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); /* -ENOMEM */
 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); /* -ENOMEM */
 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); /* -ENOMEM */
 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); /* -ENOMEM or logic error */
 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); /* -ENOMEM or logic error */
 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         if (!caching_ctl)
 478                 return -ENOMEM;
 479
 480         INIT_LIST_HEAD(&caching_ctl->list);
 481         mutex_init(&caching_ctl->mutex);
 482         init_waitqueue_head(&caching_ctl->wait);
 483         caching_ctl->block_group = cache;
 484         caching_ctl->progress = cache->key.objectid;
 485         atomic_set(&caching_ctl->count, 1);
 486         caching_ctl->work.func = caching_thread;
 487
 488         spin_lock(&cache->lock);
 489         /*
 490          * This should be a rare occasion, but this could happen I think in the
 491          * case where one thread starts to load the space cache info, and then
 492          * some other thread starts a transaction commit which tries to do an
 493          * allocation while the other thread is still loading the space cache
 494          * info.  The previous loop should have kept us from choosing this block
 495          * group, but if we've moved to the state where we will wait on caching
 496          * block groups we need to first check if we're doing a fast load here,
 497          * so we can wait for it to finish, otherwise we could end up allocating
 498          * from a block group who's cache gets evicted for one reason or
 499          * another.
 500          */
 501         while (cache->cached == BTRFS_CACHE_FAST) {
 502                 struct btrfs_caching_control *ctl;
 503
 504                 ctl = cache->caching_ctl;
 505                 atomic_inc(&ctl->count);
 506                 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
 507                 spin_unlock(&cache->lock);
 508
 509                 schedule();
 510
 511                 finish_wait(&ctl->wait, &wait);
 512                 put_caching_control(ctl);
 513                 spin_lock(&cache->lock);
 514         }
 515
 516         if (cache->cached != BTRFS_CACHE_NO) {
 517                 spin_unlock(&cache->lock);
 518                 kfree(caching_ctl);
 519                 return 0;
 520         }
 521         WARN_ON(cache->caching_ctl);
 522         cache->caching_ctl = caching_ctl;
 523         cache->cached = BTRFS_CACHE_FAST;
 524         spin_unlock(&cache->lock);
 525
 526         /*
 527          * We can't do the read from on-disk cache during a commit since we need
 528          * to have the normal tree locking.  Also if we are currently trying to
 529          * allocate blocks for the tree root we can't do the fast caching since
 530          * we likely hold important locks.
 531          */
 532         if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
 533                 ret = load_free_space_cache(fs_info, cache);
 534
 535                 spin_lock(&cache->lock);
 536                 if (ret == 1) {
 537                         cache->caching_ctl = NULL;
 538                         cache->cached = BTRFS_CACHE_FINISHED;
 539                         cache->last_byte_to_unpin = (u64)-1;
 540                 } else {
 541                         if (load_cache_only) {
 542                                 cache->caching_ctl = NULL;
 543                                 cache->cached = BTRFS_CACHE_NO;
 544                         } else {
 545                                 cache->cached = BTRFS_CACHE_STARTED;
 546                         }
 547                 }
 548                 spin_unlock(&cache->lock);
 549                 wake_up(&caching_ctl->wait);
 550                 if (ret == 1) {
 551                         put_caching_control(caching_ctl);
 552                         free_excluded_extents(fs_info->extent_root, cache);
 553                         return 0;
 554                 }
 555         } else {
 556                 /*
 557                  * We are not going to do the fast caching, set cached to the
 558                  * appropriate value and wakeup any waiters.
 559                  */
 560                 spin_lock(&cache->lock);
 561                 if (load_cache_only) {
 562                         cache->caching_ctl = NULL;
 563                         cache->cached = BTRFS_CACHE_NO;
 564                 } else {
 565                         cache->cached = BTRFS_CACHE_STARTED;
 566                 }
 567                 spin_unlock(&cache->lock);
 568                 wake_up(&caching_ctl->wait);
 569         }
 570
 571         if (load_cache_only) {
 572                 put_caching_control(caching_ctl);
 573                 return 0;
 574         }
 575
 576         down_write(&fs_info->extent_commit_sem);
 577         atomic_inc(&caching_ctl->count);
 578         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 579         up_write(&fs_info->extent_commit_sem);
 580
 581         btrfs_get_block_group(cache);
 582
 583         btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
 584
 585         return ret;
 586 }
 587
 588 /*
 589  * return the block group that starts at or after bytenr
 590  */
 591 static struct btrfs_block_group_cache *
 592 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 593 {
 594         struct btrfs_block_group_cache *cache;
 595
 596         cache = block_group_cache_tree_search(info, bytenr, 0);
 597
 598         return cache;
 599 }
 600
 601 /*
 602  * return the block group that contains the given bytenr
 603  */
 604 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 605                                                  struct btrfs_fs_info *info,
 606                                                  u64 bytenr)
 607 {
 608         struct btrfs_block_group_cache *cache;
 609
 610         cache = block_group_cache_tree_search(info, bytenr, 1);
 611
 612         return cache;
 613 }
 614
 615 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 616                                                   u64 flags)
 617 {
 618         struct list_head *head = &info->space_info;
 619         struct btrfs_space_info *found;
 620
 621         flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
 622
 623         rcu_read_lock();
 624         list_for_each_entry_rcu(found, head, list) {
 625                 if (found->flags & flags) {
 626                         rcu_read_unlock();
 627                         return found;
 628                 }
 629         }
 630         rcu_read_unlock();
 631         return NULL;
 632 }
 633
 634 /*
 635  * after adding space to the filesystem, we need to clear the full flags
 636  * on all the space infos.
 637  */
 638 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 639 {
 640         struct list_head *head = &info->space_info;
 641         struct btrfs_space_info *found;
 642
 643         rcu_read_lock();
 644         list_for_each_entry_rcu(found, head, list)
 645                 found->full = 0;
 646         rcu_read_unlock();
 647 }
 648
 649 static u64 div_factor(u64 num, int factor)
 650 {
 651         if (factor == 10)
 652                 return num;
 653         num *= factor;
 654         do_div(num, 10);
 655         return num;
 656 }
 657
 658 static u64 div_factor_fine(u64 num, int factor)
 659 {
 660         if (factor == 100)
 661                 return num;
 662         num *= factor;
 663         do_div(num, 100);
 664         return num;
 665 }
 666
 667 u64 btrfs_find_block_group(struct btrfs_root *root,
 668                            u64 search_start, u64 search_hint, int owner)
 669 {
 670         struct btrfs_block_group_cache *cache;
 671         u64 used;
 672         u64 last = max(search_hint, search_start);
 673         u64 group_start = 0;
 674         int full_search = 0;
 675         int factor = 9;
 676         int wrapped = 0;
 677 again:
 678         while (1) {
 679                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 680                 if (!cache)
 681                         break;
 682
 683                 spin_lock(&cache->lock);
 684                 last = cache->key.objectid + cache->key.offset;
 685                 used = btrfs_block_group_used(&cache->item);
 686
 687                 if ((full_search || !cache->ro) &&
 688                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 689                         if (used + cache->pinned + cache->reserved <
 690                             div_factor(cache->key.offset, factor)) {
 691                                 group_start = cache->key.objectid;
 692                                 spin_unlock(&cache->lock);
 693                                 btrfs_put_block_group(cache);
 694                                 goto found;
 695                         }
 696                 }
 697                 spin_unlock(&cache->lock);
 698                 btrfs_put_block_group(cache);
 699                 cond_resched();
 700         }
 701         if (!wrapped) {
 702                 last = search_start;
 703                 wrapped = 1;
 704                 goto again;
 705         }
 706         if (!full_search && factor < 10) {
 707                 last = search_start;
 708                 full_search = 1;
 709                 factor = 10;
 710                 goto again;
 711         }
 712 found:
 713         return group_start;
 714 }
 715
 716 /* simple helper to search for an existing extent at a given offset */
 717 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 718 {
 719         int ret;
 720         struct btrfs_key key;
 721         struct btrfs_path *path;
 722
 723         path = btrfs_alloc_path();
 724         if (!path)
 725                 return -ENOMEM;
 726
 727         key.objectid = start;
 728         key.offset = len;
 729         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 730         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 731                                 0, 0);
 732         btrfs_free_path(path);
 733         return ret;
 734 }
 735
 736 /*
 737  * helper function to lookup reference count and flags of extent.
