,
3920                                                       num_bytes, 0);
3921                         space_info->reservation_progress++;
3922                         spin_unlock(&space_info->lock);
3923                 }
3924         }
3925 }
3926
3927 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3928                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3929 {
3930         int ret;
3931
3932         ret = block_rsv_use_bytes(src, num_bytes);
3933         if (ret)
3934                 return ret;
3935
3936         block_rsv_add_bytes(dst, num_bytes, 1);
3937         return 0;
3938 }
3939
3940 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3941 {
3942         memset(rsv, 0, sizeof(*rsv));
3943         spin_lock_init(&rsv->lock);
3944 }
3945
3946 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3947 {
3948         struct btrfs_block_rsv *block_rsv;
3949         struct btrfs_fs_info *fs_info = root->fs_info;
3950
3951         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3952         if (!block_rsv)
3953                 return NULL;
3954
3955         btrfs_init_block_rsv(block_rsv);
3956         block_rsv->space_info = __find_space_info(fs_info,
3957                                                   BTRFS_BLOCK_GROUP_METADATA);
3958         return block_rsv;
3959 }
3960
3961 void btrfs_free_block_rsv(struct btrfs_root *root,
3962                           struct btrfs_block_rsv *rsv)
3963 {
3964         btrfs_block_rsv_release(root, rsv, (u64)-1);
3965         kfree(rsv);
3966 }
3967
3968 static inline int __block_rsv_add(struct btrfs_root *root,
3969                                   struct btrfs_block_rsv *block_rsv,
3970                                   u64 num_bytes, int flush)
3971 {
3972         int ret;
3973
3974         if (num_bytes == 0)
3975                 return 0;
3976
3977         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3978         if (!ret) {
3979                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3980                 return 0;
3981         }
3982
3983         return ret;
3984 }
3985
3986 int btrfs_block_rsv_add(struct btrfs_root *root,
3987                         struct btrfs_block_rsv *block_rsv,
3988                         u64 num_bytes)
3989 {
3990         return __block_rsv_add(root, block_rsv, num_bytes, 1);
3991 }
3992
3993 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3994                                 struct btrfs_block_rsv *block_rsv,
3995                                 u64 num_bytes)
3996 {
3997         return __block_rsv_add(root, block_rsv, num_bytes, 0);
3998 }
3999
4000 int btrfs_block_rsv_check(struct btrfs_root *root,
4001                           struct btrfs_block_rsv *block_rsv, int min_factor)
4002 {
4003         u64 num_bytes = 0;
4004         int ret = -ENOSPC;
4005
4006         if (!block_rsv)
4007                 return 0;
4008
4009         spin_lock(&block_rsv->lock);
4010         num_bytes = div_factor(block_rsv->size, min_factor);
4011         if (block_rsv->reserved >= num_bytes)
4012                 ret = 0;
4013         spin_unlock(&block_rsv->lock);
4014
4015         return ret;
4016 }
4017
4018 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4019                                            struct btrfs_block_rsv *block_rsv,
4020                                            u64 min_reserved, int flush)
4021 {
4022         u64 num_bytes = 0;
4023         int ret = -ENOSPC;
4024
4025         if (!block_rsv)
4026                 return 0;
4027
4028         spin_lock(&block_rsv->lock);
4029         num_bytes = min_reserved;
4030         if (block_rsv->reserved >= num_bytes)
4031                 ret = 0;
4032         else
4033                 num_bytes -= block_rsv->reserved;
4034         spin_unlock(&block_rsv->lock);
4035
4036         if (!ret)
4037                 return 0;
4038
4039         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4040         if (!ret) {
4041                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4042                 return 0;
4043         }
4044
4045         return ret;
4046 }
4047
4048 int btrfs_block_rsv_refill(struct btrfs_root *root,
4049                            struct btrfs_block_rsv *block_rsv,
4050                            u64 min_reserved)
4051 {
4052         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4053 }
4054
4055 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4056                                    struct btrfs_block_rsv *block_rsv,
4057                                    u64 min_reserved)
4058 {
4059         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4060 }
4061
4062 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4063                             struct btrfs_block_rsv *dst_rsv,
4064                             u64 num_bytes)
4065 {
4066         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4067 }
4068
4069 void btrfs_block_rsv_release(struct btrfs_root *root,
4070                              struct btrfs_block_rsv *block_rsv,
4071                              u64 num_bytes)
4072 {
4073         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4074         if (global_rsv->full || global_rsv == block_rsv ||
4075             block_rsv->space_info != global_rsv->space_info)
4076                 global_rsv = NULL;
4077         block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4078                                 num_bytes);
4079 }
4080
4081 /*
4082  * helper to calculate size of global block reservation.
4083  * the desired value is sum of space used by extent tree,
4084  * checksum tree and root tree
4085  */
4086 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4087 {
4088         struct btrfs_space_info *sinfo;
4089         u64 num_bytes;
4090         u64 meta_used;
4091         u64 data_used;
4092         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4093
4094         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4095         spin_lock(&sinfo->lock);
4096         data_used = sinfo->bytes_used;
4097         spin_unlock(&sinfo->lock);
4098
4099         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4100         spin_lock(&sinfo->lock);
4101         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4102                 data_used = 0;
4103         meta_used = sinfo->bytes_used;
4104         spin_unlock(&sinfo->lock);
4105
4106         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4107                     csum_size * 2;
4108         num_bytes += div64_u64(data_used + meta_used, 50);
4109
4110         if (num_bytes * 3 > meta_used)
4111                 num_bytes = div64_u64(meta_used, 3);
4112
4113         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4114 }
4115
4116 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4117 {
4118         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4119         struct btrfs_space_info *sinfo = block_rsv->space_info;
4120         u64 num_bytes;
4121
4122         num_bytes = calc_global_metadata_size(fs_info);
4123
4124         spin_lock(&block_rsv->lock);
4125         spin_lock(&sinfo->lock);
4126
4127         block_rsv->size = num_bytes;
4128
4129         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4130                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4131                     sinfo->bytes_may_use;
4132
4133         if (sinfo->total_bytes > num_bytes) {
4134                 num_bytes = sinfo->total_bytes - num_bytes;
4135                 block_rsv->reserved += num_bytes;
4136                 sinfo->bytes_may_use += num_bytes;
4137                 trace_btrfs_space_reservation(fs_info, "space_info",
4138                                               (u64)sinfo, num_bytes, 1);
4139         }
4140
4141         if (block_rsv->reserved >= block_rsv->size) {
4142                 num_bytes = block_rsv->reserved - block_rsv->size;
4143                 sinfo->bytes_may_use -= num_bytes;
4144                 trace_btrfs_space_reservation(fs_info, "space_info",
4145                                               (u64)sinfo, num_bytes, 0);
4146                 sinfo->reservation_progress++;
4147                 block_rsv->reserved = block_rsv->size;
4148                 block_rsv->full = 1;
4149         }
4150
4151         spin_unlock(&sinfo->lock);
4152         spin_unlock(&block_rsv->lock);
4153 }
4154
4155 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4156 {
4157         struct btrfs_space_info *space_info;
4158
4159         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4160         fs_info->chunk_block_rsv.space_info = space_info;
4161
4162         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4163         fs_info->global_block_rsv.space_info = space_info;
4164         fs_info->delalloc_block_rsv.space_info = space_info;
4165         fs_info->trans_block_rsv.space_info = space_info;
4166         fs_info->empty_block_rsv.space_info = space_info;
4167         fs_info->delayed_block_rsv.space_info = space_info;
4168
4169         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4170         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4171         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4172         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4173         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4174
4175         update_global_block_rsv(fs_info);
4176 }
4177
4178 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4179 {
4180         block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4181                                 (u64)-1);
4182         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4183         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4184         WARN_ON(fs_info->trans_block_rsv.size > 0);
4185         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4186         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4187         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4188         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4189         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4190 }
4191
4192 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4193                                   struct btrfs_root *root)
4194 {
4195         if (!trans->bytes_reserved)
4196                 return;
4197
4198         trace_btrfs_space_reservation(root->fs_info, "transaction", (u64)trans,
4199                                       trans->bytes_reserved, 0);
4200         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4201         trans->bytes_reserved = 0;
4202 }
4203
4204 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4205                                   struct inode *inode)
4206 {
4207         struct btrfs_root *root = BTRFS_I(inode)->root;
4208         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4209         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4210
4211         /*
4212          * We need to hold space in order to delete our orphan item once we've
4213          * added it, so this takes the reservation so we can release it later
4214          * when we are truly done with the orphan item.
4215          */
4216         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4217         trace_btrfs_space_reservation(root->fs_info, "orphan",
4218                                       btrfs_ino(inode), num_bytes, 1);
4219         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4220 }
4221
4222 void btrfs_orphan_release_metadata(struct inode *inode)
4223 {
4224         struct btrfs_root *root = BTRFS_I(inode)->root;
4225         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4226         trace_btrfs_space_reservation(root->fs_info, "orphan",
4227                                       btrfs_ino(inode), num_bytes, 0);
4228         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4229 }
4230
4231 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4232                                 struct btrfs_pending_snapshot *pending)
4233 {
4234         struct btrfs_root *root = pending->root;
4235         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4236         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4237         /*
4238          * two for root back/forward refs, two for directory entries
4239          * and one for root of the snapshot.
4240          */
4241         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4242         dst_rsv->space_info = src_rsv->space_info;
4243         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4244 }
4245
4246 /**
4247  * drop_outstanding_extent - drop an outstanding extent
4248  * @inode: the inode we're dropping the extent for
4249  *
4250  * This is called when we are freeing up an outstanding extent, either called
4251  * after an error or after an extent is written.  This will return the number of
4252  * reserved extents that need to be freed.  This must be called with
4253  * BTRFS_I(inode)->lock held.
4254  */
4255 static unsigned drop_outstanding_extent(struct inode *inode)
4256 {
4257         unsigned drop_inode_space = 0;
4258         unsigned dropped_extents = 0;
4259
4260         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4261         BTRFS_I(inode)->outstanding_extents--;
4262
4263         if (BTRFS_I(inode)->outstanding_extents == 0 &&
4264             BTRFS_I(inode)->delalloc_meta_reserved) {
4265                 drop_inode_space = 1;
4266                 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4267         }
4268
4269         /*
4270          * If we have more or the same amount of outsanding extents than we have
4271          * reserved then we need to leave the reserved extents count alone.
4272          */
4273         if (BTRFS_I(inode)->outstanding_extents >=
4274             BTRFS_I(inode)->reserved_extents)
4275                 return drop_inode_space;
4276
4277         dropped_extents = BTRFS_I(inode)->reserved_extents -
4278                 BTRFS_I(inode)->outstanding_extents;
4279         BTRFS_I(inode)->reserved_extents -= dropped_extents;
4280         return dropped_extents + drop_inode_space;
4281 }
4282
4283 /**
4284  * calc_csum_metadata_size - return the amount of metada space that must be
4285  *      reserved/free'd for the given bytes.
4286  * @inode: the inode we're manipulating
4287  * @num_bytes: the number of bytes in question
4288  * @reserve: 1 if we are reserving space, 0 if we are freeing space
4289  *
4290  * This adjusts the number of csum_bytes in the inode and then returns the
4291  * correct amount of metadata that must either be reserved or freed.  We
4292  * calculate how many checksums we can fit into one leaf and then divide the
4293  * number of bytes that will need to be checksumed by this value to figure out
4294  * how many checksums will be required.  If we are adding bytes then the number
4295  * may go up and we will return the number of additional bytes that must be
4296  * reserved.  If it is going down we will return the number of bytes that must
4297  * be freed.
4298  *
4299  * This must be called with BTRFS_I(inode)->lock held.
4300  */
4301 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4302                                    int reserve)
4303 {
4304         struct btrfs_root *root = BTRFS_I(inode)->root;
4305         u64 csum_size;
4306         int num_csums_per_leaf;
4307         int num_csums;
4308         int old_csums;
4309
4310         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4311             BTRFS_I(inode)->csum_bytes == 0)
4312                 return 0;
4313
4314         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4315         if (reserve)
4316                 BTRFS_I(inode)->csum_bytes += num_bytes;
4317         else
4318                 BTRFS_I(inode)->csum_bytes -= num_bytes;
4319         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4320         num_csums_per_leaf = (int)div64_u64(csum_size,
4321                                             sizeof(struct btrfs_csum_item) +
4322                                             sizeof(struct btrfs_disk_key));
4323         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4324         num_csums = num_csums + num_csums_per_leaf - 1;
4325         num_csums = num_csums / num_csums_per_leaf;
4326
4327         old_csums = old_csums + num_csums_per_leaf - 1;
4328         old_csums = old_csums / num_csums_per_leaf;
4329
4330         /* No change, no need to reserve more */
4331         if (old_csums == num_csums)
4332                 return 0;
4333
4334         if (reserve)
4335                 return btrfs_calc_trans_metadata_size(root,
4336                                                       num_csums - old_csums);
4337
4338         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4339 }
4340
4341 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4342 {
4343         struct btrfs_root *root = BTRFS_I(inode)->root;
4344         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4345         u64 to_reserve = 0;
4346         u64 csum_bytes;
4347         unsigned nr_extents = 0;
4348         int extra_reserve = 0;
4349         int flush = 1;
4350         int ret;
4351
4352         /* Need to be holding the i_mutex here if we aren't free space cache */
4353         if (btrfs_is_free_space_inode(root, inode))
4354                 flush = 0;
4355
4356         if (flush && btrfs_transaction_in_commit(root->fs_info))
4357                 schedule_timeout(1);
4358
4359         mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4360         num_bytes = ALIGN(num_bytes, root->sectorsize);
4361
4362         spin_lock(&BTRFS_I(inode)->lock);
4363         BTRFS_I(inode)->outstanding_extents++;
4364
4365         if (BTRFS_I(inode)->outstanding_extents >
4366             BTRFS_I(inode)->reserved_extents)
4367                 nr_extents = BTRFS_I(inode)->outstanding_extents -
4368                         BTRFS_I(inode)->reserved_extents;
4369
4370         /*
4371          * Add an item to reserve for updating the inode when we complete the
4372          * delalloc io.
4373          */
4374         if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4375                 nr_extents++;
4376                 extra_reserve = 1;
4377         }
4378
4379         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4380         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4381         csum_bytes = BTRFS_I(inode)->csum_bytes;
4382         spin_unlock(&BTRFS_I(inode)->lock);
4383
4384         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4385         if (ret) {
4386                 u64 to_free = 0;
4387                 unsigned dropped;
4388
4389                 spin_lock(&BTRFS_I(inode)->lock);
4390                 dropped = drop_outstanding_extent(inode);
4391                 /*
4392                  * If the inodes csum_bytes is the same as the original
4393                  * csum_bytes then we know we haven't raced with any free()ers
4394                  * so we can just reduce our inodes csum bytes and carry on.
4395                  * Otherwise we have to do the normal free thing to account for
4396                  * the case that the free side didn't free up its reserve
4397                  * because of this outstanding reservation.
4398                  */
4399                 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4400                         calc_csum_metadata_size(inode, num_bytes, 0);
4401                 else
4402                         to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4403                 spin_unlock(&BTRFS_I(inode)->lock);
4404                 if (dropped)
4405                         to_free += btrfs_calc_trans_metadata_size(root, dropped);
4406
4407                 if (to_free) {
4408                         btrfs_block_rsv_release(root, block_rsv, to_free);
4409                         trace_btrfs_space_reservation(root->fs_info,
4410                                                       "delalloc",
4411                                                       btrfs_ino(inode),
4412                                                       to_free, 0);
4413                 }
4414                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4415                 return ret;
4416         }
4417
4418         spin_lock(&BTRFS_I(inode)->lock);
4419         if (extra_reserve) {
4420                 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4421                 nr_extents--;
4422         }
4423         BTRFS_I(inode)->reserved_extents += nr_extents;
4424         spin_unlock(&BTRFS_I(inode)->lock);
4425         mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4426
4427         if (to_reserve)
4428                 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4429                                               btrfs_ino(inode), to_reserve, 1);
4430         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4431
4432         return 0;
4433 }
4434
4435 /**
4436  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4437  * @inode: the inode to release the reservation for
4438  * @num_bytes: the number of bytes we're releasing
4439  *
4440  * This will release the metadata reservation for an inode.  This can be called
4441  * once we complete IO for a given set of bytes to release their metadata
4442  * reservations.
4443  */
4444 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4445 {
4446         struct btrfs_root *root = BTRFS_I(inode)->root;
4447         u64 to_free = 0;
4448         unsigned dropped;
4449
4450         num_bytes = ALIGN(num_bytes, root->sectorsize);
4451         spin_lock(&BTRFS_I(inode)->lock);
4452         dropped = drop_outstanding_extent(inode);
4453
4454         to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4455         spin_unlock(&BTRFS_I(inode)->lock);
4456         if (dropped > 0)
4457                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4458
4459         trace_btrfs_space_reservation(root->fs_info, "delalloc",
4460                                       btrfs_ino(inode), to_free, 0);
4461         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4462                                 to_free);
4463 }
4464
4465 /**
4466  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4467  * @inode: inode we're writing to
4468  * @num_bytes: the number of bytes we want to allocate
4469  *
4470  * This will do the following things
4471  *
4472  * o reserve space in the data space info for num_bytes
4473  * o reserve space in the metadata space info based on number of outstanding
4474  *   extents and how much csums will be needed
4475  * o add to the inodes ->delalloc_bytes
4476  * o add it to the fs_info's delalloc inodes list.
4477  *
4478  * This will return 0 for success and -ENOSPC if there is no space left.
4479  */
4480 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4481 {
4482         int ret;
4483
4484         ret = btrfs_check_data_free_space(inode, num_bytes);
4485         if (ret)
4486                 return ret;
4487
4488         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4489         if (ret) {
4490                 btrfs_free_reserved_data_space(inode, num_bytes);
4491                 return ret;
4492         }
4493
4494         return 0;
4495 }
4496
4497 /**
4498  * btrfs_delalloc_release_space - release data and metadata space for delalloc
4499  * @inode: inode we're releasing space for
4500  * @num_bytes: the number of bytes we want to free up
4501  *
4502  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
4503  * called in the case that we don't need the metadata AND data reservations
4504  * anymore.  So if there is an error or we insert an inline extent.
4505  *
4506  * This function will release the metadata space that was not used and will
4507  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4508  * list if there are no delalloc bytes left.
4509  */
4510 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4511 {
4512         btrfs_delalloc_release_metadata(inode, num_bytes);
4513         btrfs_free_reserved_data_space(inode, num_bytes);
4514 }
4515
4516 static int update_block_group(struct btrfs_trans_handle *trans,
4517                               struct btrfs_root *root,
4518                               u64 bytenr, u64 num_bytes, int alloc)
4519 {
4520         struct btrfs_block_group_cache *cache = NULL;
4521         struct btrfs_fs_info *info = root->fs_info;
4522         u64 total = num_bytes;
4523         u64 old_val;
4524         u64 byte_in_group;
4525         int factor;
4526
4527         /* block accounting for super block */
4528         spin_lock(&info->delalloc_lock);
4529         old_val = btrfs_super_bytes_used(info->super_copy);
4530         if (alloc)
4531                 old_val += num_bytes;
4532         else
4533                 old_val -= num_bytes;
4534         btrfs_set_super_bytes_used(info->super_copy, old_val);
4535         spin_unlock(&info->delalloc_lock);
4536
4537         while (total) {
4538                 cache = btrfs_lookup_block_group(info, bytenr);
4539                 if (!cache)
4540                         return -1;
4541                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4542                                     BTRFS_BLOCK_GROUP_RAID1 |
4543                                     BTRFS_BLOCK_GROUP_RAID10))
4544                         factor = 2;
4545                 else
4546                         factor = 1;
4547                 /*
4548                  * If this block group has free space cache written out, we
4549                  * need to make sure to load it if we are removing space.  This
4550                  * is because we need the unpinning stage to actually add the
4551                  * space back to the block group, otherwise we will leak space.
4552                  */
4553                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4554                         cache_block_group(cache, trans, NULL, 1);
4555
4556                 byte_in_group = bytenr - cache->key.objectid;
4557                 WARN_ON(byte_in_group > cache->key.offset);
4558
4559                 spin_lock(&cache->space_info->lock);
4560                 spin_lock(&cache->lock);
4561
4562                 if (btrfs_test_opt(root, SPACE_CACHE) &&
4563                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4564                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4565
4566                 cache->dirty = 1;
4567                 old_val = btrfs_block_group_used(&cache->item);
4568                 num_bytes = min(total, cache->key.offset - byte_in_group);
4569                 if (alloc) {
4570                         old_val += num_bytes;
4571                         btrfs_set_block_group_used(&cache->item, old_val);
4572                         cache->reserved -= num_bytes;
4573                         cache->space_info->bytes_reserved -= num_bytes;
4574                         cache->space_info->bytes_used += num_bytes;
4575                         cache->space_info->disk_used += num_bytes * factor;
4576                         spin_unlock(&cache->lock);
4577                         spin_unlock(&cache->space_info->lock);
4578                 } else {
4579                         old_val -= num_bytes;
4580                         btrfs_set_block_group_used(&cache->item, old_val);
4581                         cache->pinned += num_bytes;
4582                         cache->space_info->bytes_pinned += num_bytes;
4583                         cache->space_info->bytes_used -= num_bytes;
4584                         cache->space_info->disk_used -= num_bytes * factor;
4585                         spin_unlock(&cache->lock);
4586                         spin_unlock(&cache->space_info->lock);
4587
4588                         set_extent_dirty(info->pinned_extents,
4589                                          bytenr, bytenr + num_bytes - 1,
4590                                          GFP_NOFS | __GFP_NOFAIL);
4591                 }
4592                 btrfs_put_block_group(cache);
4593                 total -= num_bytes;
4594                 bytenr += num_bytes;
4595         }
4596         return 0;
4597 }
4598
4599 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4600 {
4601         struct btrfs_block_group_cache *cache;
4602         u64 bytenr;
4603
4604         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4605         if (!cache)
4606                 return 0;
4607
4608         bytenr = cache->key.objectid;
4609         btrfs_put_block_group(cache);
4610
4611         return bytenr;
4612 }
4613
4614 static int pin_down_extent(struct btrfs_root *root,
4615                            struct btrfs_block_group_cache *cache,
4616                            u64 bytenr, u64 num_bytes, int reserved)
4617 {
4618         spin_lock(&cache->space_info->lock);
4619         spin_lock(&cache->lock);
4620         cache->pinned += num_bytes;
4621         cache->space_info->bytes_pinned += num_bytes;
4622         if (reserved) {
4623                 cache->reserved -= num_bytes;
4624                 cache->space_info->bytes_reserved -= num_bytes;
4625         }
4626         spin_unlock(&cache->lock);
4627         spin_unlock(&cache->space_info->lock);
4628
4629         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4630                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4631         return 0;
4632 }
4633
4634 /*
4635  * this function must be called within transaction
4636  */
4637 int btrfs_pin_extent(struct btrfs_root *root,
4638                      u64 bytenr, u64 num_bytes, int reserved)
4639 {
4640         struct btrfs_block_group_cache *cache;
4641
4642         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4643         BUG_ON(!cache);
4644
4645         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4646
4647         btrfs_put_block_group(cache);
4648         return 0;
4649 }
4650
4651 /*
4652  * this function must be called within transaction
4653  */
4654 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4655                                     struct btrfs_root *root,
4656                                     u64 bytenr, u64 num_bytes)
4657 {
4658         struct btrfs_block_group_cache *cache;
4659
4660         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4661         BUG_ON(!cache);
4662
4663         /*
4664          * pull in the free space cache (if any) so that our pin
4665          * removes the free space from the cache.  We have load_only set
4666          * to one because the slow code to read in the free extents does check
4667          * the pinned extents.
4668          */
4669         cache_block_group(cache, trans, root, 1);
4670
4671         pin_down_extent(root, cache, bytenr, num_bytes, 0);
4672
4673         /* remove us from the free space cache (if we're there at all) */
4674         btrfs_remove_free_space(cache, bytenr, num_bytes);
4675         btrfs_put_block_group(cache);
4676         return 0;
4677 }
4678
4679 /**
4680  * btrfs_update_reserved_bytes - update the block_group and space info counters
4681  * @cache:      The cache we are manipulating
4682  * @num_bytes:  The number of bytes in question
4683  * @reserve:    One of the reservation enums
4684  *
4685  * This is called by the allocator when it reserves space, or by somebody who is
4686  * freeing space that was never actually used on disk.  For example if you
4687  * reserve some space for a new leaf in transaction A and before transaction A
4688  * commits you free that leaf, you call this with reserve set to 0 in order to
4689  * clear the reservation.
4690  *
4691  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4692  * ENOSPC accounting.  For data we handle the reservation through clearing the
4693  * delalloc bits in the io_tree.  We have to do this since we could end up
4694  * allocating less disk space for the amount of data we have reserved in the
4695  * case of compression.
4696  *
4697  * If this is a reservation and the block group has become read only we cannot
4698  * make the reservation and return -EAGAIN, otherwise this function always
4699  * succeeds.
4700  */
4701 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4702                                        u64 num_bytes, int reserve)
4703 {
4704         struct btrfs_space_info *space_info = cache->space_info;
4705         int ret = 0;
4706         spin_lock(&space_info->lock);
4707         spin_lock(&cache->lock);
4708         if (reserve != RESERVE_FREE) {
4709                 if (cache->ro) {
4710                         ret = -EAGAIN;
4711                 } else {
4712                         cache->reserved += num_bytes;
4713                         space_info->bytes_reserved += num_bytes;
4714                         if (reserve == RESERVE_ALLOC) {
4715                                 trace_btrfs_space_reservation(cache->fs_info,
4716                                                               "space_info",
4717                                                               (u64)space_info,
4718                                                               num_bytes, 0);
4719                                 space_info->bytes_may_use -= num_bytes;
4720                         }
4721                 }
4722         } else {
4723                 if (cache->ro)
4724                         space_info->bytes_readonly += num_bytes;
4725                 cache->reserved -= num_bytes;
4726                 space_info->bytes_reserved -= num_bytes;
4727                 space_info->reservation_progress++;
4728         }
4729         spin_unlock(&cache->lock);
4730         spin_unlock(&space_info->lock);
4731         return ret;
4732 }
4733
4734 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4735                                 struct btrfs_root *root)
4736 {
4737         struct btrfs_fs_info *fs_info = root->fs_info;
4738         struct btrfs_caching_control *next;
4739         struct btrfs_caching_control *caching_ctl;
4740         struct btrfs_block_group_cache *cache;
4741
4742         down_write(&fs_info->extent_commit_sem);
4743
4744         list_for_each_entry_safe(caching_ctl, next,
4745                                  &fs_info->caching_block_groups, list) {
4746                 cache = caching_ctl->block_group;
4747                 if (block_group_cache_done(cache)) {
4748                         cache->last_byte_to_unpin = (u64)-1;
4749                         list_del_init(&caching_ctl->list);
4750                         put_caching_control(caching_ctl);
4751                 } else {
4752                         cache->last_byte_to_unpin = caching_ctl->progress;
4753                 }
4754         }
4755
4756         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4757                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4758         else
4759                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4760
4761         up_write(&fs_info->extent_commit_sem);
4762
4763         update_global_block_rsv(fs_info);
4764         return 0;
4765 }
4766
4767 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4768 {
4769         struct btrfs_fs_info *fs_info = root->fs_info;
4770         struct btrfs_block_group_cache *cache = NULL;
4771         u64 len;
4772
4773         while (start <= end) {
4774                 if (!cache ||
4775                     start >= cache->key.objectid + cache->key.offset) {
4776                         if (cache)
4777                                 btrfs_put_block_group(cache);
4778                         cache = btrfs_lookup_block_group(fs_info, start);
4779                         BUG_ON(!cache);
4780                 }
4781
4782                 len = cache->key.objectid + cache->key.offset - start;
4783                 len = min(len, end + 1 - start);
4784
4785                 if (start < cache->last_byte_to_unpin) {
4786                         len = min(len, cache->last_byte_to_unpin - start);
4787                         btrfs_add_free_space(cache, start, len);
4788                 }
4789
4790                 start += len;
4791
4792                 spin_lock(&cache->space_info->lock);
4793                 spin_lock(&cache->lock);
4794                 cache->pinned -= len;
4795                 cache->space_info->bytes_pinned -= len;
4796                 if (cache->ro)
4797                         cache->space_info->bytes_readonly += len;
4798                 spin_unlock(&cache->lock);
4799                 spin_unlock(&cache->space_info->lock);
4800         }
4801
4802         if (cache)
4803                 btrfs_put_block_group(cache);
4804         return 0;
4805 }
4806
4807 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4808                                struct btrfs_root *root)
4809 {
4810         struct btrfs_fs_info *fs_info = root->fs_info;
4811         struct extent_io_tree *unpin;
4812         u64 start;
4813         u64 end;
4814         int ret;
4815
4816         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4817                 unpin = &fs_info->freed_extents[1];
4818         else
4819                 unpin = &fs_info->freed_extents[0];
4820
4821         while (1) {
4822                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4823                                             EXTENT_DIRTY);
4824                 if (ret)
4825                         break;
4826
4827                 if (btrfs_test_opt(root, DISCARD))
4828                         ret = btrfs_discard_extent(root, start,
4829                                                    end + 1 - start, NULL);
4830
4831                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4832                 unpin_extent_range(root, start, end);
4833                 cond_resched();
4834         }
4835
4836         return 0;
4837 }
4838
4839 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4840                                 struct btrfs_root *root,
4841                                 u64 bytenr, u64 num_bytes, u64 parent,
4842                                 u64 root_objectid, u64 owner_objectid,
4843                                 u64 owner_offset, int refs_to_drop,
4844                                 struct btrfs_delayed_extent_op *extent_op)
4845 {
4846         struct btrfs_key key;
4847         struct btrfs_path *path;
4848         struct btrfs_fs_info *info = root->fs_info;
4849         struct btrfs_root *extent_root = info->extent_root;
4850         struct extent_buffer *leaf;
4851         struct btrfs_extent_item *ei;
4852         struct btrfs_extent_inline_ref *iref;
4853         int ret;
4854         int is_data;
4855         int extent_slot = 0;
4856         int found_extent = 0;
4857         int num_to_del = 1;
4858         u32 item_size;
4859         u64 refs;
4860
4861         path = btrfs_alloc_path();
4862         if (!path)
4863                 return -ENOMEM;
4864
4865         path->reada = 1;
4866         path->leave_spinning = 1;
4867
4868         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4869         BUG_ON(!is_data && refs_to_drop != 1);
4870
4871         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4872                                     bytenr, num_bytes, parent,
4873                                     root_objectid, owner_objectid,
4874                                     owner_offset);
4875         if (ret == 0) {
4876                 extent_slot = path->slots[0];
4877                 while (extent_slot >= 0) {
4878                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4879                                               extent_slot);
4880                         if (key.objectid != bytenr)
4881                                 break;
4882                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4883                             key.offset == num_bytes) {
4884                                 found_extent = 1;
4885                                 break;
4886                         }
4887                         if (path->slots[0] - extent_slot > 5)
4888                                 break;
4889                         extent_slot--;
4890                 }
4891 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4892                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4893                 if (found_extent && item_size < sizeof(*ei))
4894                         found_extent = 0;
4895 #endif
4896                 if (!found_extent) {
4897                         BUG_ON(iref);
4898                         ret = remove_extent_backref(trans, extent_root, path,
4899                                                     NULL, refs_to_drop,
4900                                                     is_data);
4901                         BUG_ON(ret);
4902                         btrfs_release_path(path);
4903                         path->leave_spinning = 1;
4904
4905                         key.objectid = bytenr;
4906                         key.type = BTRFS_EXTENT_ITEM_KEY;
4907                         key.offset = num_bytes;
4908
4909                         ret = btrfs_search_slot(trans, extent_root,
4910                                                 &key, path, -1, 1);
4911                         if (ret) {
4912                                 printk(KERN_ERR "umm, got %d back from search"
4913                                        ", was looking for %llu\n", ret,
4914                                        (unsigned long long)bytenr);
4915                                 if (ret > 0)
4916                                         btrfs_print_leaf(extent_root,
4917                                                          path->nodes[0]);
4918                         }
4919                         BUG_ON(ret);
4920                         extent_slot = path->slots[0];
4921                 }
4922         } else {
4923                 btrfs_print_leaf(extent_root, path->nodes[0]);
4924                 WARN_ON(1);
4925                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4926                        "parent %llu root %llu  owner %llu offset %llu\n",
4927                        (unsigned long long)bytenr,
4928                        (unsigned long long)parent,
4929                        (unsigned long long)root_objectid,
4930                        (unsigned long long)owner_objectid,
4931                        (unsigned long long)owner_offset);
4932         }
4933
4934         leaf = path->nodes[0];
4935         item_size = btrfs_item_size_nr(leaf, extent_slot);
4936 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4937         if (item_size < sizeof(*ei)) {
4938                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4939                 ret = convert_extent_item_v0(trans, extent_root, path,
4940                                              owner_objectid, 0);
4941                 BUG_ON(ret < 0);
4942
4943                 btrfs_release_path(path);
4944                 path->leave_spinning = 1;
4945
4946                 key.objectid = bytenr;
4947                 key.type = BTRFS_EXTENT_ITEM_KEY;
4948                 key.offset = num_bytes;
4949
4950                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4951                                         -1, 1);
4952                 if (ret) {
4953                         printk(KERN_ERR "umm, got %d back from search"
4954                                ", was looking for %llu\n", ret,
4955                                (unsigned long long)bytenr);
4956                         btrfs_print_leaf(extent_root, path->nodes[0]);
4957                 }
4958                 BUG_ON(ret);
4959                 extent_slot = path->slots[0];
4960                 leaf = path->nodes[0];
4961                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4962         }
4963 #endif
4964         BUG_ON(item_size < sizeof(*ei));
4965         ei = btrfs_item_ptr(leaf, extent_slot,
4966                             struct btrfs_extent_item);
4967         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4968                 struct btrfs_tree_block_info *bi;
4969                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4970                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4971                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4972         }
4973
4974         refs = btrfs_extent_refs(leaf, ei);
4975         BUG_ON(refs < refs_to_drop);
4976         refs -= refs_to_drop;
4977
4978         if (refs > 0) {
4979                 if (extent_op)
4980                         __run_delayed_extent_op(extent_op, leaf, ei);
4981                 /*
4982                  * In the case of inline back ref, reference count will
4983                  * be updated by remove_extent_backref
4984                  */
4985                 if (iref) {
4986                         BUG_ON(!found_extent);
4987                 } else {
4988                         btrfs_set_extent_refs(leaf, ei, refs);
4989                         btrfs_mark_buffer_dirty(leaf);
4990                 }
4991                 if (found_extent) {
4992                         ret = remove_extent_backref(trans, extent_root, path,
4993                                                     iref, refs_to_drop,
4994                                                     is_data);
4995                         BUG_ON(ret);
4996                 }
4997         } else {
4998                 if (found_extent) {
4999                         BUG_ON(is_data && refs_to_drop !=
5000                                extent_data_ref_count(root, path, iref));
5001                         if (iref) {
5002                                 BUG_ON(path->slots[0] != extent_slot);
5003                         } else {
5004                                 BUG_ON(path->slots[0] != extent_slot + 1);
5005                                 path->slots[0] = extent_slot;
5006                                 num_to_del = 2;
5007                         }
5008                 }
5009
5010                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5011                                       num_to_del);
5012                 BUG_ON(ret);
5013                 btrfs_release_path(path);
5014
5015                 if (is_data) {
5016                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5017                         BUG_ON(ret);
5018                 } else {
5019                         invalidate_mapping_pages(info->btree_inode->i_mapping,
5020                              bytenr >> PAGE_CACHE_SHIFT,
5021                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
5022                 }
5023
5024                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5025                 BUG_ON(ret);
5026         }
5027         btrfs_free_path(path);
5028         return ret;
5029 }
5030
5031 /*
5032  * when we free an block, it is possible (and likely) that we free the last
5033  * delayed ref for that extent as well.  This searches the delayed ref tree for
5034  * a given extent, and if there are no other delayed refs to be processed, it
5035  * removes it from the tree.