 738  *
 739  * the head node for delayed ref is used to store the sum of all the
 740  * reference count modifications queued up in the rbtree. the head
 741  * node may also store the extent flags to set. This way you can check
 742  * to see what the reference count and extent flags would be if all of
 743  * the delayed refs are not processed.
 744  */
 745 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 746                              struct btrfs_root *root, u64 bytenr,
 747                              u64 num_bytes, u64 *refs, u64 *flags)
 748 {
 749         struct btrfs_delayed_ref_head *head;
 750         struct btrfs_delayed_ref_root *delayed_refs;
 751         struct btrfs_path *path;
 752         struct btrfs_extent_item *ei;
 753         struct extent_buffer *leaf;
 754         struct btrfs_key key;
 755         u32 item_size;
 756         u64 num_refs;
 757         u64 extent_flags;
 758         int ret;
 759
 760         path = btrfs_alloc_path();
 761         if (!path)
 762                 return -ENOMEM;
 763
 764         key.objectid = bytenr;
 765         key.type = BTRFS_EXTENT_ITEM_KEY;
 766         key.offset = num_bytes;
 767         if (!trans) {
 768                 path->skip_locking = 1;
 769                 path->search_commit_root = 1;
 770         }
 771 again:
 772         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 773                                 &key, path, 0, 0);
 774         if (ret < 0)
 775                 goto out_free;
 776
 777         if (ret == 0) {
 778                 leaf = path->nodes[0];
 779                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 780                 if (item_size >= sizeof(*ei)) {
 781                         ei = btrfs_item_ptr(leaf, path->slots[0],
 782                                             struct btrfs_extent_item);
 783                         num_refs = btrfs_extent_refs(leaf, ei);
 784                         extent_flags = btrfs_extent_flags(leaf, ei);
 785                 } else {
 786 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 787                         struct btrfs_extent_item_v0 *ei0;
 788                         BUG_ON(item_size != sizeof(*ei0));
 789                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 790                                              struct btrfs_extent_item_v0);
 791                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 792                         /* FIXME: this isn't correct for data */
 793                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 794 #else
 795                         BUG();
 796 #endif
 797                 }
 798                 BUG_ON(num_refs == 0);
 799         } else {
 800                 num_refs = 0;
 801                 extent_flags = 0;
 802                 ret = 0;
 803         }
 804
 805         if (!trans)
 806                 goto out;
 807
 808         delayed_refs = &trans->transaction->delayed_refs;
 809         spin_lock(&delayed_refs->lock);
 810         head = btrfs_find_delayed_ref_head(trans, bytenr);
 811         if (head) {
 812                 if (!mutex_trylock(&head->mutex)) {
 813                         atomic_inc(&head->node.refs);
 814                         spin_unlock(&delayed_refs->lock);
 815
 816                         btrfs_release_path(path);
 817
 818                         /*
 819                          * Mutex was contended, block until it's released and try
 820                          * again
 821                          */
 822                         mutex_lock(&head->mutex);
 823                         mutex_unlock(&head->mutex);
 824                         btrfs_put_delayed_ref(&head->node);
 825                         goto again;
 826                 }
 827                 if (head->extent_op && head->extent_op->update_flags)
 828                         extent_flags |= head->extent_op->flags_to_set;
 829                 else
 830                         BUG_ON(num_refs == 0);
 831
 832                 num_refs += head->node.ref_mod;
 833                 mutex_unlock(&head->mutex);
 834         }
 835         spin_unlock(&delayed_refs->lock);
 836 out:
 837         WARN_ON(num_refs == 0);
 838         if (refs)
 839                 *refs = num_refs;
 840         if (flags)
 841                 *flags = extent_flags;
 842 out_free:
 843         btrfs_free_path(path);
 844         return ret;
 845 }
 846
 847 /*
 848  * Back reference rules.  Back refs have three main goals:
 849  *
 850  * 1) differentiate between all holders of references to an extent so that
 851  *    when a reference is dropped we can make sure it was a valid reference
 852  *    before freeing the extent.
 853  *
 854  * 2) Provide enough information to quickly find the holders of an extent
 855  *    if we notice a given block is corrupted or bad.
 856  *
 857  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 858  *    maintenance.  This is actually the same as #2, but with a slightly
 859  *    different use case.
 860  *
 861  * There are two kinds of back refs. The implicit back refs is optimized
 862  * for pointers in non-shared tree blocks. For a given pointer in a block,
 863  * back refs of this kind provide information about the block's owner tree
 864  * and the pointer's key. These information allow us to find the block by
 865  * b-tree searching. The full back refs is for pointers in tree blocks not
 866  * referenced by their owner trees. The location of tree block is recorded
 867  * in the back refs. Actually the full back refs is generic, and can be
 868  * used in all cases the implicit back refs is used. The major shortcoming
 869  * of the full back refs is its overhead. Every time a tree block gets
 870  * COWed, we have to update back refs entry for all pointers in it.
 871  *
 872  * For a newly allocated tree block, we use implicit back refs for
 873  * pointers in it. This means most tree related operations only involve
 874  * implicit back refs. For a tree block created in old transaction, the
 875  * only way to drop a reference to it is COW it. So we can detect the
 876  * event that tree block loses its owner tree's reference and do the
 877  * back refs conversion.
 878  *
 879  * When a tree block is COW'd through a tree, there are four cases:
 880  *
 881  * The reference count of the block is one and the tree is the block's
 882  * owner tree. Nothing to do in this case.
 883  *
 884  * The reference count of the block is one and the tree is not the
 885  * block's owner tree. In this case, full back refs is used for pointers
 886  * in the block. Remove these full back refs, add implicit back refs for
 887  * every pointers in the new block.
 888  *
 889  * The reference count of the block is greater than one and the tree is
 890  * the block's owner tree. In this case, implicit back refs is used for
 891  * pointers in the block. Add full back refs for every pointers in the
 892  * block, increase lower level extents' reference counts. The original
 893  * implicit back refs are entailed to the new block.
 894  *
 895  * The reference count of the block is greater than one and the tree is
 896  * not the block's owner tree. Add implicit back refs for every pointer in
 897  * the new block, increase lower level extents' reference count.
 898  *
 899  * Back Reference Key composing:
 900  *
 901  * The key objectid corresponds to the first byte in the extent,
 902  * The key type is used to differentiate between types of back refs.
 903  * There are different meanings of the key offset for different types
 904  * of back refs.
 905  *
 906  * File extents can be referenced by:
 907  *
 908  * - multiple snapshots, subvolumes, or different generations in one subvol
 909  * - different files inside a single subvolume
 910  * - different offsets inside a file (bookend extents in file.c)
 911  *
 912  * The extent ref structure for the implicit back refs has fields for:
 913  *
 914  * - Objectid of the subvolume root
 915  * - objectid of the file holding the reference
 916  * - original offset in the file
 917  * - how many bookend extents
 918  *
 919  * The key offset for the implicit back refs is hash of the first
 920  * three fields.
 921  *
 922  * The extent ref structure for the full back refs has field for:
 923  *
 924  * - number of pointers in the tree leaf
 925  *
 926  * The key offset for the implicit back refs is the first byte of
 927  * the tree leaf
 928  *
 929  * When a file extent is allocated, The implicit back refs is used.
 930  * the fields are filled in:
 931  *
 932  *     (root_key.objectid, inode objectid, offset in file, 1)
 933  *
 934  * When a file extent is removed file truncation, we find the
 935  * corresponding implicit back refs and check the following fields:
 936  *
 937  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 938  *
 939  * Btree extents can be referenced by:
 940  *
 941  * - Different subvolumes
 942  *
 943  * Both the implicit back refs and the full back refs for tree blocks
 944  * only consist of key. The key offset for the implicit back refs is
 945  * objectid of block's owner tree. The key offset for the full back refs
 946  * is the first byte of parent block.
 947  *
 948  * When implicit back refs is used, information about the lowest key and
 949  * level of the tree block are required. These information are stored in
 950  * tree block info structure.