5036  */
5037 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5038                                       struct btrfs_root *root, u64 bytenr)
5039 {
5040         struct btrfs_delayed_ref_head *head;
5041         struct btrfs_delayed_ref_root *delayed_refs;
5042         struct btrfs_delayed_ref_node *ref;
5043         struct rb_node *node;
5044         int ret = 0;
5045
5046         delayed_refs = &trans->transaction->delayed_refs;
5047         spin_lock(&delayed_refs->lock);
5048         head = btrfs_find_delayed_ref_head(trans, bytenr);
5049         if (!head)
5050                 goto out;
5051
5052         node = rb_prev(&head->node.rb_node);
5053         if (!node)
5054                 goto out;
5055
5056         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5057
5058         /* there are still entries for this ref, we can't drop it */
5059         if (ref->bytenr == bytenr)
5060                 goto out;
5061
5062         if (head->extent_op) {
5063                 if (!head->must_insert_reserved)
5064                         goto out;
5065                 kfree(head->extent_op);
5066                 head->extent_op = NULL;
5067         }
5068
5069         /*
5070          * waiting for the lock here would deadlock.  If someone else has it
5071          * locked they are already in the process of dropping it anyway
5072          */
5073         if (!mutex_trylock(&head->mutex))
5074                 goto out;
5075
5076         /*
5077          * at this point we have a head with no other entries.  Go
5078          * ahead and process it.
5079          */
5080         head->node.in_tree = 0;
5081         rb_erase(&head->node.rb_node, &delayed_refs->root);
5082
5083         delayed_refs->num_entries--;
5084         if (waitqueue_active(&delayed_refs->seq_wait))
5085                 wake_up(&delayed_refs->seq_wait);
5086
5087         /*
5088          * we don't take a ref on the node because we're removing it from the
5089          * tree, so we just steal the ref the tree was holding.
5090          */
5091         delayed_refs->num_heads--;
5092         if (list_empty(&head->cluster))
5093                 delayed_refs->num_heads_ready--;
5094
5095         list_del_init(&head->cluster);
5096         spin_unlock(&delayed_refs->lock);
5097
5098         BUG_ON(head->extent_op);
5099         if (head->must_insert_reserved)
5100                 ret = 1;
5101
5102         mutex_unlock(&head->mutex);
5103         btrfs_put_delayed_ref(&head->node);
5104         return ret;
5105 out:
5106         spin_unlock(&delayed_refs->lock);
5107         return 0;
5108 }
5109
5110 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5111                            struct btrfs_root *root,
5112                            struct extent_buffer *buf,
5113                            u64 parent, int last_ref, int for_cow)
5114 {
5115         struct btrfs_block_group_cache *cache = NULL;
5116         int ret;
5117
5118         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5119                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5120                                         buf->start, buf->len,
5121                                         parent, root->root_key.objectid,
5122                                         btrfs_header_level(buf),
5123                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5124                 BUG_ON(ret);
5125         }
5126
5127         if (!last_ref)
5128                 return;
5129
5130         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5131
5132         if (btrfs_header_generation(buf) == trans->transid) {
5133                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5134                         ret = check_ref_cleanup(trans, root, buf->start);
5135                         if (!ret)
5136                                 goto out;
5137                 }
5138
5139                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5140                         pin_down_extent(root, cache, buf->start, buf->len, 1);
5141                         goto out;
5142                 }
5143
5144                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5145
5146                 btrfs_add_free_space(cache, buf->start, buf->len);
5147                 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5148         }
5149 out:
5150         /*
5151          * Deleting the buffer, clear the corrupt flag since it doesn't matter
5152          * anymore.
5153          */
5154         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5155         btrfs_put_block_group(cache);
5156 }
5157
5158 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5159                       u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5160                       u64 owner, u64 offset, int for_cow)
5161 {
5162         int ret;
5163         struct btrfs_fs_info *fs_info = root->fs_info;
5164
5165         /*
5166          * tree log blocks never actually go into the extent allocation
5167          * tree, just update pinning info and exit early.
5168          */
5169         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5170                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5171                 /* unlocks the pinned mutex */
5172                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5173                 ret = 0;
5174         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5175                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5176                                         num_bytes,
5177                                         parent, root_objectid, (int)owner,
5178                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5179                 BUG_ON(ret);
5180         } else {
5181                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5182                                                 num_bytes,
5183                                                 parent, root_objectid, owner,
5184                                                 offset, BTRFS_DROP_DELAYED_REF,
5185                                                 NULL, for_cow);
5186                 BUG_ON(ret);
5187         }
5188         return ret;
5189 }
5190
5191 static u64 stripe_align(struct btrfs_root *root, u64 val)
5192 {
5193         u64 mask = ((u64)root->stripesize - 1);
5194         u64 ret = (val + mask) & ~mask;
5195         return ret;
5196 }
5197
5198 /*
5199  * when we wait for progress in the block group caching, its because
5200  * our allocation attempt failed at least once.  So, we must sleep
5201  * and let some progress happen before we try again.
5202  *
5203  * This function will sleep at least once waiting for new free space to
5204  * show up, and then it will check the block group free space numbers
5205  * for our min num_bytes.  Another option is to have it go ahead
5206  * and look in the rbtree for a free extent of a given size, but this
5207  * is a good start.
5208  */
5209 static noinline int
5210 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5211                                 u64 num_bytes)
5212 {
5213         struct btrfs_caching_control *caching_ctl;
5214         DEFINE_WAIT(wait);
5215
5216         caching_ctl = get_caching_control(cache);
5217         if (!caching_ctl)
5218                 return 0;
5219
5220         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5221                    (cache->free_space_ctl->free_space >= num_bytes));
5222
5223         put_caching_control(caching_ctl);
5224         return 0;
5225 }
5226
5227 static noinline int
5228 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5229 {
5230         struct btrfs_caching_control *caching_ctl;
5231         DEFINE_WAIT(wait);
5232
5233         caching_ctl = get_caching_control(cache);
5234         if (!caching_ctl)
5235                 return 0;
5236
5237         wait_event(caching_ctl->wait, block_group_cache_done(cache));
5238
5239         put_caching_control(caching_ctl);
5240         return 0;
5241 }
5242
5243 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5244 {
5245         int index;
5246         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5247                 index = 0;
5248         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5249                 index = 1;
5250         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5251                 index = 2;
5252         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5253                 index = 3;
5254         else
5255                 index = 4;
5256         return index;
5257 }
5258
5259 enum btrfs_loop_type {
5260         LOOP_FIND_IDEAL = 0,
5261         LOOP_CACHING_NOWAIT = 1,
5262         LOOP_CACHING_WAIT = 2,
5263         LOOP_ALLOC_CHUNK = 3,
5264         LOOP_NO_EMPTY_SIZE = 4,
5265 };
5266
5267 /*
5268  * walks the btree of allocated extents and find a hole of a given size.
5269  * The key ins is changed to record the hole:
5270  * ins->objectid == block start
5271  * ins->flags = BTRFS_EXTENT_ITEM_KEY
5272  * ins->offset == number of blocks
5273  * Any available blocks before search_start are skipped.
5274  */
5275 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5276                                      struct btrfs_root *orig_root,
5277                                      u64 num_bytes, u64 empty_size,
5278                                      u64 search_start, u64 search_end,
5279                                      u64 hint_byte, struct btrfs_key *ins,
5280                                      u64 data)
5281 {
5282         int ret = 0;
5283         struct btrfs_root *root = orig_root->fs_info->extent_root;
5284         struct btrfs_free_cluster *last_ptr = NULL;
5285         struct btrfs_block_group_cache *block_group = NULL;
5286         struct btrfs_block_group_cache *used_block_group;
5287         int empty_cluster = 2 * 1024 * 1024;
5288         int allowed_chunk_alloc = 0;
5289         int done_chunk_alloc = 0;
5290         struct btrfs_space_info *space_info;
5291         int loop = 0;
5292         int index = 0;
5293         int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5294                 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5295         bool found_uncached_bg = false;
5296         bool failed_cluster_refill = false;
5297         bool failed_alloc = false;
5298         bool use_cluster = true;
5299         bool have_caching_bg = false;
5300         u64 ideal_cache_percent = 0;
5301         u64 ideal_cache_offset = 0;
5302
5303         WARN_ON(num_bytes < root->sectorsize);
5304         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5305         ins->objectid = 0;
5306         ins->offset = 0;
5307
5308         trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5309
5310         space_info = __find_space_info(root->fs_info, data);
5311         if (!space_info) {
5312                 printk(KERN_ERR "No space info for %llu\n", data);
5313                 return -ENOSPC;
5314         }
5315
5316         /*
5317          * If the space info is for both data and metadata it means we have a
5318          * small filesystem and we can't use the clustering stuff.
5319          */
5320         if (btrfs_mixed_space_info(space_info))
5321                 use_cluster = false;
5322
5323         if (orig_root->ref_cows || empty_size)
5324                 allowed_chunk_alloc = 1;
5325
5326         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5327                 last_ptr = &root->fs_info->meta_alloc_cluster;
5328                 if (!btrfs_test_opt(root, SSD))
5329                         empty_cluster = 64 * 1024;
5330         }
5331
5332         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5333             btrfs_test_opt(root, SSD)) {
5334                 last_ptr = &root->fs_info->data_alloc_cluster;
5335         }
5336
5337         if (last_ptr) {
5338                 spin_lock(&last_ptr->lock);
5339                 if (last_ptr->block_group)
5340                         hint_byte = last_ptr->window_start;
5341                 spin_unlock(&last_ptr->lock);
5342         }
5343
5344         search_start = max(search_start, first_logical_byte(root, 0));
5345         search_start = max(search_start, hint_byte);
5346
5347         if (!last_ptr)
5348                 empty_cluster = 0;
5349
5350         if (search_start == hint_byte) {
5351 ideal_cache:
5352                 block_group = btrfs_lookup_block_group(root->fs_info,
5353                                                        search_start);
5354                 used_block_group = block_group;
5355                 /*
5356                  * we don't want to use the block group if it doesn't match our
5357                  * allocation bits, or if its not cached.
5358                  *
5359                  * However if we are re-searching with an ideal block group
5360                  * picked out then we don't care that the block group is cached.
5361                  */
5362                 if (block_group && block_group_bits(block_group, data) &&
5363                     (block_group->cached != BTRFS_CACHE_NO ||
5364                      search_start == ideal_cache_offset)) {
5365                         down_read(&space_info->groups_sem);
5366                         if (list_empty(&block_group->list) ||
5367                             block_group->ro) {
5368                                 /*
5369                                  * someone is removing this block group,
5370                                  * we can't jump into the have_block_group
5371                                  * target because our list pointers are not
5372                                  * valid
5373                                  */
5374                                 btrfs_put_block_group(block_group);
5375                                 up_read(&space_info->groups_sem);
5376                         } else {
5377                                 index = get_block_group_index(block_group);
5378                                 goto have_block_group;
5379                         }
5380                 } else if (block_group) {
5381                         btrfs_put_block_group(block_group);
5382                 }
5383         }
5384 search:
5385         have_caching_bg = false;
5386         down_read(&space_info->groups_sem);
5387         list_for_each_entry(block_group, &space_info->block_groups[index],
5388                             list) {
5389                 u64 offset;
5390                 int cached;
5391
5392                 used_block_group = block_group;
5393                 btrfs_get_block_group(block_group);
5394                 search_start = block_group->key.objectid;
5395
5396                 /*
5397                  * this can happen if we end up cycling through all the
5398                  * raid types, but we want to make sure we only allocate
5399                  * for the proper type.
5400                  */
5401                 if (!block_group_bits(block_group, data)) {
5402                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
5403                                 BTRFS_BLOCK_GROUP_RAID1 |
5404                                 BTRFS_BLOCK_GROUP_RAID10;
5405
5406                         /*
5407                          * if they asked for extra copies and this block group
5408                          * doesn't provide them, bail.  This does allow us to
5409                          * fill raid0 from raid1.
5410                          */
5411                         if ((data & extra) && !(block_group->flags & extra))
5412                                 goto loop;
5413                 }
5414
5415 have_block_group:
5416                 cached = block_group_cache_done(block_group);
5417                 if (unlikely(!cached)) {
5418                         u64 free_percent;
5419
5420                         found_uncached_bg = true;
5421                         ret = cache_block_group(block_group, trans,
5422                                                 orig_root, 1);
5423                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5424                                 goto alloc;
5425
5426                         free_percent = btrfs_block_group_used(&block_group->item);
5427                         free_percent *= 100;
5428                         free_percent = div64_u64(free_percent,
5429                                                  block_group->key.offset);
5430                         free_percent = 100 - free_percent;
5431                         if (free_percent > ideal_cache_percent &&
5432                             likely(!block_group->ro)) {
5433                                 ideal_cache_offset = block_group->key.objectid;
5434                                 ideal_cache_percent = free_percent;
5435                         }
5436
5437                         /*
5438                          * The caching workers are limited to 2 threads, so we
5439                          * can queue as much work as we care to.
5440                          */
5441                         if (loop > LOOP_FIND_IDEAL) {
5442                                 ret = cache_block_group(block_group, trans,
5443                                                         orig_root, 0);
5444                                 BUG_ON(ret);
5445                         }
5446
5447                         /*
5448                          * If loop is set for cached only, try the next block
5449                          * group.
5450                          */
5451                         if (loop == LOOP_FIND_IDEAL)
5452                                 goto loop;
5453                 }
5454
5455 alloc:
5456                 if (unlikely(block_group->ro))
5457                         goto loop;
5458
5459                 /*
5460                  * Ok we want to try and use the cluster allocator, so
5461                  * lets look there
5462                  */
5463                 if (last_ptr) {
5464                         /*
5465                          * the refill lock keeps out other
5466                          * people trying to start a new cluster
5467                          */
5468                         spin_lock(&last_ptr->refill_lock);
5469                         used_block_group = last_ptr->block_group;
5470                         if (used_block_group != block_group &&
5471                             (!used_block_group ||
5472                              used_block_group->ro ||
5473                              !block_group_bits(used_block_group, data))) {
5474                                 used_block_group = block_group;
5475                                 goto refill_cluster;
5476                         }
5477
5478                         if (used_block_group != block_group)
5479                                 btrfs_get_block_group(used_block_group);
5480
5481                         offset = btrfs_alloc_from_cluster(used_block_group,
5482                           last_ptr, num_bytes, used_block_group->key.objectid);
5483                         if (offset) {
5484                                 /* we have a block, we're done */
5485                                 spin_unlock(&last_ptr->refill_lock);
5486                                 trace_btrfs_reserve_extent_cluster(root,
5487                                         block_group, search_start, num_bytes);
5488                                 goto checks;
5489                         }
5490
5491                         WARN_ON(last_ptr->block_group != used_block_group);
5492                         if (used_block_group != block_group) {
5493                                 btrfs_put_block_group(used_block_group);
5494                                 used_block_group = block_group;
5495                         }
5496 refill_cluster:
5497                         BUG_ON(used_block_group != block_group);
5498                         /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5499                          * set up a new clusters, so lets just skip it
5500                          * and let the allocator find whatever block
5501                          * it can find.  If we reach this point, we
5502                          * will have tried the cluster allocator
5503                          * plenty of times and not have found
5504                          * anything, so we are likely way too
5505                          * fragmented for the clustering stuff to find
5506                          * anything.
5507                          *
5508                          * However, if the cluster is taken from the
5509                          * current block group, release the cluster
5510                          * first, so that we stand a better chance of
5511                          * succeeding in the unclustered
5512                          * allocation.  */
5513                         if (loop >= LOOP_NO_EMPTY_SIZE &&
5514                             last_ptr->block_group != block_group) {
5515                                 spin_unlock(&last_ptr->refill_lock);
5516                                 goto unclustered_alloc;
5517                         }
5518
5519                         /*
5520                          * this cluster didn't work out, free it and
5521                          * start over
5522                          */
5523                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5524
5525                         if (loop >= LOOP_NO_EMPTY_SIZE) {
5526                                 spin_unlock(&last_ptr->refill_lock);
5527                                 goto unclustered_alloc;
5528                         }
5529
5530                         /* allocate a cluster in this block group */
5531                         ret = btrfs_find_space_cluster(trans, root,
5532                                                block_group, last_ptr,
5533                                                search_start, num_bytes,
5534                                                empty_cluster + empty_size);
5535                         if (ret == 0) {
5536                                 /*
5537                                  * now pull our allocation out of this
5538                                  * cluster
5539                                  */
5540                                 offset = btrfs_alloc_from_cluster(block_group,
5541                                                   last_ptr, num_bytes,
5542                                                   search_start);
5543                                 if (offset) {
5544                                         /* we found one, proceed */
5545                                         spin_unlock(&last_ptr->refill_lock);
5546                                         trace_btrfs_reserve_extent_cluster(root,
5547                                                 block_group, search_start,
5548                                                 num_bytes);
5549                                         goto checks;
5550                                 }
5551                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5552                                    && !failed_cluster_refill) {
5553                                 spin_unlock(&last_ptr->refill_lock);
5554
5555                                 failed_cluster_refill = true;
5556                                 wait_block_group_cache_progress(block_group,
5557                                        num_bytes + empty_cluster + empty_size);
5558                                 goto have_block_group;
5559                         }
5560
5561                         /*
5562                          * at this point we either didn't find a cluster
5563                          * or we weren't able to allocate a block from our
5564                          * cluster.  Free the cluster we've been trying
5565                          * to use, and go to the next block group
5566                          */
5567                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5568                         spin_unlock(&last_ptr->refill_lock);
5569                         goto loop;
5570                 }
5571
5572 unclustered_alloc:
5573                 spin_lock(&block_group->free_space_ctl->tree_lock);
5574                 if (cached &&
5575                     block_group->free_space_ctl->free_space <
5576                     num_bytes + empty_cluster + empty_size) {
5577                         spin_unlock(&block_group->free_space_ctl->tree_lock);
5578                         goto loop;
5579                 }
5580                 spin_unlock(&block_group->free_space_ctl->tree_lock);
5581
5582                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5583                                                     num_bytes, empty_size);
5584                 /*
5585                  * If we didn't find a chunk, and we haven't failed on this
5586                  * block group before, and this block group is in the middle of
5587                  * caching and we are ok with waiting, then go ahead and wait
5588                  * for progress to be made, and set failed_alloc to true.
5589                  *
5590                  * If failed_alloc is true then we've already waited on this
5591                  * block group once and should move on to the next block group.
5592                  */
5593                 if (!offset && !failed_alloc && !cached &&
5594                     loop > LOOP_CACHING_NOWAIT) {
5595                         wait_block_group_cache_progress(block_group,
5596                                                 num_bytes + empty_size);
5597                         failed_alloc = true;
5598                         goto have_block_group;
5599                 } else if (!offset) {
5600                         if (!cached)
5601                                 have_caching_bg = true;
5602                         goto loop;
5603                 }
5604 checks:
5605                 search_start = stripe_align(root, offset);
5606                 /* move on to the next group */
5607                 if (search_start + num_bytes >= search_end) {
5608                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5609                         goto loop;
5610                 }
5611
5612                 /* move on to the next group */
5613                 if (search_start + num_bytes >
5614                     used_block_group->key.objectid + used_block_group->key.offset) {
5615                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5616                         goto loop;
5617                 }
5618
5619                 if (offset < search_start)
5620                         btrfs_add_free_space(used_block_group, offset,
5621                                              search_start - offset);
5622                 BUG_ON(offset > search_start);
5623
5624                 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5625                                                   alloc_type);
5626                 if (ret == -EAGAIN) {
5627                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5628                         goto loop;
5629                 }
5630
5631                 /* we are all good, lets return */
5632                 ins->objectid = search_start;
5633                 ins->offset = num_bytes;
5634
5635                 trace_btrfs_reserve_extent(orig_root, block_group,
5636                                            search_start, num_bytes);
5637                 if (offset < search_start)
5638                         btrfs_add_free_space(used_block_group, offset,
5639                                              search_start - offset);
5640                 BUG_ON(offset > search_start);
5641                 if (used_block_group != block_group)
5642                         btrfs_put_block_group(used_block_group);
5643                 btrfs_put_block_group(block_group);
5644                 break;
5645 loop:
5646                 failed_cluster_refill = false;
5647                 failed_alloc = false;
5648                 BUG_ON(index != get_block_group_index(block_group));
5649                 if (used_block_group != block_group)
5650                         btrfs_put_block_group(used_block_group);
5651                 btrfs_put_block_group(block_group);
5652         }
5653         up_read(&space_info->groups_sem);
5654
5655         if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5656                 goto search;
5657
5658         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5659                 goto search;
5660
5661         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5662          *                      for them to make caching progress.  Also
5663          *                      determine the best possible bg to cache
5664          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5665          *                      caching kthreads as we move along
5666          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5667          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5668          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5669          *                      again
5670          */
5671         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5672                 index = 0;
5673                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5674                         found_uncached_bg = false;
5675                         loop++;
5676                         if (!ideal_cache_percent)
5677                                 goto search;
5678
5679                         /*
5680                          * 1 of the following 2 things have happened so far
5681                          *
5682                          * 1) We found an ideal block group for caching that
5683                          * is mostly full and will cache quickly, so we might
5684                          * as well wait for it.
5685                          *
5686                          * 2) We searched for cached only and we didn't find
5687                          * anything, and we didn't start any caching kthreads
5688                          * either, so chances are we will loop through and
5689                          * start a couple caching kthreads, and then come back
5690                          * around and just wait for them.  This will be slower
5691                          * because we will have 2 caching kthreads reading at
5692                          * the same time when we could have just started one
5693                          * and waited for it to get far enough to give us an
5694                          * allocation, so go ahead and go to the wait caching
5695                          * loop.
5696                          */
5697                         loop = LOOP_CACHING_WAIT;
5698                         search_start = ideal_cache_offset;
5699                         ideal_cache_percent = 0;
5700                         goto ideal_cache;
5701                 } else if (loop == LOOP_FIND_IDEAL) {
5702                         /*
5703                          * Didn't find a uncached bg, wait on anything we find
5704                          * next.
5705                          */
5706                         loop = LOOP_CACHING_WAIT;
5707                         goto search;
5708                 }
5709
5710                 loop++;
5711
5712                 if (loop == LOOP_ALLOC_CHUNK) {
5713                        if (allowed_chunk_alloc) {
5714                                 ret = do_chunk_alloc(trans, root, num_bytes +
5715                                                      2 * 1024 * 1024, data,
5716                                                      CHUNK_ALLOC_LIMITED);
5717                                 allowed_chunk_alloc = 0;
5718                                 if (ret == 1)
5719                                         done_chunk_alloc = 1;
5720                         } else if (!done_chunk_alloc &&
5721                                    space_info->force_alloc ==
5722                                    CHUNK_ALLOC_NO_FORCE) {
5723                                 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5724                         }
5725
5726                        /*
5727                         * We didn't allocate a chunk, go ahead and drop the
5728                         * empty size and loop again.
5729                         */
5730                        if (!done_chunk_alloc)
5731                                loop = LOOP_NO_EMPTY_SIZE;
5732                 }
5733
5734                 if (loop == LOOP_NO_EMPTY_SIZE) {
5735                         empty_size = 0;
5736                         empty_cluster = 0;
5737                 }
5738
5739                 goto search;
5740         } else if (!ins->objectid) {
5741                 ret = -ENOSPC;
5742         } else if (ins->objectid) {
5743                 ret = 0;
5744         }
5745
5746         return ret;
5747 }
5748
5749 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5750                             int dump_block_groups)
5751 {
5752         struct btrfs_block_group_cache *cache;
5753         int index = 0;
5754
5755         spin_lock(&info->lock);
5756         printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5757                (unsigned long long)info->flags,
5758                (unsigned long long)(info->total_bytes - info->bytes_used -
5759                                     info->bytes_pinned - info->bytes_reserved -
5760                                     info->bytes_readonly),
5761                (info->full) ? "" : "not ");
5762         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5763                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5764                (unsigned long long)info->total_bytes,
5765                (unsigned long long)info->bytes_used,
5766                (unsigned long long)info->bytes_pinned,
5767                (unsigned long long)info->bytes_reserved,
5768                (unsigned long long)info->bytes_may_use,
5769                (unsigned long long)info->bytes_readonly);
5770         spin_unlock(&info->lock);
5771
5772         if (!dump_block_groups)
5773                 return;
5774
5775         down_read(&info->groups_sem);
5776 again:
5777         list_for_each_entry(cache, &info->block_groups[index], list) {
5778                 spin_lock(&cache->lock);
5779                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5780                        "%llu pinned %llu reserved\n",
5781                        (unsigned long long)cache->key.objectid,
5782                        (unsigned long long)cache->key.offset,
5783                        (unsigned long long)btrfs_block_group_used(&cache->item),
5784                        (unsigned long long)cache->pinned,
5785                        (unsigned long long)cache->reserved);
5786                 btrfs_dump_free_space(cache, bytes);
5787                 spin_unlock(&cache->lock);
5788         }
5789         if (++index < BTRFS_NR_RAID_TYPES)
5790                 goto again;
5791         up_read(&info->groups_sem);
5792 }
5793
5794 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5795                          struct btrfs_root *root,
5796                          u64 num_bytes, u64 min_alloc_size,
5797                          u64 empty_size, u64 hint_byte,
5798                          u64 search_end, struct btrfs_key *ins,
5799                          u64 data)
5800 {
5801         bool final_tried = false;
5802         int ret;
5803         u64 search_start = 0;
5804
5805         data = btrfs_get_alloc_profile(root, data);
5806 again:
5807         /*
5808          * the only place that sets empty_size is btrfs_realloc_node, which
5809          * is not called recursively on allocations
5810          */
5811         if (empty_size || root->ref_cows)
5812                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5813                                      num_bytes + 2 * 1024 * 1024, data,
5814                                      CHUNK_ALLOC_NO_FORCE);
5815
5816         WARN_ON(num_bytes < root->sectorsize);
5817         ret = find_free_extent(trans, root, num_bytes, empty_size,
5818                                search_start, search_end, hint_byte,
5819                                ins, data);
5820
5821         if (ret == -ENOSPC) {
5822                 if (!final_tried) {
5823                         num_bytes = num_bytes >> 1;
5824                         num_bytes = num_bytes & ~(root->sectorsize - 1);
5825                         num_bytes = max(num_bytes, min_alloc_size);
5826                         do_chunk_alloc(trans, root->fs_info->extent_root,
5827                                        num_bytes, data, CHUNK_ALLOC_FORCE);
5828                         if (num_bytes == min_alloc_size)
5829                                 final_tried = true;
5830                         goto again;
5831                 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5832                         struct btrfs_space_info *sinfo;
5833
5834                         sinfo = __find_space_info(root->fs_info, data);
5835                         printk(KERN_ERR "btrfs allocation failed flags %llu, "
5836                                "wanted %llu\n", (unsigned long long)data,
5837                                (unsigned long long)num_bytes);
5838                         dump_space_info(sinfo, num_bytes, 1);
5839                 }
5840         }
5841
5842         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5843
5844         return ret;
5845 }
5846
5847 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5848                                         u64 start, u64 len, int pin)
5849 {
5850         struct btrfs_block_group_cache *cache;