 951  */
 952
 953 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 954 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 955                                   struct btrfs_root *root,
 956                                   struct btrfs_path *path,
 957                                   u64 owner, u32 extra_size)
 958 {
 959         struct btrfs_extent_item *item;
 960         struct btrfs_extent_item_v0 *ei0;
 961         struct btrfs_extent_ref_v0 *ref0;
 962         struct btrfs_tree_block_info *bi;
 963         struct extent_buffer *leaf;
 964         struct btrfs_key key;
 965         struct btrfs_key found_key;
 966         u32 new_size = sizeof(*item);
 967         u64 refs;
 968         int ret;
 969
 970         leaf = path->nodes[0];
 971         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 972
 973         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 974         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 975                              struct btrfs_extent_item_v0);
 976         refs = btrfs_extent_refs_v0(leaf, ei0);
 977
 978         if (owner == (u64)-1) {
 979                 while (1) {
 980                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 981                                 ret = btrfs_next_leaf(root, path);
 982                                 if (ret < 0)
 983                                         return ret;
 984                                 BUG_ON(ret > 0); /* Corruption */
 985                                 leaf = path->nodes[0];
 986                         }
 987                         btrfs_item_key_to_cpu(leaf, &found_key,
 988                                               path->slots[0]);
 989                         BUG_ON(key.objectid != found_key.objectid);
 990                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 991                                 path->slots[0]++;
 992                                 continue;
 993                         }
 994                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 995                                               struct btrfs_extent_ref_v0);
 996                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 997                         break;
 998                 }
 999         }
1000         btrfs_release_path(path);
1001
1002         if (owner < BTRFS_FIRST_FREE_OBJECTID)
1003                 new_size += sizeof(*bi);
1004
1005         new_size -= sizeof(*ei0);
1006         ret = btrfs_search_slot(trans, root, &key, path,
1007                                 new_size + extra_size, 1);
1008         if (ret < 0)
1009                 return ret;
1010         BUG_ON(ret); /* Corruption */
1011
1012         btrfs_extend_item(trans, root, path, new_size);
1013
1014         leaf = path->nodes[0];
1015         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1016         btrfs_set_extent_refs(leaf, item, refs);
1017         /* FIXME: get real generation */
1018         btrfs_set_extent_generation(leaf, item, 0);
1019         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1020                 btrfs_set_extent_flags(leaf, item,
1021                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
1022                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
1023                 bi = (struct btrfs_tree_block_info *)(item + 1);
1024                 /* FIXME: get first key of the block */
1025                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1026                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1027         } else {
1028                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1029         }
1030         btrfs_mark_buffer_dirty(leaf);
1031         return 0;
1032 }
1033 #endif
1034
1035 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1036 {
1037         u32 high_crc = ~(u32)0;
1038         u32 low_crc = ~(u32)0;
1039         __le64 lenum;
1040
1041         lenum = cpu_to_le64(root_objectid);
1042         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1043         lenum = cpu_to_le64(owner);
1044         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1045         lenum = cpu_to_le64(offset);
1046         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1047
1048         return ((u64)high_crc << 31) ^ (u64)low_crc;
1049 }
1050
1051 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1052                                      struct btrfs_extent_data_ref *ref)
1053 {
1054         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1055                                     btrfs_extent_data_ref_objectid(leaf, ref),
1056                                     btrfs_extent_data_ref_offset(leaf, ref));
1057 }
1058
1059 static int match_extent_data_ref(struct extent_buffer *leaf,
1060                                  struct btrfs_extent_data_ref *ref,
1061                                  u64 root_objectid, u64 owner, u64 offset)
1062 {
1063         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1064             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1065             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1066                 return 0;
1067         return 1;
1068 }
1069
1070 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1071                                            struct btrfs_root *root,
1072                                            struct btrfs_path *path,
1073                                            u64 bytenr, u64 parent,
1074                                            u64 root_objectid,
1075                                            u64 owner, u64 offset)
1076 {
1077         struct btrfs_key key;
1078         struct btrfs_extent_data_ref *ref;
1079         struct extent_buffer *leaf;
1080         u32 nritems;
1081         int ret;
1082         int recow;
1083         int err = -ENOENT;
1084
1085         key.objectid = bytenr;
1086         if (parent) {
1087                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1088                 key.offset = parent;
1089         } else {
1090                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1091                 key.offset = hash_extent_data_ref(root_objectid,
1092                                                   owner, offset);
1093         }
1094 again:
1095         recow = 0;
1096         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1097         if (ret < 0) {
1098                 err = ret;
1099                 goto fail;
1100         }
1101
1102         if (parent) {
1103                 if (!ret)
1104                         return 0;
1105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1106                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1107                 btrfs_release_path(path);
1108                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1109                 if (ret < 0) {
1110                         err = ret;
1111                         goto fail;
1112                 }
1113                 if (!ret)
1114                         return 0;
1115 #endif
1116                 goto fail;
1117         }
1118
1119         leaf = path->nodes[0];
1120         nritems = btrfs_header_nritems(leaf);
1121         while (1) {
1122                 if (path->slots[0] >= nritems) {
1123                         ret = btrfs_next_leaf(root, path);
1124                         if (ret < 0)
1125                                 err = ret;
1126                         if (ret)
1127                                 goto fail;
1128
1129                         leaf = path->nodes[0];
1130                         nritems = btrfs_header_nritems(leaf);
1131                         recow = 1;
1132                 }
1133
1134                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1135                 if (key.objectid != bytenr ||
1136                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1137                         goto fail;
1138
1139                 ref = btrfs_item_ptr(leaf, path->slots[0],
1140                                      struct btrfs_extent_data_ref);
1141
1142                 if (match_extent_data_ref(leaf, ref, root_objectid,
1143                                           owner, offset)) {
1144                         if (recow) {
1145                                 btrfs_release_path(path);
1146                                 goto again;
1147                         }
1148                         err = 0;
1149                         break;
1150                 }
1151                 path->slots[0]++;
1152         }
1153 fail:
1154         return err;
1155 }
1156
1157 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1158                                            struct btrfs_root *root,
1159                                            struct btrfs_path *path,
1160                                            u64 bytenr, u64 parent,
1161                                            u64 root_objectid, u64 owner,
1162                                            u64 offset, int refs_to_add)
1163 {
1164         struct btrfs_key key;
1165         struct extent_buffer *leaf;
1166         u32 size;
1167         u32 num_refs;
1168         int ret;
1169
1170         key.objectid = bytenr;
1171         if (parent) {
1172                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1173                 key.offset = parent;
1174                 size = sizeof(struct btrfs_shared_data_ref);
1175         } else {
1176                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1177                 key.offset = hash_extent_data_ref(root_objectid,
1178                                                   owner, offset);
1179                 size = sizeof(struct btrfs_extent_data_ref);
1180         }
1181
1182         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1183         if (ret && ret != -EEXIST)
1184                 goto fail;
1185
1186         leaf = path->nodes[0];
1187         if (parent) {
1188                 struct btrfs_shared_data_ref *ref;
1189                 ref = btrfs_item_ptr(leaf, path->slots[0],
1190                                      struct btrfs_shared_data_ref);
1191                 if (ret == 0) {
1192                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1193                 } else {
1194                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1195                         num_refs += refs_to_add;
1196                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1197                 }
1198         } else {
1199                 struct btrfs_extent_data_ref *ref;
1200                 while (ret == -EEXIST) {
1201                         ref = btrfs_item_ptr(leaf, path->slots[0],
1202                                              struct btrfs_extent_data_ref);
1203                         if (match_extent_data_ref(leaf, ref, root_objectid,
1204                                                   owner, offset))
1205                                 break;
1206                         btrfs_release_path(path);
1207                         key.offset++;
1208                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1209                                                       size);
1210                         if (ret && ret != -EEXIST)
1211                                 goto fail;
1212
1213                         leaf = path->nodes[0];
1214                 }
1215                 ref = btrfs_item_ptr(leaf, path->slots[0],
1216                                      struct btrfs_extent_data_ref);
1217                 if (ret == 0) {
1218                         btrfs_set_extent_data_ref_root(leaf, ref,
1219                                                        root_objectid);
1220                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1221                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1222                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1223                 } else {
1224                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1225                         num_refs += refs_to_add;
1226                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1227                 }
1228         }
1229         btrfs_mark_buffer_dirty(leaf);
1230         ret = 0;
1231 fail:
1232         btrfs_release_path(path);
1233         return ret;
1234 }
1235
1236 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1237                                            struct btrfs_root *root,
1238                                            struct btrfs_path *path,
1239                                            int refs_to_drop)
1240 {
1241         struct btrfs_key key;
1242         struct btrfs_extent_data_ref *ref1 = NULL;
1243         struct btrfs_shared_data_ref *ref2 = NULL;
1244         struct extent_buffer *leaf;
1245         u32 num_refs = 0;
1246         int ret = 0;
1247
1248         leaf = path->nodes[0];
1249         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1250
1251         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1252                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1253                                       struct btrfs_extent_data_ref);
1254                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1255         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1256                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1257                                       struct btrfs_shared_data_ref);