5851         int ret = 0;
5852
5853         cache = btrfs_lookup_block_group(root->fs_info, start);
5854         if (!cache) {
5855                 printk(KERN_ERR "Unable to find block group for %llu\n",
5856                        (unsigned long long)start);
5857                 return -ENOSPC;
5858         }
5859
5860         if (btrfs_test_opt(root, DISCARD))
5861                 ret = btrfs_discard_extent(root, start, len, NULL);
5862
5863         if (pin)
5864                 pin_down_extent(root, cache, start, len, 1);
5865         else {
5866                 btrfs_add_free_space(cache, start, len);
5867                 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5868         }
5869         btrfs_put_block_group(cache);
5870
5871         trace_btrfs_reserved_extent_free(root, start, len);
5872
5873         return ret;
5874 }
5875
5876 int btrfs_free_reserved_extent(struct btrfs_root *root,
5877                                         u64 start, u64 len)
5878 {
5879         return __btrfs_free_reserved_extent(root, start, len, 0);
5880 }
5881
5882 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5883                                        u64 start, u64 len)
5884 {
5885         return __btrfs_free_reserved_extent(root, start, len, 1);
5886 }
5887
5888 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5889                                       struct btrfs_root *root,
5890                                       u64 parent, u64 root_objectid,
5891                                       u64 flags, u64 owner, u64 offset,
5892                                       struct btrfs_key *ins, int ref_mod)
5893 {
5894         int ret;
5895         struct btrfs_fs_info *fs_info = root->fs_info;
5896         struct btrfs_extent_item *extent_item;
5897         struct btrfs_extent_inline_ref *iref;
5898         struct btrfs_path *path;
5899         struct extent_buffer *leaf;
5900         int type;
5901         u32 size;
5902
5903         if (parent > 0)
5904                 type = BTRFS_SHARED_DATA_REF_KEY;
5905         else
5906                 type = BTRFS_EXTENT_DATA_REF_KEY;
5907
5908         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5909
5910         path = btrfs_alloc_path();
5911         if (!path)
5912                 return -ENOMEM;
5913
5914         path->leave_spinning = 1;
5915         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5916                                       ins, size);
5917         BUG_ON(ret);
5918
5919         leaf = path->nodes[0];
5920         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5921                                      struct btrfs_extent_item);
5922         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5923         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5924         btrfs_set_extent_flags(leaf, extent_item,
5925                                flags | BTRFS_EXTENT_FLAG_DATA);
5926
5927         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5928         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5929         if (parent > 0) {
5930                 struct btrfs_shared_data_ref *ref;
5931                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5932                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5933                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5934         } else {
5935                 struct btrfs_extent_data_ref *ref;
5936                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5937                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5938                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5939                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5940                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5941         }
5942
5943         btrfs_mark_buffer_dirty(path->nodes[0]);
5944         btrfs_free_path(path);
5945
5946         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5947         if (ret) {
5948                 printk(KERN_ERR "btrfs update block group failed for %llu "
5949                        "%llu\n", (unsigned long long)ins->objectid,
5950                        (unsigned long long)ins->offset);
5951                 BUG();
5952         }
5953         return ret;
5954 }
5955
5956 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5957                                      struct btrfs_root *root,
5958                                      u64 parent, u64 root_objectid,
5959                                      u64 flags, struct btrfs_disk_key *key,
5960                                      int level, struct btrfs_key *ins)
5961 {
5962         int ret;
5963         struct btrfs_fs_info *fs_info = root->fs_info;
5964         struct btrfs_extent_item *extent_item;
5965         struct btrfs_tree_block_info *block_info;
5966         struct btrfs_extent_inline_ref *iref;
5967         struct btrfs_path *path;
5968         struct extent_buffer *leaf;
5969         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5970
5971         path = btrfs_alloc_path();
5972         if (!path)
5973                 return -ENOMEM;
5974
5975         path->leave_spinning = 1;
5976         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5977                                       ins, size);
5978         BUG_ON(ret);
5979
5980         leaf = path->nodes[0];
5981         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5982                                      struct btrfs_extent_item);
5983         btrfs_set_extent_refs(leaf, extent_item, 1);
5984         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5985         btrfs_set_extent_flags(leaf, extent_item,
5986                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5987         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5988
5989         btrfs_set_tree_block_key(leaf, block_info, key);
5990         btrfs_set_tree_block_level(leaf, block_info, level);
5991
5992         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5993         if (parent > 0) {
5994                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5995                 btrfs_set_extent_inline_ref_type(leaf, iref,
5996                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5997                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5998         } else {
5999                 btrfs_set_extent_inline_ref_type(leaf, iref,
6000                                                  BTRFS_TREE_BLOCK_REF_KEY);
6001                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6002         }
6003
6004         btrfs_mark_buffer_dirty(leaf);
6005         btrfs_free_path(path);
6006
6007         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6008         if (ret) {
6009                 printk(KERN_ERR "btrfs update block group failed for %llu "
6010                        "%llu\n", (unsigned long long)ins->objectid,
6011                        (unsigned long long)ins->offset);
6012                 BUG();
6013         }
6014         return ret;
6015 }
6016
6017 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6018                                      struct btrfs_root *root,
6019                                      u64 root_objectid, u64 owner,
6020                                      u64 offset, struct btrfs_key *ins)
6021 {
6022         int ret;
6023
6024         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6025
6026         ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6027                                          ins->offset, 0,
6028                                          root_objectid, owner, offset,
6029                                          BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6030         return ret;
6031 }
6032
6033 /*
6034  * this is used by the tree logging recovery code.  It records that
6035  * an extent has been allocated and makes sure to clear the free
6036  * space cache bits as well
6037  */
6038 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6039                                    struct btrfs_root *root,
6040                                    u64 root_objectid, u64 owner, u64 offset,
6041                                    struct btrfs_key *ins)
6042 {
6043         int ret;
6044         struct btrfs_block_group_cache *block_group;
6045         struct btrfs_caching_control *caching_ctl;
6046         u64 start = ins->objectid;
6047         u64 num_bytes = ins->offset;
6048
6049         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6050         cache_block_group(block_group, trans, NULL, 0);
6051         caching_ctl = get_caching_control(block_group);
6052
6053         if (!caching_ctl) {
6054                 BUG_ON(!block_group_cache_done(block_group));
6055                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6056                 BUG_ON(ret);
6057         } else {
6058                 mutex_lock(&caching_ctl->mutex);
6059
6060                 if (start >= caching_ctl->progress) {
6061                         ret = add_excluded_extent(root, start, num_bytes);
6062                         BUG_ON(ret);
6063                 } else if (start + num_bytes <= caching_ctl->progress) {
6064                         ret = btrfs_remove_free_space(block_group,
6065                                                       start, num_bytes);
6066                         BUG_ON(ret);
6067                 } else {
6068                         num_bytes = caching_ctl->progress - start;
6069                         ret = btrfs_remove_free_space(block_group,
6070                                                       start, num_bytes);
6071                         BUG_ON(ret);
6072
6073                         start = caching_ctl->progress;
6074                         num_bytes = ins->objectid + ins->offset -
6075                                     caching_ctl->progress;
6076                         ret = add_excluded_extent(root, start, num_bytes);
6077                         BUG_ON(ret);
6078                 }
6079
6080                 mutex_unlock(&caching_ctl->mutex);
6081                 put_caching_control(caching_ctl);
6082         }
6083
6084         ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6085                                           RESERVE_ALLOC_NO_ACCOUNT);
6086         BUG_ON(ret);
6087         btrfs_put_block_group(block_group);
6088         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6089                                          0, owner, offset, ins, 1);
6090         return ret;
6091 }
6092
6093 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6094                                             struct btrfs_root *root,
6095                                             u64 bytenr, u32 blocksize,
6096                                             int level)
6097 {
6098         struct extent_buffer *buf;
6099
6100         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6101         if (!buf)
6102                 return ERR_PTR(-ENOMEM);
6103         btrfs_set_header_generation(buf, trans->transid);
6104         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6105         btrfs_tree_lock(buf);
6106         clean_tree_block(trans, root, buf);
6107
6108         btrfs_set_lock_blocking(buf);
6109         btrfs_set_buffer_uptodate(buf);
6110
6111         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6112                 /*
6113                  * we allow two log transactions at a time, use different
6114                  * EXENT bit to differentiate dirty pages.
6115                  */
6116                 if (root->log_transid % 2 == 0)
6117                         set_extent_dirty(&root->dirty_log_pages, buf->start,
6118                                         buf->start + buf->len - 1, GFP_NOFS);
6119                 else
6120                         set_extent_new(&root->dirty_log_pages, buf->start,
6121                                         buf->start + buf->len - 1, GFP_NOFS);
6122         } else {
6123                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6124                          buf->start + buf->len - 1, GFP_NOFS);
6125         }
6126         trans->blocks_used++;
6127         /* this returns a buffer locked for blocking */
6128         return buf;
6129 }
6130
6131 static struct btrfs_block_rsv *
6132 use_block_rsv(struct btrfs_trans_handle *trans,
6133               struct btrfs_root *root, u32 blocksize)
6134 {
6135         struct btrfs_block_rsv *block_rsv;
6136         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6137         int ret;
6138
6139         block_rsv = get_block_rsv(trans, root);
6140
6141         if (block_rsv->size == 0) {
6142                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6143                 /*
6144                  * If we couldn't reserve metadata bytes try and use some from
6145                  * the global reserve.
6146                  */
6147                 if (ret && block_rsv != global_rsv) {
6148                         ret = block_rsv_use_bytes(global_rsv, blocksize);
6149                         if (!ret)
6150                                 return global_rsv;
6151                         return ERR_PTR(ret);
6152                 } else if (ret) {
6153                         return ERR_PTR(ret);
6154                 }
6155                 return block_rsv;
6156         }
6157
6158         ret = block_rsv_use_bytes(block_rsv, blocksize);
6159         if (!ret)
6160                 return block_rsv;
6161         if (ret) {
6162                 static DEFINE_RATELIMIT_STATE(_rs,
6163                                 DEFAULT_RATELIMIT_INTERVAL,
6164                                 /*DEFAULT_RATELIMIT_BURST*/ 2);
6165                 if (__ratelimit(&_rs)) {
6166                         printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6167                         WARN_ON(1);
6168                 }
6169                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6170                 if (!ret) {
6171                         return block_rsv;
6172                 } else if (ret && block_rsv != global_rsv) {
6173                         ret = block_rsv_use_bytes(global_rsv, blocksize);
6174                         if (!ret)
6175                                 return global_rsv;
6176                 }
6177         }
6178
6179         return ERR_PTR(-ENOSPC);
6180 }
6181
6182 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6183                             struct btrfs_block_rsv *block_rsv, u32 blocksize)
6184 {
6185         block_rsv_add_bytes(block_rsv, blocksize, 0);
6186         block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6187 }
6188
6189 /*
6190  * finds a free extent and does all the dirty work required for allocation
6191  * returns the key for the extent through ins, and a tree buffer for
6192  * the first block of the extent through buf.
6193  *
6194  * returns the tree buffer or NULL.
6195  */
6196 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6197                                         struct btrfs_root *root, u32 blocksize,
6198                                         u64 parent, u64 root_objectid,
6199                                         struct btrfs_disk_key *key, int level,
6200                                         u64 hint, u64 empty_size, int for_cow)
6201 {
6202         struct btrfs_key ins;
6203         struct btrfs_block_rsv *block_rsv;
6204         struct extent_buffer *buf;
6205         u64 flags = 0;
6206         int ret;
6207
6208
6209         block_rsv = use_block_rsv(trans, root, blocksize);
6210         if (IS_ERR(block_rsv))
6211                 return ERR_CAST(block_rsv);
6212
6213         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6214                                    empty_size, hint, (u64)-1, &ins, 0);
6215         if (ret) {
6216                 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6217                 return ERR_PTR(ret);
6218         }
6219
6220         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6221                                     blocksize, level);
6222         BUG_ON(IS_ERR(buf));
6223
6224         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6225                 if (parent == 0)
6226                         parent = ins.objectid;
6227                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6228         } else
6229                 BUG_ON(parent > 0);
6230
6231         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6232                 struct btrfs_delayed_extent_op *extent_op;
6233                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6234                 BUG_ON(!extent_op);
6235                 if (key)
6236                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
6237                 else
6238                         memset(&extent_op->key, 0, sizeof(extent_op->key));
6239                 extent_op->flags_to_set = flags;
6240                 extent_op->update_key = 1;
6241                 extent_op->update_flags = 1;
6242                 extent_op->is_data = 0;
6243
6244                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6245                                         ins.objectid,
6246                                         ins.offset, parent, root_objectid,
6247                                         level, BTRFS_ADD_DELAYED_EXTENT,
6248                                         extent_op, for_cow);
6249                 BUG_ON(ret);
6250         }
6251         return buf;
6252 }
6253
6254 struct walk_control {
6255         u64 refs[BTRFS_MAX_LEVEL];
6256         u64 flags[BTRFS_MAX_LEVEL];
6257         struct btrfs_key update_progress;
6258         int stage;
6259         int level;
6260         int shared_level;
6261         int update_ref;
6262         int keep_locks;
6263         int reada_slot;
6264         int reada_count;
6265         int for_reloc;
6266 };
6267
6268 #define DROP_REFERENCE  1
6269 #define UPDATE_BACKREF  2
6270
6271 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6272                                      struct btrfs_root *root,
6273                                      struct walk_control *wc,
6274                                      struct btrfs_path *path)
6275 {
6276         u64 bytenr;
6277         u64 generation;
6278         u64 refs;
6279         u64 flags;
6280         u32 nritems;
6281         u32 blocksize;
6282         struct btrfs_key key;
6283         struct extent_buffer *eb;
6284         int ret;
6285         int slot;
6286         int nread = 0;
6287
6288         if (path->slots[wc->level] < wc->reada_slot) {
6289                 wc->reada_count = wc->reada_count * 2 / 3;
6290                 wc->reada_count = max(wc->reada_count, 2);
6291         } else {
6292                 wc->reada_count = wc->reada_count * 3 / 2;
6293                 wc->reada_count = min_t(int, wc->reada_count,
6294                                         BTRFS_NODEPTRS_PER_BLOCK(root));
6295         }
6296
6297         eb = path->nodes[wc->level];
6298         nritems = btrfs_header_nritems(eb);
6299         blocksize = btrfs_level_size(root, wc->level - 1);
6300
6301         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6302                 if (nread >= wc->reada_count)
6303                         break;
6304
6305                 cond_resched();
6306                 bytenr = btrfs_node_blockptr(eb, slot);
6307                 generation = btrfs_node_ptr_generation(eb, slot);
6308
6309                 if (slot == path->slots[wc->level])
6310                         goto reada;
6311
6312                 if (wc->stage == UPDATE_BACKREF &&
6313                     generation <= root->root_key.offset)
6314                         continue;
6315
6316                 /* We don't lock the tree block, it's OK to be racy here */
6317                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6318                                                &refs, &flags);
6319                 BUG_ON(ret);
6320                 BUG_ON(refs == 0);
6321
6322                 if (wc->stage == DROP_REFERENCE) {
6323                         if (refs == 1)
6324                                 goto reada;
6325
6326                         if (wc->level == 1 &&
6327                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6328                                 continue;
6329                         if (!wc->update_ref ||
6330                             generation <= root->root_key.offset)
6331                                 continue;
6332                         btrfs_node_key_to_cpu(eb, &key, slot);
6333                         ret = btrfs_comp_cpu_keys(&key,
6334                                                   &wc->update_progress);
6335                         if (ret < 0)
6336                                 continue;
6337                 } else {
6338                         if (wc->level == 1 &&
6339                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6340                                 continue;
6341                 }
6342 reada:
6343                 ret = readahead_tree_block(root, bytenr, blocksize,
6344                                            generation);
6345                 if (ret)
6346                         break;
6347                 nread++;
6348         }
6349         wc->reada_slot = slot;
6350 }
6351
6352 /*
6353  * hepler to process tree block while walking down the tree.
6354  *
6355  * when wc->stage == UPDATE_BACKREF, this function updates
6356  * back refs for pointers in the block.
6357  *
6358  * NOTE: return value 1 means we should stop walking down.
6359  */
6360 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6361                                    struct btrfs_root *root,
6362                                    struct btrfs_path *path,
6363                                    struct walk_control *wc, int lookup_info)
6364 {
6365         int level = wc->level;
6366         struct extent_buffer *eb = path->nodes[level];
6367         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6368         int ret;
6369
6370         if (wc->stage == UPDATE_BACKREF &&
6371             btrfs_header_owner(eb) != root->root_key.objectid)
6372                 return 1;
6373
6374         /*
6375          * when reference count of tree block is 1, it won't increase
6376          * again. once full backref flag is set, we never clear it.
6377          */
6378         if (lookup_info &&
6379             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6380              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6381                 BUG_ON(!path->locks[level]);
6382                 ret = btrfs_lookup_extent_info(trans, root,
6383                                                eb->start, eb->len,
6384                                                &wc->refs[level],
6385                                                &wc->flags[level]);
6386                 BUG_ON(ret);
6387                 BUG_ON(wc->refs[level] == 0);
6388         }
6389
6390         if (wc->stage == DROP_REFERENCE) {
6391                 if (wc->refs[level] > 1)
6392                         return 1;
6393
6394                 if (path->locks[level] && !wc->keep_locks) {
6395                         btrfs_tree_unlock_rw(eb, path->locks[level]);
6396                         path->locks[level] = 0;
6397                 }
6398                 return 0;
6399         }
6400
6401         /* wc->stage == UPDATE_BACKREF */
6402         if (!(wc->flags[level] & flag)) {
6403                 BUG_ON(!path->locks[level]);
6404                 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6405                 BUG_ON(ret);
6406                 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6407                 BUG_ON(ret);
6408                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6409                                                   eb->len, flag, 0);
6410                 BUG_ON(ret);
6411                 wc->flags[level] |= flag;
6412         }
6413
6414         /*
6415          * the block is shared by multiple trees, so it's not good to
6416          * keep the tree lock
6417          */
6418         if (path->locks[level] && level > 0) {
6419                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6420                 path->locks[level] = 0;
6421         }
6422         return 0;
6423 }
6424
6425 /*
6426  * hepler to process tree block pointer.
6427  *
6428  * when wc->stage == DROP_REFERENCE, this function checks
6429  * reference count of the block pointed to. if the block
6430  * is shared and we need update back refs for the subtree
6431  * rooted at the block, this function changes wc->stage to
6432  * UPDATE_BACKREF. if the block is shared and there is no
6433  * need to update back, this function drops the reference
6434  * to the block.
6435  *
6436  * NOTE: return value 1 means we should stop walking down.
6437  */
6438 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6439                                  struct btrfs_root *root,
6440                                  struct btrfs_path *path,
6441                                  struct walk_control *wc, int *lookup_info)
6442 {
6443         u64 bytenr;
6444         u64 generation;
6445         u64 parent;
6446         u32 blocksize;
6447         struct btrfs_key key;
6448         struct extent_buffer *next;
6449         int level = wc->level;
6450         int reada = 0;
6451         int ret = 0;
6452
6453         generation = btrfs_node_ptr_generation(path->nodes[level],
6454                                                path->slots[level]);
6455         /*
6456          * if the lower level block was created before the snapshot
6457          * was created, we know there is no need to update back refs
6458          * for the subtree
6459          */
6460         if (wc->stage == UPDATE_BACKREF &&
6461             generation <= root->root_key.offset) {
6462                 *lookup_info = 1;
6463                 return 1;
6464         }
6465
6466         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6467         blocksize = btrfs_level_size(root, level - 1);
6468
6469         next = btrfs_find_tree_block(root, bytenr, blocksize);
6470         if (!next) {
6471                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6472                 if (!next)
6473                         return -ENOMEM;
6474                 reada = 1;
6475         }
6476         btrfs_tree_lock(next);
6477         btrfs_set_lock_blocking(next);
6478
6479         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6480                                        &wc->refs[level - 1],
6481                                        &wc->flags[level - 1]);
6482         BUG_ON(ret);
6483         BUG_ON(wc->refs[level - 1] == 0);
6484         *lookup_info = 0;
6485
6486         if (wc->stage == DROP_REFERENCE) {
6487                 if (wc->refs[level - 1] > 1) {
6488                         if (level == 1 &&
6489                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6490                                 goto skip;
6491
6492                         if (!wc->update_ref ||
6493                             generation <= root->root_key.offset)
6494                                 goto skip;
6495
6496                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6497                                               path->slots[level]);
6498                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6499                         if (ret < 0)
6500                                 goto skip;
6501
6502                         wc->stage = UPDATE_BACKREF;
6503                         wc->shared_level = level - 1;
6504                 }
6505         } else {
6506                 if (level == 1 &&
6507                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6508                         goto skip;
6509         }
6510
6511         if (!btrfs_buffer_uptodate(next, generation)) {
6512                 btrfs_tree_unlock(next);
6513                 free_extent_buffer(next);
6514                 next = NULL;
6515                 *lookup_info = 1;
6516         }
6517
6518         if (!next) {
6519                 if (reada && level == 1)
6520                         reada_walk_down(trans, root, wc, path);
6521                 next = read_tree_block(root, bytenr, blocksize, generation);
6522                 if (!next)
6523                         return -EIO;
6524                 btrfs_tree_lock(next);
6525                 btrfs_set_lock_blocking(next);
6526         }
6527
6528         level--;
6529         BUG_ON(level != btrfs_header_level(next));
6530         path->nodes[level] = next;
6531         path->slots[level] = 0;
6532         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6533         wc->level = level;
6534         if (wc->level == 1)
6535                 wc->reada_slot = 0;
6536         return 0;
6537 skip:
6538         wc->refs[level - 1] = 0;
6539         wc->flags[level - 1] = 0;
6540         if (wc->stage == DROP_REFERENCE) {
6541                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6542                         parent = path->nodes[level]->start;
6543                 } else {
6544                         BUG_ON(root->root_key.objectid !=
6545                                btrfs_header_owner(path->nodes[level]));
6546                         parent = 0;
6547                 }
6548
6549                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6550                                 root->root_key.objectid, level - 1, 0, 0);
6551                 BUG_ON(ret);
6552         }
6553         btrfs_tree_unlock(next);
6554         free_extent_buffer(next);
6555         *lookup_info = 1;
6556         return 1;
6557 }
6558
6559 /*
6560  * hepler to process tree block while walking up the tree.
6561  *
6562  * when wc->stage == DROP_REFERENCE, this function drops
6563  * reference count on the block.
6564  *
6565  * when wc->stage == UPDATE_BACKREF, this function changes
6566  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6567  * to UPDATE_BACKREF previously while processing the block.
6568  *
6569  * NOTE: return value 1 means we should stop walking up.
6570  */
6571 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6572                                  struct btrfs_root *root,
6573                                  struct btrfs_path *path,
6574                                  struct walk_control *wc)
6575 {
6576         int ret;
6577         int level = wc->level;
6578         struct extent_buffer *eb = path->nodes[level];
6579         u64 parent = 0;
6580
6581         if (wc->stage == UPDATE_BACKREF) {
6582                 BUG_ON(wc->shared_level < level);
6583                 if (level < wc->shared_level)
6584                         goto out;
6585
6586                 ret = find_next_key(path, level + 1, &wc->update_progress);
6587                 if (ret > 0)
6588                         wc->update_ref = 0;
6589
6590                 wc->stage = DROP_REFERENCE;
6591                 wc->shared_level = -1;
6592                 path->slots[level] = 0;
6593
6594                 /*
6595                  * check reference count again if the block isn't locked.
6596                  * we should start walking down the tree again if reference
6597                  * count is one.
6598                  */
6599                 if (!path->locks[level]) {
6600                         BUG_ON(level == 0);
6601                         btrfs_tree_lock(eb);
6602                         btrfs_set_lock_blocking(eb);
6603                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6604
6605                         ret = btrfs_lookup_extent_info(trans, root,
6606                                                        eb->start, eb->len,
6607                                                        &wc->refs[level],
6608                                                        &wc->flags[level]);
6609                         BUG_ON(ret);
6610                         BUG_ON(wc->refs[level] == 0);
6611                         if (wc->refs[level] == 1) {
6612                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6613                                 return 1;
6614                         }
6615                 }
6616         }
6617
6618         /* wc->stage == DROP_REFERENCE */
6619         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6620
6621         if (wc->refs[level] == 1) {
6622                 if (level == 0) {
6623                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6624                                 ret = btrfs_dec_ref(trans, root, eb, 1,
6625                                                     wc->for_reloc);
6626                         else
6627                                 ret = btrfs_dec_ref(trans, root, eb, 0,
6628                                                     wc->for_reloc);
6629                         BUG_ON(ret);
6630                 }
6631                 /* make block locked assertion in clean_tree_block happy */
6632                 if (!path->locks[level] &&
6633                     btrfs_header_generation(eb) == trans->transid) {
6634                         btrfs_tree_lock(eb);
6635                         btrfs_set_lock_blocking(eb);
6636                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6637                 }
6638                 clean_tree_block(trans, root, eb);
6639         }
6640
6641         if (eb == root->node) {
6642                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6643                         parent = eb->start;
6644                 else
6645                         BUG_ON(root->root_key.objectid !=
6646                                btrfs_header_owner(eb));
6647         } else {
6648                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6649                         parent = path->nodes[level + 1]->start;
6650                 else
6651                         BUG_ON(root->root_key.objectid !=
6652                                btrfs_header_owner(path->nodes[level + 1]));
6653         }
6654
6655         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0);
6656 out:
6657         wc->refs[level] = 0;
6658         wc->flags[level] = 0;
6659         return 0;
6660 }
6661
6662 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6663                                    struct btrfs_root *root,
6664                                    struct btrfs_path *path,
6665                                    struct walk_control *wc)
6666 {
6667         int level = wc->level;
6668         int lookup_info = 1;
6669         int ret;
6670
6671         while (level >= 0) {
6672                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6673                 if (ret > 0)
6674                         break;
6675
6676                 if (level == 0)
6677                         break;
6678
6679                 if (path->slots[level] >=
6680                     btrfs_header_nritems(path->nodes[level]))
6681                         break;
6682
6683                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6684                 if (ret > 0) {
6685                         path->slots[level]++;
6686                         continue;
6687                 } else if (ret < 0)
6688                         return ret;
6689                 level = wc->level;
6690         }
6691         return 0;
6692 }
6693
6694 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6695                                  struct btrfs_root *root,
6696                                  struct btrfs_path *path,
6697                                  struct walk_control *wc, int max_level)
6698 {
6699         int level = wc->level;
6700         int ret;
6701
6702         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6703         while (level < max_level && path->nodes[level]) {
6704                 wc->level = level;
6705                 if (path->slots[level] + 1 <
6706                     btrfs_header_nritems(path->nodes[level])) {
6707                         path->slots[level]++;
6708                         return 0;
6709                 } else {
6710                         ret = walk_up_proc(trans, root, path, wc);
6711                         if (ret > 0)
6712                                 return 0;
6713
6714                         if (path->locks[level]) {
6715                                 btrfs_tree_unlock_rw(path->nodes[level],
6716                                                      path->locks[level]);
6717                                 path->locks[level] = 0;
6718                         }
6719                         free_extent_buffer(path->nodes[level]);
6720                         path->nodes[level] = NULL;
6721                         level++;
6722                 }
6723         }
6724         return 1;
6725 }
6726
6727 /*
6728  * drop a subvolume tree.
6729  *
6730  * this function traverses the tree freeing any blocks that only
6731  * referenced by the tree.
6732  *
6733  * when a shared tree block is found. this function decreases its
6734  * reference count by one. if update_ref is true, this function
6735  * also make sure backrefs for the shared block and all lower level
6736  * blocks are properly updated.
6737  */
6738 void btrfs_drop_snapshot(struct btrfs_root *root,
6739                          struct btrfs_block_rsv *block_rsv, int update_ref,
6740                          int for_reloc)
6741 {
6742         struct btrfs_path *path;
6743         struct btrfs_trans_handle *trans;
6744         struct btrfs_root *tree_root = root->fs_info->tree_root;
6745         struct btrfs_root_item *root_item = &root->root_item;
6746         struct walk_control *wc;
6747         struct btrfs_key key;
6748         int err = 0;
6749         int ret;
6750         int level;
6751
6752         path = btrfs_alloc_path();
6753         if (!path) {
6754                 err = -ENOMEM;
6755                 goto out;
6756         }
6757
6758         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6759         if (!wc) {
6760                 btrfs_free_path(path);
6761                 err = -ENOMEM;
6762                 goto out;
6763         }
6764
6765         trans = btrfs_start_transaction(tree_root, 0);
6766         BUG_ON(IS_ERR(trans));
6767
6768         if (block_rsv)
6769                 trans->block_rsv = block_rsv;
6770
6771         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6772                 level = btrfs_header_level(root->node);
6773                 path->nodes[level] = btrfs_lock_root_node(root);
6774                 btrfs_set_lock_blocking(path->nodes[level]);
6775                 path->slots[level] = 0;
6776                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6777                 memset(&wc->update_progress, 0,
6778                        sizeof(wc->update_progress));
6779         } else {
6780                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6781                 memcpy(&wc->update_progress, &key,
6782                        sizeof(wc->update_progress));
6783
6784                 level = root_item->drop_level;
6785                 BUG_ON(level == 0);
6786                 path->lowest_level = level;
6787                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6788                 path->lowest_level = 0;
6789                 if (ret < 0) {
6790                         err = ret;
6791                         goto out_free;
6792                 }
6793                 WARN_ON(ret > 0);
6794
6795                 /*
6796                  * unlock our path, this is safe because only this
6797                  * function is allowed to delete this snapshot
6798                  */
6799                 btrfs_unlock_up_safe(path, 0);
6800
6801                 level = btrfs_header_level(root->node);
6802                 while (1) {
6803                         btrfs_tree_lock(path->nodes[level]);
6804                         btrfs_set_lock_blocking(path->nodes[level]);
6805
6806                         ret = btrfs_lookup_extent_info(trans, root,
6807                                                 path->nodes[level]->start,
6808                                                 path->nodes[level]->len,
6809                                                 &wc->refs[level],
6810                                                 &wc->flags[level]);
6811                         BUG_ON(ret);
6812                         BUG_ON(wc->refs[level] == 0);
6813
6814                         if (level == root_item->drop_level)
6815                                 break;
6816
6817                         btrfs_tree_unlock(path->nodes[level]);
6818                         WARN_ON(wc->refs[level] != 1);
6819                         level--;
6820                 }
6821         }
6822
6823         wc->level = level;
6824         wc->shared_level = -1;
6825         wc->stage = DROP_REFERENCE;
6826         wc->update_ref = update_ref;
6827         wc->keep_locks = 0;
6828         wc->for_reloc = for_reloc;
6829         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6830
6831         while (1) {
6832                 ret = walk_down_tree(trans, root, path, wc);
6833                 if (ret < 0) {
6834                         err = ret;
6835                         break;
6836                 }
6837
6838                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6839                 if (ret < 0) {
6840                         err = ret;
6841                         break;
6842                 }
6843
6844                 if (ret > 0) {
6845                         BUG_ON(wc->stage != DROP_REFERENCE);
6846                         break;
6847                 }
6848
6849                 if (wc->stage == DROP_REFERENCE) {
6850                         level = wc->level;
6851                         btrfs_node_key(path->nodes[level],
6852                                        &root_item->drop_progress,
6853                                        path->slots[level]);
6854                         root_item->drop_level = level;
6855                 }
6856
6857                 BUG_ON(wc->level == 0);
6858                 if (btrfs_should_end_transaction(trans, tree_root)) {
6859                         ret = btrfs_update_root(trans, tree_root,
6860                                                 &root->root_key,
6861                                                 root_item);
6862                         BUG_ON(ret);
6863
6864                         btrfs_end_transaction_throttle(trans, tree_root);
6865                         trans = btrfs_start_transaction(tree_root, 0);
6866                         BUG_ON(IS_ERR(trans));
6867                         if (block_rsv)
6868                                 trans->block_rsv = block_rsv;
6869                 }
6870         }
6871         btrfs_release_path(path);
6872         BUG_ON(err);
6873
6874         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6875         BUG_ON(ret);
6876
6877         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6878                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6879                                            NULL, NULL);
6880                 BUG_ON(ret < 0);
6881                 if (ret > 0) {
6882                         /* if we fail to delete the orphan item this time
6883                          * around, it'll get picked up the next time.
6884                          *
6885                          * The most common failure here is just -ENOENT.