1258                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1259 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1260         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1261                 struct btrfs_extent_ref_v0 *ref0;
1262                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1263                                       struct btrfs_extent_ref_v0);
1264                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1265 #endif
1266         } else {
1267                 BUG();
1268         }
1269
1270         BUG_ON(num_refs < refs_to_drop);
1271         num_refs -= refs_to_drop;
1272
1273         if (num_refs == 0) {
1274                 ret = btrfs_del_item(trans, root, path);
1275         } else {
1276                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1277                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1278                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1279                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1281                 else {
1282                         struct btrfs_extent_ref_v0 *ref0;
1283                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1284                                         struct btrfs_extent_ref_v0);
1285                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1286                 }
1287 #endif
1288                 btrfs_mark_buffer_dirty(leaf);
1289         }
1290         return ret;
1291 }
1292
1293 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1294                                           struct btrfs_path *path,
1295                                           struct btrfs_extent_inline_ref *iref)
1296 {
1297         struct btrfs_key key;
1298         struct extent_buffer *leaf;
1299         struct btrfs_extent_data_ref *ref1;
1300         struct btrfs_shared_data_ref *ref2;
1301         u32 num_refs = 0;
1302
1303         leaf = path->nodes[0];
1304         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1305         if (iref) {
1306                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1307                     BTRFS_EXTENT_DATA_REF_KEY) {
1308                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1309                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1310                 } else {
1311                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1312                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1313                 }
1314         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1315                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1316                                       struct btrfs_extent_data_ref);
1317                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1318         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1319                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1320                                       struct btrfs_shared_data_ref);
1321                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1322 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1323         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1324                 struct btrfs_extent_ref_v0 *ref0;
1325                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1326                                       struct btrfs_extent_ref_v0);
1327                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1328 #endif
1329         } else {
1330                 WARN_ON(1);
1331         }
1332         return num_refs;
1333 }
1334
1335 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1336                                           struct btrfs_root *root,
1337                                           struct btrfs_path *path,
1338                                           u64 bytenr, u64 parent,
1339                                           u64 root_objectid)
1340 {
1341         struct btrfs_key key;
1342         int ret;
1343
1344         key.objectid = bytenr;
1345         if (parent) {
1346                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1347                 key.offset = parent;
1348         } else {
1349                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1350                 key.offset = root_objectid;
1351         }
1352
1353         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1354         if (ret > 0)
1355                 ret = -ENOENT;
1356 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1357         if (ret == -ENOENT && parent) {
1358                 btrfs_release_path(path);
1359                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1360                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1361                 if (ret > 0)
1362                         ret = -ENOENT;
1363         }
1364 #endif
1365         return ret;
1366 }
1367
1368 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1369                                           struct btrfs_root *root,
1370                                           struct btrfs_path *path,
1371                                           u64 bytenr, u64 parent,
1372                                           u64 root_objectid)
1373 {
1374         struct btrfs_key key;
1375         int ret;
1376
1377         key.objectid = bytenr;
1378         if (parent) {
1379                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1380                 key.offset = parent;
1381         } else {
1382                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1383                 key.offset = root_objectid;
1384         }
1385
1386         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1387         btrfs_release_path(path);
1388         return ret;
1389 }
1390
1391 static inline int extent_ref_type(u64 parent, u64 owner)
1392 {
1393         int type;
1394         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1395                 if (parent > 0)
1396                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1397                 else
1398                         type = BTRFS_TREE_BLOCK_REF_KEY;
1399         } else {
1400                 if (parent > 0)
1401                         type = BTRFS_SHARED_DATA_REF_KEY;
1402                 else
1403                         type = BTRFS_EXTENT_DATA_REF_KEY;
1404         }
1405         return type;
1406 }
1407
1408 static int find_next_key(struct btrfs_path *path, int level,
1409                          struct btrfs_key *key)
1410
1411 {
1412         for (; level < BTRFS_MAX_LEVEL; level++) {
1413                 if (!path->nodes[level])
1414                         break;
1415                 if (path->slots[level] + 1 >=
1416                     btrfs_header_nritems(path->nodes[level]))
1417                         continue;
1418                 if (level == 0)
1419                         btrfs_item_key_to_cpu(path->nodes[level], key,
1420                                               path->slots[level] + 1);
1421                 else
1422                         btrfs_node_key_to_cpu(path->nodes[level], key,
1423                                               path->slots[level] + 1);
1424                 return 0;
1425         }
1426         return 1;
1427 }
1428
1429 /*
1430  * look for inline back ref. if back ref is found, *ref_ret is set
1431  * to the address of inline back ref, and 0 is returned.
1432  *
1433  * if back ref isn't found, *ref_ret is set to the address where it
1434  * should be inserted, and -ENOENT is returned.
1435  *
1436  * if insert is true and there are too many inline back refs, the path
1437  * points to the extent item, and -EAGAIN is returned.
1438  *
1439  * NOTE: inline back refs are ordered in the same way that back ref
1440  *       items in the tree are ordered.
1441  */
1442 static noinline_for_stack
1443 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1444                                  struct btrfs_root *root,
1445                                  struct btrfs_path *path,
1446                                  struct btrfs_extent_inline_ref **ref_ret,
1447                                  u64 bytenr, u64 num_bytes,
1448                                  u64 parent, u64 root_objectid,
1449                                  u64 owner, u64 offset, int insert)
1450 {
1451         struct btrfs_key key;
1452         struct extent_buffer *leaf;
1453         struct btrfs_extent_item *ei;
1454         struct btrfs_extent_inline_ref *iref;
1455         u64 flags;
1456         u64 item_size;
1457         unsigned long ptr;
1458         unsigned long end;
1459         int extra_size;
1460         int type;
1461         int want;
1462         int ret;
1463         int err = 0;
1464
1465         key.objectid = bytenr;
1466         key.type = BTRFS_EXTENT_ITEM_KEY;
1467         key.offset = num_bytes;
1468
1469         want = extent_ref_type(parent, owner);
1470         if (insert) {
1471                 extra_size = btrfs_extent_inline_ref_size(want);
1472                 path->keep_locks = 1;
1473         } else
1474                 extra_size = -1;
1475         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1476         if (ret < 0) {
1477                 err = ret;
1478                 goto out;
1479         }
1480         if (ret && !insert) {
1481                 err = -ENOENT;
1482                 goto out;
1483         }
1484         BUG_ON(ret); /* Corruption */
1485
1486         leaf = path->nodes[0];
1487         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1488 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1489         if (item_size < sizeof(*ei)) {
1490                 if (!insert) {
1491                         err = -ENOENT;
1492                         goto out;
1493                 }
1494                 ret = convert_extent_item_v0(trans, root, path, owner,
1495                                              extra_size);
1496                 if (ret < 0) {
1497                         err = ret;
1498                         goto out;
1499                 }
1500                 leaf = path->nodes[0];
1501                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1502         }
1503 #endif
1504         BUG_ON(item_size < sizeof(*ei));
1505
1506         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1507         flags = btrfs_extent_flags(leaf, ei);
1508
1509         ptr = (unsigned long)(ei + 1);
1510         end = (unsigned long)ei + item_size;
1511
1512         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1513                 ptr += sizeof(struct btrfs_tree_block_info);
1514                 BUG_ON(ptr > end);
1515         } else {
1516                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1517         }
1518
1519         err = -ENOENT;
1520         while (1) {
1521                 if (ptr >= end) {
1522                         WARN_ON(ptr > end);
1523                         break;
1524                 }
1525                 iref = (struct btrfs_extent_inline_ref *)ptr;
1526                 type = btrfs_extent_inline_ref_type(leaf, iref);
1527                 if (want < type)
1528                         break;
1529                 if (want > type) {
1530                         ptr += btrfs_extent_inline_ref_size(type);
1531                         continue;
1532                 }
1533
1534                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1535                         struct btrfs_extent_data_ref *dref;
1536                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1537                         if (match_extent_data_ref(leaf, dref, root_objectid,
1538                                                   owner, offset)) {
1539                                 err = 0;
1540                                 break;
1541                         }
1542                         if (hash_extent_data_ref_item(leaf, dref) <
1543                             hash_extent_data_ref(root_objectid, owner, offset))
1544                                 break;
1545                 } else {
1546                         u64 ref_offset;
1547                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1548                         if (parent > 0) {
1549                                 if (parent == ref_offset) {
1550                                         err = 0;
1551                                         break;
1552                                 }
1553                                 if (ref_offset < parent)
1554                                         break;
1555                         } else {
1556                                 if (root_objectid == ref_offset) {
1557                                         err = 0;
1558                                         break;
1559                                 }
1560                                 if (ref_offset < root_objectid)
1561                                         break;
1562                         }
1563                 }
1564                 ptr += btrfs_extent_inline_ref_size(type);
1565         }
1566         if (err == -ENOENT && insert) {
1567                 if (item_size + extra_size >=
1568                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1569                         err = -EAGAIN;
1570                         goto out;
1571                 }
1572                 /*
1573                  * To add new inline back ref, we have to make sure
1574                  * there is no corresponding back ref item.