6886                          */
6887                         btrfs_del_orphan_item(trans, tree_root,
6888                                               root->root_key.objectid);
6889                 }
6890         }
6891
6892         if (root->in_radix) {
6893                 btrfs_free_fs_root(tree_root->fs_info, root);
6894         } else {
6895                 free_extent_buffer(root->node);
6896                 free_extent_buffer(root->commit_root);
6897                 kfree(root);
6898         }
6899 out_free:
6900         btrfs_end_transaction_throttle(trans, tree_root);
6901         kfree(wc);
6902         btrfs_free_path(path);
6903 out:
6904         if (err)
6905                 btrfs_std_error(root->fs_info, err);
6906         return;
6907 }
6908
6909 /*
6910  * drop subtree rooted at tree block 'node'.
6911  *
6912  * NOTE: this function will unlock and release tree block 'node'
6913  * only used by relocation code
6914  */
6915 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6916                         struct btrfs_root *root,
6917                         struct extent_buffer *node,
6918                         struct extent_buffer *parent)
6919 {
6920         struct btrfs_path *path;
6921         struct walk_control *wc;
6922         int level;
6923         int parent_level;
6924         int ret = 0;
6925         int wret;
6926
6927         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6928
6929         path = btrfs_alloc_path();
6930         if (!path)
6931                 return -ENOMEM;
6932
6933         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6934         if (!wc) {
6935                 btrfs_free_path(path);
6936                 return -ENOMEM;
6937         }
6938
6939         btrfs_assert_tree_locked(parent);
6940         parent_level = btrfs_header_level(parent);
6941         extent_buffer_get(parent);
6942         path->nodes[parent_level] = parent;
6943         path->slots[parent_level] = btrfs_header_nritems(parent);
6944
6945         btrfs_assert_tree_locked(node);
6946         level = btrfs_header_level(node);
6947         path->nodes[level] = node;
6948         path->slots[level] = 0;
6949         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6950
6951         wc->refs[parent_level] = 1;
6952         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6953         wc->level = level;
6954         wc->shared_level = -1;
6955         wc->stage = DROP_REFERENCE;
6956         wc->update_ref = 0;
6957         wc->keep_locks = 1;
6958         wc->for_reloc = 1;
6959         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6960
6961         while (1) {
6962                 wret = walk_down_tree(trans, root, path, wc);
6963                 if (wret < 0) {
6964                         ret = wret;
6965                         break;
6966                 }
6967
6968                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6969                 if (wret < 0)
6970                         ret = wret;
6971                 if (wret != 0)
6972                         break;
6973         }
6974
6975         kfree(wc);
6976         btrfs_free_path(path);
6977         return ret;
6978 }
6979
6980 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6981 {
6982         u64 num_devices;
6983         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6984                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6985
6986         if (root->fs_info->balance_ctl) {
6987                 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
6988                 u64 tgt = 0;
6989
6990                 /* pick restriper's target profile and return */
6991                 if (flags & BTRFS_BLOCK_GROUP_DATA &&
6992                     bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6993                         tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
6994                 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
6995                            bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6996                         tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
6997                 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
6998                            bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6999                         tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
7000                 }
7001
7002                 if (tgt) {
7003                         /* extended -> chunk profile */
7004                         tgt &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
7005                         return tgt;
7006                 }
7007         }
7008
7009         /*
7010          * we add in the count of missing devices because we want
7011          * to make sure that any RAID levels on a degraded FS
7012          * continue to be honored.
7013          */
7014         num_devices = root->fs_info->fs_devices->rw_devices +
7015                 root->fs_info->fs_devices->missing_devices;
7016
7017         if (num_devices == 1) {
7018                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7019                 stripped = flags & ~stripped;
7020
7021                 /* turn raid0 into single device chunks */
7022                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7023                         return stripped;
7024
7025                 /* turn mirroring into duplication */
7026                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7027                              BTRFS_BLOCK_GROUP_RAID10))
7028                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7029                 return flags;
7030         } else {
7031                 /* they already had raid on here, just return */
7032                 if (flags & stripped)
7033                         return flags;
7034
7035                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7036                 stripped = flags & ~stripped;
7037
7038                 /* switch duplicated blocks with raid1 */
7039                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7040                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7041
7042                 /* turn single device chunks into raid0 */
7043                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7044         }
7045         return flags;
7046 }
7047
7048 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7049 {
7050         struct btrfs_space_info *sinfo = cache->space_info;
7051         u64 num_bytes;
7052         u64 min_allocable_bytes;
7053         int ret = -ENOSPC;
7054
7055
7056         /*
7057          * We need some metadata space and system metadata space for
7058          * allocating chunks in some corner cases until we force to set
7059          * it to be readonly.
7060          */
7061         if ((sinfo->flags &
7062              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7063             !force)
7064                 min_allocable_bytes = 1 * 1024 * 1024;
7065         else
7066                 min_allocable_bytes = 0;
7067
7068         spin_lock(&sinfo->lock);
7069         spin_lock(&cache->lock);
7070
7071         if (cache->ro) {
7072                 ret = 0;
7073                 goto out;
7074         }
7075
7076         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7077                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7078
7079         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7080             sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7081             min_allocable_bytes <= sinfo->total_bytes) {
7082                 sinfo->bytes_readonly += num_bytes;
7083                 cache->ro = 1;
7084                 ret = 0;
7085         }
7086 out:
7087         spin_unlock(&cache->lock);
7088         spin_unlock(&sinfo->lock);
7089         return ret;
7090 }
7091
7092 int btrfs_set_block_group_ro(struct btrfs_root *root,
7093                              struct btrfs_block_group_cache *cache)
7094
7095 {
7096         struct btrfs_trans_handle *trans;
7097         u64 alloc_flags;
7098         int ret;
7099
7100         BUG_ON(cache->ro);
7101
7102         trans = btrfs_join_transaction(root);
7103         BUG_ON(IS_ERR(trans));
7104
7105         alloc_flags = update_block_group_flags(root, cache->flags);
7106         if (alloc_flags != cache->flags)
7107                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7108                                CHUNK_ALLOC_FORCE);
7109
7110         ret = set_block_group_ro(cache, 0);
7111         if (!ret)
7112                 goto out;
7113         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7114         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7115                              CHUNK_ALLOC_FORCE);
7116         if (ret < 0)
7117                 goto out;
7118         ret = set_block_group_ro(cache, 0);
7119 out:
7120         btrfs_end_transaction(trans, root);
7121         return ret;
7122 }
7123
7124 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7125                             struct btrfs_root *root, u64 type)
7126 {
7127         u64 alloc_flags = get_alloc_profile(root, type);
7128         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7129                               CHUNK_ALLOC_FORCE);
7130 }
7131
7132 /*
7133  * helper to account the unused space of all the readonly block group in the
7134  * list. takes mirrors into account.
7135  */
7136 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7137 {
7138         struct btrfs_block_group_cache *block_group;
7139         u64 free_bytes = 0;
7140         int factor;
7141
7142         list_for_each_entry(block_group, groups_list, list) {
7143                 spin_lock(&block_group->lock);
7144
7145                 if (!block_group->ro) {
7146                         spin_unlock(&block_group->lock);
7147                         continue;
7148                 }
7149
7150                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7151                                           BTRFS_BLOCK_GROUP_RAID10 |
7152                                           BTRFS_BLOCK_GROUP_DUP))
7153                         factor = 2;
7154                 else
7155                         factor = 1;
7156
7157                 free_bytes += (block_group->key.offset -
7158                                btrfs_block_group_used(&block_group->item)) *
7159                                factor;
7160
7161                 spin_unlock(&block_group->lock);
7162         }
7163
7164         return free_bytes;
7165 }
7166
7167 /*
7168  * helper to account the unused space of all the readonly block group in the
7169  * space_info. takes mirrors into account.
7170  */
7171 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7172 {
7173         int i;
7174         u64 free_bytes = 0;
7175
7176         spin_lock(&sinfo->lock);
7177
7178         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7179                 if (!list_empty(&sinfo->block_groups[i]))
7180                         free_bytes += __btrfs_get_ro_block_group_free_space(
7181                                                 &sinfo->block_groups[i]);
7182
7183         spin_unlock(&sinfo->lock);
7184
7185         return free_bytes;
7186 }
7187
7188 int btrfs_set_block_group_rw(struct btrfs_root *root,
7189                               struct btrfs_block_group_cache *cache)
7190 {
7191         struct btrfs_space_info *sinfo = cache->space_info;
7192         u64 num_bytes;
7193
7194         BUG_ON(!cache->ro);
7195
7196         spin_lock(&sinfo->lock);
7197         spin_lock(&cache->lock);
7198         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7199                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7200         sinfo->bytes_readonly -= num_bytes;
7201         cache->ro = 0;
7202         spin_unlock(&cache->lock);
7203         spin_unlock(&sinfo->lock);
7204         return 0;
7205 }
7206
7207 /*
7208  * checks to see if its even possible to relocate this block group.
7209  *
7210  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7211  * ok to go ahead and try.
7212  */
7213 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7214 {
7215         struct btrfs_block_group_cache *block_group;
7216         struct btrfs_space_info *space_info;
7217         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7218         struct btrfs_device *device;
7219         u64 min_free;
7220         u64 dev_min = 1;
7221         u64 dev_nr = 0;
7222         int index;
7223         int full = 0;
7224         int ret = 0;
7225
7226         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7227
7228         /* odd, couldn't find the block group, leave it alone */
7229         if (!block_group)
7230                 return -1;
7231
7232         min_free = btrfs_block_group_used(&block_group->item);
7233
7234         /* no bytes used, we're good */
7235         if (!min_free)
7236                 goto out;
7237
7238         space_info = block_group->space_info;
7239         spin_lock(&space_info->lock);
7240
7241         full = space_info->full;
7242
7243         /*
7244          * if this is the last block group we have in this space, we can't
7245          * relocate it unless we're able to allocate a new chunk below.
7246          *
7247          * Otherwise, we need to make sure we have room in the space to handle
7248          * all of the extents from this block group.  If we can, we're good
7249          */
7250         if ((space_info->total_bytes != block_group->key.offset) &&
7251             (space_info->bytes_used + space_info->bytes_reserved +
7252              space_info->bytes_pinned + space_info->bytes_readonly +
7253              min_free < space_info->total_bytes)) {
7254                 spin_unlock(&space_info->lock);
7255                 goto out;
7256         }
7257         spin_unlock(&space_info->lock);
7258
7259         /*
7260          * ok we don't have enough space, but maybe we have free space on our
7261          * devices to allocate new chunks for relocation, so loop through our
7262          * alloc devices and guess if we have enough space.  However, if we
7263          * were marked as full, then we know there aren't enough chunks, and we
7264          * can just return.
7265          */
7266         ret = -1;
7267         if (full)
7268                 goto out;
7269
7270         /*
7271          * index:
7272          *      0: raid10
7273          *      1: raid1
7274          *      2: dup
7275          *      3: raid0
7276          *      4: single
7277          */
7278         index = get_block_group_index(block_group);
7279         if (index == 0) {
7280                 dev_min = 4;
7281                 /* Divide by 2 */
7282                 min_free >>= 1;
7283         } else if (index == 1) {
7284                 dev_min = 2;
7285         } else if (index == 2) {
7286                 /* Multiply by 2 */
7287                 min_free <<= 1;
7288         } else if (index == 3) {
7289                 dev_min = fs_devices->rw_devices;
7290                 do_div(min_free, dev_min);
7291         }
7292
7293         mutex_lock(&root->fs_info->chunk_mutex);
7294         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7295                 u64 dev_offset;
7296
7297                 /*
7298                  * check to make sure we can actually find a chunk with enough
7299                  * space to fit our block group in.
7300                  */
7301                 if (device->total_bytes > device->bytes_used + min_free) {
7302                         ret = find_free_dev_extent(device, min_free,
7303                                                    &dev_offset, NULL);
7304                         if (!ret)
7305                                 dev_nr++;
7306
7307                         if (dev_nr >= dev_min)
7308                                 break;
7309
7310                         ret = -1;
7311                 }
7312         }
7313         mutex_unlock(&root->fs_info->chunk_mutex);
7314 out:
7315         btrfs_put_block_group(block_group);
7316         return ret;
7317 }
7318
7319 static int find_first_block_group(struct btrfs_root *root,
7320                 struct btrfs_path *path, struct btrfs_key *key)
7321 {
7322         int ret = 0;
7323         struct btrfs_key found_key;
7324         struct extent_buffer *leaf;
7325         int slot;
7326
7327         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7328         if (ret < 0)
7329                 goto out;
7330
7331         while (1) {
7332                 slot = path->slots[0];
7333                 leaf = path->nodes[0];
7334                 if (slot >= btrfs_header_nritems(leaf)) {
7335                         ret = btrfs_next_leaf(root, path);
7336                         if (ret == 0)
7337                                 continue;
7338                         if (ret < 0)
7339                                 goto out;
7340                         break;
7341                 }
7342                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7343
7344                 if (found_key.objectid >= key->objectid &&
7345                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7346                         ret = 0;
7347                         goto out;
7348                 }
7349                 path->slots[0]++;
7350         }
7351 out:
7352         return ret;
7353 }
7354
7355 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7356 {
7357         struct btrfs_block_group_cache *block_group;
7358         u64 last = 0;
7359
7360         while (1) {
7361                 struct inode *inode;
7362
7363                 block_group = btrfs_lookup_first_block_group(info, last);
7364                 while (block_group) {
7365                         spin_lock(&block_group->lock);
7366                         if (block_group->iref)
7367                                 break;
7368                         spin_unlock(&block_group->lock);
7369                         block_group = next_block_group(info->tree_root,
7370                                                        block_group);
7371                 }
7372                 if (!block_group) {
7373                         if (last == 0)
7374                                 break;
7375                         last = 0;
7376                         continue;
7377                 }
7378
7379                 inode = block_group->inode;
7380                 block_group->iref = 0;
7381                 block_group->inode = NULL;
7382                 spin_unlock(&block_group->lock);
7383                 iput(inode);
7384                 last = block_group->key.objectid + block_group->key.offset;
7385                 btrfs_put_block_group(block_group);
7386         }
7387 }
7388
7389 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7390 {
7391         struct btrfs_block_group_cache *block_group;
7392         struct btrfs_space_info *space_info;
7393         struct btrfs_caching_control *caching_ctl;
7394         struct rb_node *n;
7395
7396         down_write(&info->extent_commit_sem);
7397         while (!list_empty(&info->caching_block_groups)) {
7398                 caching_ctl = list_entry(info->caching_block_groups.next,
7399                                          struct btrfs_caching_control, list);
7400                 list_del(&caching_ctl->list);
7401                 put_caching_control(caching_ctl);
7402         }
7403         up_write(&info->extent_commit_sem);
7404
7405         spin_lock(&info->block_group_cache_lock);
7406         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7407                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7408                                        cache_node);
7409                 rb_erase(&block_group->cache_node,
7410                          &info->block_group_cache_tree);
7411                 spin_unlock(&info->block_group_cache_lock);
7412
7413                 down_write(&block_group->space_info->groups_sem);
7414                 list_del(&block_group->list);
7415                 up_write(&block_group->space_info->groups_sem);
7416
7417                 if (block_group->cached == BTRFS_CACHE_STARTED)
7418                         wait_block_group_cache_done(block_group);
7419
7420                 /*
7421                  * We haven't cached this block group, which means we could
7422                  * possibly have excluded extents on this block group.
7423                  */
7424                 if (block_group->cached == BTRFS_CACHE_NO)
7425                         free_excluded_extents(info->extent_root, block_group);
7426
7427                 btrfs_remove_free_space_cache(block_group);
7428                 btrfs_put_block_group(block_group);
7429
7430                 spin_lock(&info->block_group_cache_lock);
7431         }
7432         spin_unlock(&info->block_group_cache_lock);
7433
7434         /* now that all the block groups are freed, go through and
7435          * free all the space_info structs.  This is only called during
7436          * the final stages of unmount, and so we know nobody is
7437          * using them.  We call synchronize_rcu() once before we start,
7438          * just to be on the safe side.
7439          */
7440         synchronize_rcu();
7441
7442         release_global_block_rsv(info);
7443
7444         while(!list_empty(&info->space_info)) {
7445                 space_info = list_entry(info->space_info.next,
7446                                         struct btrfs_space_info,
7447                                         list);
7448                 if (space_info->bytes_pinned > 0 ||
7449                     space_info->bytes_reserved > 0 ||
7450                     space_info->bytes_may_use > 0) {
7451                         WARN_ON(1);
7452                         dump_space_info(space_info, 0, 0);
7453                 }
7454                 list_del(&space_info->list);
7455                 kfree(space_info);
7456         }
7457         return 0;
7458 }
7459
7460 static void __link_block_group(struct btrfs_space_info *space_info,
7461                                struct btrfs_block_group_cache *cache)
7462 {
7463         int index = get_block_group_index(cache);
7464
7465         down_write(&space_info->groups_sem);
7466         list_add_tail(&cache->list, &space_info->block_groups[index]);
7467         up_write(&space_info->groups_sem);
7468 }
7469
7470 int btrfs_read_block_groups(struct btrfs_root *root)
7471 {
7472         struct btrfs_path *path;
7473         int ret;
7474         struct btrfs_block_group_cache *cache;
7475         struct btrfs_fs_info *info = root->fs_info;
7476         struct btrfs_space_info *space_info;
7477         struct btrfs_key key;
7478         struct btrfs_key found_key;
7479         struct extent_buffer *leaf;
7480         int need_clear = 0;
7481         u64 cache_gen;
7482
7483         root = info->extent_root;
7484         key.objectid = 0;
7485         key.offset = 0;
7486         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7487         path = btrfs_alloc_path();
7488         if (!path)
7489                 return -ENOMEM;
7490         path->reada = 1;
7491
7492         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7493         if (btrfs_test_opt(root, SPACE_CACHE) &&
7494             btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7495                 need_clear = 1;
7496         if (btrfs_test_opt(root, CLEAR_CACHE))
7497                 need_clear = 1;
7498
7499         while (1) {
7500                 ret = find_first_block_group(root, path, &key);
7501                 if (ret > 0)
7502                         break;
7503                 if (ret != 0)
7504                         goto error;
7505                 leaf = path->nodes[0];
7506                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7507                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7508                 if (!cache) {
7509                         ret = -ENOMEM;
7510                         goto error;
7511                 }
7512                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7513                                                 GFP_NOFS);
7514                 if (!cache->free_space_ctl) {
7515                         kfree(cache);
7516                         ret = -ENOMEM;
7517                         goto error;
7518                 }
7519
7520                 atomic_set(&cache->count, 1);
7521                 spin_lock_init(&cache->lock);
7522                 cache->fs_info = info;
7523                 INIT_LIST_HEAD(&cache->list);
7524                 INIT_LIST_HEAD(&cache->cluster_list);
7525
7526                 if (need_clear)
7527                         cache->disk_cache_state = BTRFS_DC_CLEAR;
7528
7529                 read_extent_buffer(leaf, &cache->item,
7530                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7531                                    sizeof(cache->item));
7532                 memcpy(&cache->key, &found_key, sizeof(found_key));
7533
7534                 key.objectid = found_key.objectid + found_key.offset;
7535                 btrfs_release_path(path);
7536                 cache->flags = btrfs_block_group_flags(&cache->item);
7537                 cache->sectorsize = root->sectorsize;
7538
7539                 btrfs_init_free_space_ctl(cache);
7540
7541                 /*
7542                  * We need to exclude the super stripes now so that the space
7543                  * info has super bytes accounted for, otherwise we'll think
7544                  * we have more space than we actually do.
7545                  */
7546                 exclude_super_stripes(root, cache);
7547
7548                 /*
7549                  * check for two cases, either we are full, and therefore
7550                  * don't need to bother with the caching work since we won't
7551                  * find any space, or we are empty, and we can just add all
7552                  * the space in and be done with it.  This saves us _alot_ of
7553                  * time, particularly in the full case.
7554                  */
7555                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7556                         cache->last_byte_to_unpin = (u64)-1;
7557                         cache->cached = BTRFS_CACHE_FINISHED;
7558                         free_excluded_extents(root, cache);
7559                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7560                         cache->last_byte_to_unpin = (u64)-1;
7561                         cache->cached = BTRFS_CACHE_FINISHED;
7562                         add_new_free_space(cache, root->fs_info,
7563                                            found_key.objectid,
7564                                            found_key.objectid +
7565                                            found_key.offset);
7566                         free_excluded_extents(root, cache);
7567                 }
7568
7569                 ret = update_space_info(info, cache->flags, found_key.offset,
7570                                         btrfs_block_group_used(&cache->item),
7571                                         &space_info);
7572                 BUG_ON(ret);
7573                 cache->space_info = space_info;
7574                 spin_lock(&cache->space_info->lock);
7575                 cache->space_info->bytes_readonly += cache->bytes_super;
7576                 spin_unlock(&cache->space_info->lock);
7577
7578                 __link_block_group(space_info, cache);
7579
7580                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7581                 BUG_ON(ret);
7582
7583                 set_avail_alloc_bits(root->fs_info, cache->flags);
7584                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7585                         set_block_group_ro(cache, 1);
7586         }
7587
7588         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7589                 if (!(get_alloc_profile(root, space_info->flags) &
7590                       (BTRFS_BLOCK_GROUP_RAID10 |
7591                        BTRFS_BLOCK_GROUP_RAID1 |
7592                        BTRFS_BLOCK_GROUP_DUP)))
7593                         continue;
7594                 /*
7595                  * avoid allocating from un-mirrored block group if there are
7596                  * mirrored block groups.
7597                  */
7598                 list_for_each_entry(cache, &space_info->block_groups[3], list)
7599                         set_block_group_ro(cache, 1);
7600                 list_for_each_entry(cache, &space_info->block_groups[4], list)
7601                         set_block_group_ro(cache, 1);
7602         }
7603
7604         init_global_block_rsv(info);
7605         ret = 0;
7606 error:
7607         btrfs_free_path(path);
7608         return ret;
7609 }
7610
7611 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7612                            struct btrfs_root *root, u64 bytes_used,
7613                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7614                            u64 size)
7615 {
7616         int ret;
7617         struct btrfs_root *extent_root;
7618         struct btrfs_block_group_cache *cache;
7619
7620         extent_root = root->fs_info->extent_root;
7621
7622         root->fs_info->last_trans_log_full_commit = trans->transid;
7623
7624         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7625         if (!cache)
7626                 return -ENOMEM;
7627         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7628                                         GFP_NOFS);
7629         if (!cache->free_space_ctl) {
7630                 kfree(cache);
7631                 return -ENOMEM;
7632         }
7633
7634         cache->key.objectid = chunk_offset;
7635         cache->key.offset = size;
7636         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7637         cache->sectorsize = root->sectorsize;
7638         cache->fs_info = root->fs_info;
7639
7640         atomic_set(&cache->count, 1);
7641         spin_lock_init(&cache->lock);
7642         INIT_LIST_HEAD(&cache->list);
7643         INIT_LIST_HEAD(&cache->cluster_list);
7644
7645         btrfs_init_free_space_ctl(cache);
7646
7647         btrfs_set_block_group_used(&cache->item, bytes_used);
7648         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7649         cache->flags = type;
7650         btrfs_set_block_group_flags(&cache->item, type);
7651
7652         cache->last_byte_to_unpin = (u64)-1;
7653         cache->cached = BTRFS_CACHE_FINISHED;
7654         exclude_super_stripes(root, cache);
7655
7656         add_new_free_space(cache, root->fs_info, chunk_offset,
7657                            chunk_offset + size);
7658
7659         free_excluded_extents(root, cache);
7660
7661         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7662                                 &cache->space_info);
7663         BUG_ON(ret);
7664         update_global_block_rsv(root->fs_info);
7665
7666         spin_lock(&cache->space_info->lock);
7667         cache->space_info->bytes_readonly += cache->bytes_super;
7668         spin_unlock(&cache->space_info->lock);
7669
7670         __link_block_group(cache->space_info, cache);
7671
7672         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7673         BUG_ON(ret);
7674
7675         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7676                                 sizeof(cache->item));
7677         BUG_ON(ret);
7678
7679         set_avail_alloc_bits(extent_root->fs_info, type);
7680
7681         return 0;
7682 }
7683
7684 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7685 {
7686         u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
7687
7688         /* chunk -> extended profile */
7689         if (extra_flags == 0)
7690                 extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
7691
7692         if (flags & BTRFS_BLOCK_GROUP_DATA)
7693                 fs_info->avail_data_alloc_bits &= ~extra_flags;
7694         if (flags & BTRFS_BLOCK_GROUP_METADATA)
7695                 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7696         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7697                 fs_info->avail_system_alloc_bits &= ~extra_flags;
7698 }
7699
7700 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7701                              struct btrfs_root *root, u64 group_start)
7702 {
7703         struct btrfs_path *path;
7704         struct btrfs_block_group_cache *block_group;
7705         struct btrfs_free_cluster *cluster;
7706         struct btrfs_root *tree_root = root->fs_info->tree_root;
7707         struct btrfs_key key;
7708         struct inode *inode;
7709         int ret;
7710         int index;
7711         int factor;
7712
7713         root = root->fs_info->extent_root;
7714
7715         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7716         BUG_ON(!block_group);
7717         BUG_ON(!block_group->ro);
7718
7719         /*
7720          * Free the reserved super bytes from this block group before
7721          * remove it.
7722          */
7723         free_excluded_extents(root, block_group);
7724
7725         memcpy(&key, &block_group->key, sizeof(key));
7726         index = get_block_group_index(block_group);
7727         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7728                                   BTRFS_BLOCK_GROUP_RAID1 |
7729                                   BTRFS_BLOCK_GROUP_RAID10))
7730                 factor = 2;
7731         else
7732                 factor = 1;
7733
7734         /* make sure this block group isn't part of an allocation cluster */
7735         cluster = &root->fs_info->data_alloc_cluster;
7736         spin_lock(&cluster->refill_lock);
7737         btrfs_return_cluster_to_free_space(block_group, cluster);
7738         spin_unlock(&cluster->refill_lock);
7739
7740         /*
7741          * make sure this block group isn't part of a metadata
7742          * allocation cluster
7743          */
7744         cluster = &root->fs_info->meta_alloc_cluster;
7745         spin_lock(&cluster->refill_lock);
7746         btrfs_return_cluster_to_free_space(block_group, cluster);
7747         spin_unlock(&cluster->refill_lock);
7748
7749         path = btrfs_alloc_path();
7750         if (!path) {
7751                 ret = -ENOMEM;
7752                 goto out;
7753         }
7754
7755         inode = lookup_free_space_inode(tree_root, block_group, path);
7756         if (!IS_ERR(inode)) {
7757                 ret = btrfs_orphan_add(trans, inode);
7758                 BUG_ON(ret);
7759                 clear_nlink(inode);
7760                 /* One for the block groups ref */
7761                 spin_lock(&block_group->lock);
7762                 if (block_group->iref) {
7763                         block_group->iref = 0;
7764                         block_group->inode = NULL;
7765                         spin_unlock(&block_group->lock);
7766                         iput(inode);
7767                 } else {
7768                         spin_unlock(&block_group->lock);
7769                 }
7770                 /* One for our lookup ref */
7771                 btrfs_add_delayed_iput(inode);
7772         }
7773
7774         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7775         key.offset = block_group->key.objectid;
7776         key.type = 0;
7777
7778         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7779         if (ret < 0)
7780                 goto out;
7781         if (ret > 0)
7782                 btrfs_release_path(path);
7783         if (ret == 0) {
7784                 ret = btrfs_del_item(trans, tree_root, path);
7785                 if (ret)
7786                         goto out;
7787                 btrfs_release_path(path);
7788         }
7789
7790         spin_lock(&root->fs_info->block_group_cache_lock);
7791         rb_erase(&block_group->cache_node,
7792                  &root->fs_info->block_group_cache_tree);
7793         spin_unlock(&root->fs_info->block_group_cache_lock);
7794
7795         down_write(&block_group->space_info->groups_sem);
7796         /*
7797          * we must use list_del_init so people can check to see if they
7798          * are still on the list after taking the semaphore
7799          */
7800         list_del_init(&block_group->list);
7801         if (list_empty(&block_group->space_info->block_groups[index]))
7802                 clear_avail_alloc_bits(root->fs_info, block_group->flags);
7803         up_write(&block_group->space_info->groups_sem);
7804
7805         if (block_group->cached == BTRFS_CACHE_STARTED)
7806                 wait_block_group_cache_done(block_group);
7807
7808         btrfs_remove_free_space_cache(block_group);
7809
7810         spin_lock(&block_group->space_info->lock);
7811         block_group->space_info->total_bytes -= block_group->key.offset;
7812         block_group->space_info->bytes_readonly -= block_group->key.offset;
7813         block_group->space_info->disk_total -= block_group->key.offset * factor;
7814         spin_unlock(&block_group->space_info->lock);
7815
7816         memcpy(&key, &block_group->key, sizeof(key));
7817
7818         btrfs_clear_space_info_full(root->fs_info);
7819
7820         btrfs_put_block_group(block_group);
7821         btrfs_put_block_group(block_group);
7822
7823         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7824         if (ret > 0)
7825                 ret = -EIO;
7826         if (ret < 0)
7827                 goto out;
7828
7829         ret = btrfs_del_item(trans, root, path);
7830 out:
7831         btrfs_free_path(path);
7832         return ret;
7833 }
7834
7835 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7836 {
7837         struct btrfs_space_info *space_info;
7838         struct btrfs_super_block *disk_super;
7839         u64 features;
7840         u64 flags;
7841         int mixed = 0;
7842         int ret;
7843
7844         disk_super = fs_info->super_copy;
7845         if (!btrfs_super_root(disk_super))
7846                 return 1;
7847
7848         features = btrfs_super_incompat_flags(disk_super);
7849         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7850                 mixed = 1;
7851
7852         flags = BTRFS_BLOCK_GROUP_SYSTEM;
7853         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7854         if (ret)
7855                 goto out;
7856
7857         if (mixed) {
7858                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7859                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7860         } else {
7861                 flags = BTRFS_BLOCK_GROUP_METADATA;
7862                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7863                 if (ret)
7864                         goto out;
7865
7866                 flags = BTRFS_BLOCK_GROUP_DATA;
7867                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7868         }
7869 out:
7870         return ret;
7871 }
7872
7873 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7874 {
7875         return unpin_extent_range(root, start, end);
7876 }
7877
7878 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7879                                u64 num_bytes, u64 *actual_bytes)
7880 {
7881         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7882 }
7883
7884 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7885 {
7886         struct btrfs_fs_info *fs_info = root->fs_info;
7887         struct btrfs_block_group_cache *cache = NULL;
7888         u64 group_trimmed;
7889         u64 start;
7890         u64 end;
7891         u64 trimmed = 0;
7892         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
7893         int ret = 0;
7894
7895         /*
7896          * try to trim all FS space, our block group may start from non-zero.
7897          */
7898         if (range->len == total_bytes)
7899                 cache = btrfs_lookup_first_block_group(fs_info, range->start);
7900         else
7901                 cache = btrfs_lookup_block_group(fs_info, range->start);
7902
7903         while (cache) {
7904                 if (cache->key.objectid >= (range->start + range->len)) {
7905                         btrfs_put_block_group(cache);
7906                         break;
7907                 }
7908
7909                 start = max(range->start, cache->key.objectid);
7910                 end = min(range->start + range->len,
7911                                 cache->key.objectid + cache->key.offset);
7912
7913                 if (end - start >= range->minlen) {
7914                         if (!block_group_cache_done(cache)) {
7915                                 ret = cache_block_group(cache, NULL, root, 0);
7916                                 if (!ret)
7917                                         wait_block_group_cache_done(cache);
7918                         }
7919                         ret = btrfs_trim_block_group(cache,
7920                                                      &group_trimmed,
7921                                                      start,
7922                                                      end,
7923                                                      range->minlen);
7924
7925                         trimmed += group_trimmed;
7926                         if (ret) {
7927                                 btrfs_put_block_group(cache);
7928                                 break;
7929                         }
7930                 }
7931
7932                 cache = next_block_group(fs_info->tree_root, cache);
7933         }
7934
7935         range->len = trimmed;
7936         return ret;
7937 }