1575                  * For simplicity, we just do not add new inline back
1576                  * ref if there is any kind of item for this block
1577                  */
1578                 if (find_next_key(path, 0, &key) == 0 &&
1579                     key.objectid == bytenr &&
1580                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1581                         err = -EAGAIN;
1582                         goto out;
1583                 }
1584         }
1585         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1586 out:
1587         if (insert) {
1588                 path->keep_locks = 0;
1589                 btrfs_unlock_up_safe(path, 1);
1590         }
1591         return err;
1592 }
1593
1594 /*
1595  * helper to add new inline back ref
1596  */
1597 static noinline_for_stack
1598 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1599                                  struct btrfs_root *root,
1600                                  struct btrfs_path *path,
1601                                  struct btrfs_extent_inline_ref *iref,
1602                                  u64 parent, u64 root_objectid,
1603                                  u64 owner, u64 offset, int refs_to_add,
1604                                  struct btrfs_delayed_extent_op *extent_op)
1605 {
1606         struct extent_buffer *leaf;
1607         struct btrfs_extent_item *ei;
1608         unsigned long ptr;
1609         unsigned long end;
1610         unsigned long item_offset;
1611         u64 refs;
1612         int size;
1613         int type;
1614
1615         leaf = path->nodes[0];
1616         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1617         item_offset = (unsigned long)iref - (unsigned long)ei;
1618
1619         type = extent_ref_type(parent, owner);
1620         size = btrfs_extent_inline_ref_size(type);
1621
1622         btrfs_extend_item(trans, root, path, size);
1623
1624         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1625         refs = btrfs_extent_refs(leaf, ei);
1626         refs += refs_to_add;
1627         btrfs_set_extent_refs(leaf, ei, refs);
1628         if (extent_op)
1629                 __run_delayed_extent_op(extent_op, leaf, ei);
1630
1631         ptr = (unsigned long)ei + item_offset;
1632         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1633         if (ptr < end - size)
1634                 memmove_extent_buffer(leaf, ptr + size, ptr,
1635                                       end - size - ptr);
1636
1637         iref = (struct btrfs_extent_inline_ref *)ptr;
1638         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1639         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1640                 struct btrfs_extent_data_ref *dref;
1641                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1642                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1643                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1644                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1645                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1646         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1647                 struct btrfs_shared_data_ref *sref;
1648                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1649                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1650                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1651         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1652                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1653         } else {
1654                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1655         }
1656         btrfs_mark_buffer_dirty(leaf);
1657 }
1658
1659 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1660                                  struct btrfs_root *root,
1661                                  struct btrfs_path *path,
1662                                  struct btrfs_extent_inline_ref **ref_ret,
1663                                  u64 bytenr, u64 num_bytes, u64 parent,
1664                                  u64 root_objectid, u64 owner, u64 offset)
1665 {
1666         int ret;
1667
1668         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1669                                            bytenr, num_bytes, parent,
1670                                            root_objectid, owner, offset, 0);
1671         if (ret != -ENOENT)
1672                 return ret;
1673
1674         btrfs_release_path(path);
1675         *ref_ret = NULL;
1676
1677         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1678                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1679                                             root_objectid);
1680         } else {
1681                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1682                                              root_objectid, owner, offset);
1683         }
1684         return ret;
1685 }
1686
1687 /*
1688  * helper to update/remove inline back ref
1689  */
1690 static noinline_for_stack
1691 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1692                                   struct btrfs_root *root,
1693                                   struct btrfs_path *path,
1694                                   struct btrfs_extent_inline_ref *iref,
1695                                   int refs_to_mod,
1696                                   struct btrfs_delayed_extent_op *extent_op)
1697 {
1698         struct extent_buffer *leaf;
1699         struct btrfs_extent_item *ei;
1700         struct btrfs_extent_data_ref *dref = NULL;
1701         struct btrfs_shared_data_ref *sref = NULL;
1702         unsigned long ptr;
1703         unsigned long end;
1704         u32 item_size;
1705         int size;
1706         int type;
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                 btrfs_truncate_item(trans, root, path, item_size, 1);
1749         }
1750         btrfs_mark_buffer_dirty(leaf);
1751 }
1752
1753 static noinline_for_stack
1754 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1755                                  struct btrfs_root *root,
1756                                  struct btrfs_path *path,
1757                                  u64 bytenr, u64 num_bytes, u64 parent,
1758                                  u64 root_objectid, u64 owner,
1759                                  u64 offset, int refs_to_add,
1760                                  struct btrfs_delayed_extent_op *extent_op)
1761 {
1762         struct btrfs_extent_inline_ref *iref;
1763         int ret;
1764
1765         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1766                                            bytenr, num_bytes, parent,
1767                                            root_objectid, owner, offset, 1);
1768         if (ret == 0) {
1769                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1770                 update_inline_extent_backref(trans, root, path, iref,
1771                                              refs_to_add, extent_op);
1772         } else if (ret == -ENOENT) {
1773                 setup_inline_extent_backref(trans, root, path, iref, parent,
1774                                             root_objectid, owner, offset,
1775                                             refs_to_add, extent_op);
1776                 ret = 0;
1777         }
1778         return ret;
1779 }
1780
1781 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1782                                  struct btrfs_root *root,
1783                                  struct btrfs_path *path,
1784                                  u64 bytenr, u64 parent, u64 root_objectid,
1785                                  u64 owner, u64 offset, int refs_to_add)
1786 {
1787         int ret;
1788         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1789                 BUG_ON(refs_to_add != 1);
1790                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1791                                             parent, root_objectid);
1792         } else {
1793                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1794                                              parent, root_objectid,
1795                                              owner, offset, refs_to_add);
1796         }
1797         return ret;
1798 }
1799
1800 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1801                                  struct btrfs_root *root,
1802                                  struct btrfs_path *path,
1803                                  struct btrfs_extent_inline_ref *iref,
1804                                  int refs_to_drop, int is_data)
1805 {
1806         int ret = 0;
1807
1808         BUG_ON(!is_data && refs_to_drop != 1);
1809         if (iref) {
1810                 update_inline_extent_backref(trans, root, path, iref,
1811                                              -refs_to_drop, NULL);
1812         } else if (is_data) {
1813                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1814         } else {
1815                 ret = btrfs_del_item(trans, root, path);
1816         }
1817         return ret;
1818 }
1819
1820 static int btrfs_issue_discard(struct block_device *bdev,
1821                                 u64 start, u64 len)
1822 {
1823         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1824 }
1825
1826 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1827                                 u64 num_bytes, u64 *actual_bytes)
1828 {
1829         int ret;
1830         u64 discarded_bytes = 0;
1831         struct btrfs_bio *bbio = NULL;
1832
1833
1834         /* Tell the block device(s) that the sectors can be discarded */
1835         ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1836                               bytenr, &num_bytes, &bbio, 0);
1837         /* Error condition is -ENOMEM */
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; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
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 /* Can return -ENOMEM */
1873 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1874                          struct btrfs_root *root,
1875                          u64 bytenr, u64 num_bytes, u64 parent,
1876                          u64 root_objectid, u64 owner, u64 offset, int for_cow)
1877 {
1878         int ret;
1879         struct btrfs_fs_info *fs_info = root->fs_info;
1880
1881         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1882                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1883
1884         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1885                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1886                                         num_bytes,
1887                                         parent, root_objectid, (int)owner,
1888                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1889         } else {
1890                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1891                                         num_bytes,
1892                                         parent, root_objectid, owner, offset,
1893                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1894         }
1895         return ret;
1896 }
1897
1898 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1899                                   struct btrfs_root *root,
1900                                   u64 bytenr, u64 num_bytes,
1901                                   u64 parent, u64 root_objectid,
1902                                   u64 owner, u64 offset, int refs_to_add,
1903                                   struct btrfs_delayed_extent_op *extent_op)
1904 {
1905         struct btrfs_path *path;
1906         struct extent_buffer *leaf;
1907         struct btrfs_extent_item *item;
1908         u64 refs;
1909         int ret;
1910         int err = 0;
1911
1912         path = btrfs_alloc_path();
1913         if (!path)
1914                 return -ENOMEM;
1915
1916         path->reada = 1;
1917         path->leave_spinning = 1;
1918         /* this will setup the path even if it fails to insert the back ref */
1919         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1920                                            path, bytenr, num_bytes, parent,
1921                                            root_objectid, owner, offset,
1922                                            refs_to_add, extent_op);
1923         if (ret == 0)
1924                 goto out;
1925
1926         if (ret != -EAGAIN) {
1927                 err = ret;
1928                 goto out;
1929         }
1930
1931         leaf = path->nodes[0];
1932         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1933         refs = btrfs_extent_refs(leaf, item);
1934         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1935         if (extent_op)
1936                 __run_delayed_extent_op(extent_op, leaf, item);
1937
1938         btrfs_mark_buffer_dirty(leaf);
1939         btrfs_release_path(path);
1940
1941         path->reada = 1;
1942         path->leave_spinning = 1;
1943
1944         /* now insert the actual backref */
1945         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1946                                     path, bytenr, parent, root_objectid,
1947                                     owner, offset, refs_to_add);
1948         if (ret)
1949                 btrfs_abort_transaction(trans, root, ret);
1950 out:
1951         btrfs_free_path(path);
1952         return err;
1953 }
1954
1955 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1956                                 struct btrfs_root *root,
1957                                 struct btrfs_delayed_ref_node *node,
1958                                 struct btrfs_delayed_extent_op *extent_op,
1959                                 int insert_reserved)
1960 {
1961         int ret = 0;
1962         struct btrfs_delayed_data_ref *ref;
1963         struct btrfs_key ins;
1964         u64 parent = 0;
1965         u64 ref_root = 0;
1966         u64 flags = 0;
1967
1968         ins.objectid = node->bytenr;
1969         ins.offset = node->num_bytes;
1970         ins.type = BTRFS_EXTENT_ITEM_KEY;
1971
1972         ref = btrfs_delayed_node_to_data_ref(node);
1973         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1974                 parent = ref->parent;
1975         else
1976                 ref_root = ref->root;
1977
1978         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1979                 if (extent_op) {
1980                         BUG_ON(extent_op->update_key);
1981                         flags |= extent_op->flags_to_set;
1982                 }
1983                 ret = alloc_reserved_file_extent(trans, root,
1984                                                  parent, ref_root, flags,
1985                                                  ref->objectid, ref->offset,
1986                                                  &ins, node->ref_mod);
1987         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1988                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1989                                              node->num_bytes, parent,
1990                                              ref_root, ref->objectid,
1991                                              ref->offset, node->ref_mod,
1992                                              extent_op);
1993         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1994                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1995                                           node->num_bytes, parent,
1996                                           ref_root, ref->objectid,
1997                                           ref->offset, node->ref_mod,
1998                                           extent_op);
1999         } else {
2000                 BUG();
2001         }
2002         return ret;
2003 }
2004
2005 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2006                                     struct extent_buffer *leaf,
2007                                     struct btrfs_extent_item *ei)
2008 {
2009         u64 flags = btrfs_extent_flags(leaf, ei);
2010         if (extent_op->update_flags) {
2011                 flags |= extent_op->flags_to_set;
2012                 btrfs_set_extent_flags(leaf, ei, flags);
2013         }
2014
2015         if (extent_op->update_key) {
2016                 struct btrfs_tree_block_info *bi;
2017                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2018                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2019                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2020         }
2021 }
2022
2023 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2024                                  struct btrfs_root *root,
2025                                  struct btrfs_delayed_ref_node *node,
2026                                  struct btrfs_delayed_extent_op *extent_op)
2027 {
2028         struct btrfs_key key;
2029         struct btrfs_path *path;
2030         struct btrfs_extent_item *ei;
2031         struct extent_buffer *leaf;
2032         u32 item_size;
2033         int ret;
2034         int err = 0;
2035
2036         if (trans->aborted)
2037                 return 0;
2038
2039         path = btrfs_alloc_path();
2040         if (!path)
2041                 return -ENOMEM;
2042
2043         key.objectid = node->bytenr;
2044         key.type = BTRFS_EXTENT_ITEM_KEY;
2045         key.offset = node->num_bytes;
2046
2047         path->reada = 1;
2048         path->leave_spinning = 1;
2049         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2050                                 path, 0, 1);
2051         if (ret < 0) {
2052                 err = ret;
2053                 goto out;
2054         }
2055         if (ret > 0) {
2056                 err = -EIO;
2057                 goto out;
2058         }
2059
2060         leaf = path->nodes[0];
2061         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2062 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2063         if (item_size < sizeof(*ei)) {
2064                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2065                                              path, (u64)-1, 0);
2066                 if (ret < 0) {
2067                         err = ret;
2068                         goto out;
2069                 }
2070                 leaf = path->nodes[0];
2071                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2072         }
2073 #endif
2074         BUG_ON(item_size < sizeof(*ei));
2075         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2076         __run_delayed_extent_op(extent_op, leaf, ei);
2077
2078         btrfs_mark_buffer_dirty(leaf);
2079 out:
2080         btrfs_free_path(path);
2081         return err;
2082 }
2083
2084 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2085                                 struct btrfs_root *root,
2086                                 struct btrfs_delayed_ref_node *node,
2087                                 struct btrfs_delayed_extent_op *extent_op,
2088                                 int insert_reserved)
2089 {
2090         int ret = 0;
2091         struct btrfs_delayed_tree_ref *ref;
2092         struct btrfs_key ins;
2093         u64 parent = 0;
2094         u64 ref_root = 0;
2095
2096         ins.objectid = node->bytenr;
2097         ins.offset = node->num_bytes;
2098         ins.type = BTRFS_EXTENT_ITEM_KEY;
2099
2100         ref = btrfs_delayed_node_to_tree_ref(node);
2101         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2102                 parent = ref->parent;
2103         else
2104                 ref_root = ref->root;
2105
2106         BUG_ON(node->ref_mod != 1);
2107         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2108                 BUG_ON(!extent_op || !extent_op->update_flags ||
2109                        !extent_op->update_key);
2110                 ret = alloc_reserved_tree_block(trans, root,
2111                                                 parent, ref_root,
2112                                                 extent_op->flags_to_set,
2113                                                 &extent_op->key,
2114                                                 ref->level, &ins);
2115         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2116                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2117                                              node->num_bytes, parent, ref_root,
2118                                              ref->level, 0, 1, extent_op);
2119         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2120                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2121                                           node->num_bytes, parent, ref_root,
2122                                           ref->level, 0, 1, extent_op);
2123         } else {
2124                 BUG();
2125         }
2126         return ret;
2127 }
2128
2129 /* helper function to actually process a single delayed ref entry */
2130 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2131                                struct btrfs_root *root,
2132                                struct btrfs_delayed_ref_node *node,
2133                                struct btrfs_delayed_extent_op *extent_op,
2134                                int insert_reserved)
2135 {
2136         int ret = 0;
2137
2138         if (trans->aborted)
2139                 return 0;
2140
2141         if (btrfs_delayed_ref_is_head(node)) {
2142                 struct btrfs_delayed_ref_head *head;
2143                 /*
2144                  * we've hit the end of the chain and we were supposed
2145                  * to insert this extent into the tree.  But, it got
2146                  * deleted before we ever needed to insert it, so all
2147                  * we have to do is clean up the accounting
2148                  */
2149                 BUG_ON(extent_op);
2150                 head = btrfs_delayed_node_to_head(node);
2151                 if (insert_reserved) {
2152                         btrfs_pin_extent(root, node->bytenr,
2153                                          node->num_bytes, 1);
2154                         if (head->is_data) {
2155                                 ret = btrfs_del_csums(trans, root,
2156                                                       node->bytenr,
2157                                                       node->num_bytes);
2158                         }
2159                 }
2160                 mutex_unlock(&head->mutex);
2161                 return ret;
2162         }
2163
2164         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2165             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2166                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2167                                            insert_reserved);
2168         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2169                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2170                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2171                                            insert_reserved);
2172         else
2173                 BUG();
2174         return ret;
2175 }
2176
2177 static noinline struct btrfs_delayed_ref_node *
2178 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2179 {
2180         struct rb_node *node;
2181         struct btrfs_delayed_ref_node *ref;
2182         int action = BTRFS_ADD_DELAYED_REF;
2183 again:
2184         /*
2185          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2186          * this prevents ref count from going down to zero when
2187          * there still are pending delayed ref.
2188          */
2189         node = rb_prev(&head->node.rb_node);
2190         while (1) {
2191                 if (!node)
2192                         break;
2193                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2194                                 rb_node);
2195                 if (ref->bytenr != head->node.bytenr)
2196                         break;
2197                 if (ref->action == action)
2198                         return ref;
2199                 node = rb_prev(node);
2200         }
2201         if (action == BTRFS_ADD_DELAYED_REF) {
2202                 action = BTRFS_DROP_DELAYED_REF;
2203                 goto again;
2204         }
2205         return NULL;
2206 }
2207
2208 /*
2209  * Returns 0 on success or if called with an already aborted transaction.
2210  * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2211  */
2212 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2213                                        struct btrfs_root *root,
2214                                        struct list_head *cluster)
2215 {
2216         struct btrfs_delayed_ref_root *delayed_refs;
2217         struct btrfs_delayed_ref_node *ref;
2218         struct btrfs_delayed_ref_head *locked_ref = NULL;
2219         struct btrfs_delayed_extent_op *extent_op;
2220         int ret;
2221         int count = 0;
2222         int must_insert_reserved = 0;
2223
2224         delayed_refs = &trans->transaction->delayed_refs;
2225         while (1) {
2226                 if (!locked_ref) {
2227                         /* pick a new head ref from the cluster list */
2228                         if (list_empty(cluster))
2229                                 break;
2230
2231                         locked_ref = list_entry(cluster->next,
2232                                      struct btrfs_delayed_ref_head, cluster);
2233
2234                         /* grab the lock that says we are going to process
2235                          * all the refs for this head */
2236                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2237
2238                         /*
2239                          * we may have dropped the spin lock to get the head
2240                          * mutex lock, and that might have given someone else
2241                          * time to free the head.  If that's true, it has been
2242                          * removed from our list and we can move on.
2243                          */
2244                         if (ret == -EAGAIN) {
2245                                 locked_ref = NULL;
2246                                 count++;
2247                                 continue;
2248                         }
2249                 }
2250
2251                 /*
2252                  * locked_ref is the head node, so we have to go one
2253                  * node back for any delayed ref updates
2254                  */
2255                 ref = select_delayed_ref(locked_ref);
2256
2257                 if (ref && ref->seq &&
2258                     btrfs_check_delayed_seq(delayed_refs, ref->seq)) {
2259                         /*
2260                          * there are still refs with lower seq numbers in the
2261                          * process of being added. Don't run this ref yet.
2262                          */
2263                         list_del_init(&locked_ref->cluster);
2264                         mutex_unlock(&locked_ref->mutex);
2265                         locked_ref = NULL;
2266                         delayed_refs->num_heads_ready++;
2267                         spin_unlock(&delayed_refs->lock);
2268                         cond_resched();
2269                         spin_lock(&delayed_refs->lock);
2270                         continue;
2271                 }
2272
2273                 /*
2274                  * record the must insert reserved flag before we
2275                  * drop the spin lock.
2276                  */
2277                 must_insert_reserved = locked_ref->must_insert_reserved;
2278                 locked_ref->must_insert_reserved = 0;
2279
2280                 extent_op = locked_ref->extent_op;
2281                 locked_ref->extent_op = NULL;
2282
2283                 if (!ref) {
2284                         /* All delayed refs have been processed, Go ahead
2285                          * and send the head node to run_one_delayed_ref,
2286                          * so that any accounting fixes can happen
2287                          */
2288                         ref = &locked_ref->node;
2289
2290                         if (extent_op && must_insert_reserved) {
2291                                 kfree(extent_op);
2292                                 extent_op = NULL;
2293                         }
2294
2295                         if (extent_op) {
2296                                 spin_unlock(&delayed_refs->lock);
2297
2298                                 ret = run_delayed_extent_op(trans, root,
2299                                                             ref, extent_op);
2300                                 kfree(extent_op);
2301
2302                                 if (ret) {
2303                                         printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
2304                                         spin_lock(&delayed_refs->lock);
2305                                         return ret;
2306                                 }
2307
2308                                 goto next;
2309                         }
2310
2311                         list_del_init(&locked_ref->cluster);
2312                         locked_ref = NULL;
2313                 }
2314
2315                 ref->in_tree = 0;
2316                 rb_erase(&ref->rb_node, &delayed_refs->root);
2317                 delayed_refs->num_entries--;
2318                 /*
2319                  * we modified num_entries, but as we're currently running
2320                  * delayed refs, skip
2321                  *     wake_up(&delayed_refs->seq_wait);
2322                  * here.
2323                  */
2324                 spin_unlock(&delayed_refs->lock);
2325
2326                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2327                                           must_insert_reserved);
2328
2329                 btrfs_put_delayed_ref(ref);
2330                 kfree(extent_op);
2331                 count++;
2332
2333                 if (ret) {
2334                         printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
2335                         spin_lock(&delayed_refs->lock);
2336                         return ret;
2337                 }
2338
2339 next:
2340                 do_chunk_alloc(trans, root->fs_info->extent_root,
2341                                2 * 1024 * 1024,
2342                                btrfs_get_alloc_profile(root, 0),
2343                                CHUNK_ALLOC_NO_FORCE);
2344                 cond_resched();
2345                 spin_lock(&delayed_refs->lock);
2346         }
2347         return count;
2348 }
2349
2350
2351 static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs,
2352                         unsigned long num_refs)
2353 {
2354         struct list_head *first_seq = delayed_refs->seq_head.next;
2355
2356         spin_unlock(&delayed_refs->lock);
2357         pr_debug("waiting for more refs (num %ld, first %p)\n",
2358                  num_refs, first_seq);
2359         wait_event(delayed_refs->seq_wait,
2360                    num_refs != delayed_refs->num_entries ||
2361                    delayed_refs->seq_head.next != first_seq);
2362         pr_debug("done waiting for more refs (num %ld, first %p)\n",
2363                  delayed_refs->num_entries, delayed_refs->seq_head.next);
2364         spin_lock(&delayed_refs->lock);
2365 }
2366
2367 /*
2368  * this starts processing the delayed reference count updates and
2369  * extent insertions we have queued up so far.  count can be
2370  * 0, which means to process everything in the tree at the start
2371  * of the run (but not newly added entries), or it can be some target
2372  * number you'd like to process.
2373  *
2374  * Returns 0 on success or if called with an aborted transaction
2375  * Returns <0 on error and aborts the transaction
2376  */
2377 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2378                            struct btrfs_root *root, unsigned long count)
2379 {
2380         struct rb_node *node;
2381         struct btrfs_delayed_ref_root *delayed_refs;
2382         struct btrfs_delayed_ref_node *ref;
2383         struct list_head cluster;
2384         int ret;
2385         u64 delayed_start;
2386         int run_all = count == (unsigned long)-1;
2387         int run_most = 0;
2388         unsigned long num_refs = 0;
2389         int consider_waiting;
2390
2391         /* We'll clean this up in btrfs_cleanup_transaction */
2392         if (trans->aborted)
2393                 return 0;
2394
2395         if (root == root->fs_info->extent_root)
2396                 root = root->fs_info->tree_root;
2397
2398         do_chunk_alloc(trans, root->fs_info->extent_root,
2399                        2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2400                        CHUNK_ALLOC_NO_FORCE);
2401
2402         delayed_refs = &trans->transaction->delayed_refs;
2403         INIT_LIST_HEAD(&cluster);
2404 again:
2405         consider_waiting = 0;
2406         spin_lock(&delayed_refs->lock);
2407         if (count == 0) {
2408                 count = delayed_refs->num_entries * 2;
2409                 run_most = 1;
2410         }
2411         while (1) {
2412                 if (!(run_all || run_most) &&
2413                     delayed_refs->num_heads_ready < 64)
2414                         break;
2415
2416                 /*
2417                  * go find something we can process in the rbtree.  We start at
2418                  * the beginning of the tree, and then build a cluster
2419                  * of refs to process starting at the first one we are able to
2420                  * lock
2421                  */
2422                 delayed_start = delayed_refs->run_delayed_start;
2423                 ret = btrfs_find_ref_cluster(trans, &cluster,
2424                                              delayed_refs->run_delayed_start);
2425                 if (ret)
2426                         break;
2427
2428                 if (delayed_start >= delayed_refs->run_delayed_start) {
2429                         if (consider_waiting == 0) {
2430                                 /*
2431                                  * btrfs_find_ref_cluster looped. let's do one
2432                                  * more cycle. if we don't run any delayed ref
2433                                  * during that cycle (because we can't because
2434                                  * all of them are blocked) and if the number of
2435                                  * refs doesn't change, we avoid busy waiting.
2436                                  */
2437                                 consider_waiting = 1;
2438                                 num_refs = delayed_refs->num_entries;
2439                         } else {
2440                                 wait_for_more_refs(delayed_refs, num_refs);
2441                                 /*
2442                                  * after waiting, things have changed. we
2443                                  * dropped the lock and someone else might have
2444                                  * run some refs, built new clusters and so on.
2445                                  * therefore, we restart staleness detection.
2446                                  */
2447                                 consider_waiting = 0;
2448                         }
2449                 }
2450
2451                 ret = run_clustered_refs(trans, root, &cluster);
2452                 if (ret < 0) {
2453                         spin_unlock(&delayed_refs->lock);
2454                         btrfs_abort_transaction(trans, root, ret);
2455                         return ret;
2456                 }
2457
2458                 count -= min_t(unsigned long, ret, count);
2459
2460                 if (count == 0)
2461                         break;
2462
2463                 if (ret || delayed_refs->run_delayed_start == 0) {
2464                         /* refs were run, let's reset staleness detection */
2465                         consider_waiting = 0;
2466                 }
2467         }
2468
2469         if (run_all) {
2470                 node = rb_first(&delayed_refs->root);
2471                 if (!node)
2472                         goto out;
2473                 count = (unsigned long)-1;
2474
2475                 while (node) {
2476                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2477                                        rb_node);
2478                         if (btrfs_delayed_ref_is_head(ref)) {
2479                                 struct btrfs_delayed_ref_head *head;
2480
2481                                 head = btrfs_delayed_node_to_head(ref);
2482                                 atomic_inc(&ref->refs);
2483
2484                                 spin_unlock(&delayed_refs->lock);
2485                                 /*
2486                                  * Mutex was contended, block until it's
2487                                  * released and try again
2488                                  */
2489                                 mutex_lock(&head->mutex);
2490                                 mutex_unlock(&head->mutex);
2491
2492                                 btrfs_put_delayed_ref(ref);
2493                                 cond_resched();
2494                                 goto again;
2495                         }
2496                         node = rb_next(node);
2497                 }
2498                 spin_unlock(&delayed_refs->lock);
2499                 schedule_timeout(1);
2500                 goto again;
2501         }
2502 out:
2503         spin_unlock(&delayed_refs->lock);
2504         return 0;
2505 }
2506
2507 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2508                                 struct btrfs_root *root,
2509                                 u64 bytenr, u64 num_bytes, u64 flags,
2510                                 int is_data)
2511 {
2512         struct btrfs_delayed_extent_op *extent_op;
2513         int ret;
2514
2515         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2516         if (!extent_op)
2517                 return -ENOMEM;
2518
2519         extent_op->flags_to_set = flags;
2520         extent_op->update_flags = 1;
2521         extent_op->update_key = 0;
2522         extent_op->is_data = is_data ? 1 : 0;
2523
2524         ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2525                                           num_bytes, extent_op);
2526         if (ret)
2527                 kfree(extent_op);
2528         return ret;
2529 }
2530
2531 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2532                                       struct btrfs_root *root,
2533                                       struct btrfs_path *path,
2534                                       u64 objectid, u64 offset, u64 bytenr)
2535 {
2536         struct btrfs_delayed_ref_head *head;
2537         struct btrfs_delayed_ref_node *ref;
2538         struct btrfs_delayed_data_ref *data_ref;
2539         struct btrfs_delayed_ref_root *delayed_refs;
2540         struct rb_node *node;
2541         int ret = 0;
2542
2543         ret = -ENOENT;
2544         delayed_refs = &trans->transaction->delayed_refs;
2545         spin_lock(&delayed_refs->lock);
2546         head = btrfs_find_delayed_ref_head(trans, bytenr);
2547         if (!head)
2548                 goto out;
2549
2550         if (!mutex_trylock(&head->mutex)) {
2551                 atomic_inc(&head->node.refs);
2552                 spin_unlock(&delayed_refs->lock);
2553
2554                 btrfs_release_path(path);
2555
2556                 /*
2557                  * Mutex was contended, block until it's released and let
2558                  * caller try again
2559                  */
2560                 mutex_lock(&head->mutex);
2561                 mutex_unlock(&head->mutex);
2562                 btrfs_put_delayed_ref(&head->node);
2563                 return -EAGAIN;
2564         }
2565
2566         node = rb_prev(&head->node.rb_node);
2567         if (!node)
2568                 goto out_unlock;
2569
2570         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2571
2572         if (ref->bytenr != bytenr)
2573                 goto out_unlock;
2574
2575         ret = 1;
2576         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2577                 goto out_unlock;
2578
2579         data_ref = btrfs_delayed_node_to_data_ref(ref);
2580
2581         node = rb_prev(node);
2582         if (node) {
2583                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2584                 if (ref->bytenr == bytenr)
2585                         goto out_unlock;
2586         }
2587
2588         if (data_ref->root != root->root_key.objectid ||
2589             data_ref->objectid != objectid || data_ref->offset != offset)
2590                 goto out_unlock;
2591
2592         ret = 0;
2593 out_unlock:
2594         mutex_unlock(&head->mutex);
2595 out:
2596         spin_unlock(&delayed_refs->lock);
2597         return ret;
2598 }
2599
2600 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2601                                         struct btrfs_root *root,
2602                                         struct btrfs_path *path,
2603                                         u64 objectid, u64 offset, u64 bytenr)
2604 {
2605         struct btrfs_root *extent_root = root->fs_info->extent_root;
2606         struct extent_buffer *leaf;
2607         struct btrfs_extent_data_ref *ref;
2608         struct btrfs_extent_inline_ref *iref;
2609         struct btrfs_extent_item *ei;
2610         struct btrfs_key key;
2611         u32 item_size;
2612         int ret;
2613
2614         key.objectid = bytenr;
2615         key.offset = (u64)-1;
2616         key.type = BTRFS_EXTENT_ITEM_KEY;
2617
2618         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2619         if (ret < 0)
2620                 goto out;
2621         BUG_ON(ret == 0); /* Corruption */
2622
2623         ret = -ENOENT;
2624         if (path->slots[0] == 0)
2625                 goto out;
2626
2627         path->slots[0]--;
2628         leaf = path->nodes[0];
2629         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2630
2631         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2632                 goto out;
2633
2634         ret = 1;
2635         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2636 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2637         if (item_size < sizeof(*ei)) {
2638                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2639                 goto out;
2640         }
2641 #endif
2642         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2643
2644         if (item_size != sizeof(*ei) +
2645             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2646                 goto out;
2647
2648         if (btrfs_extent_generation(leaf, ei) <=
2649             btrfs_root_last_snapshot(&root->root_item))
2650                 goto out;
2651
2652         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2653         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2654             BTRFS_EXTENT_DATA_REF_KEY)
2655                 goto out;
2656
2657         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2658         if (btrfs_extent_refs(leaf, ei) !=
2659             btrfs_extent_data_ref_count(leaf, ref) ||
2660             btrfs_extent_data_ref_root(leaf, ref) !=
2661             root->root_key.objectid ||
2662             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2663             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2664                 goto out;
2665
2666         ret = 0;
2667 out:
2668         return ret;
2669 }
2670
2671 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2672                           struct btrfs_root *root,
2673                           u64 objectid, u64 offset, u64 bytenr)
2674 {
2675         struct btrfs_path *path;
2676         int ret;
2677         int ret2;
2678
2679         path = btrfs_alloc_path();
2680         if (!path)
2681                 return -ENOENT;
2682
2683         do {
2684                 ret = check_committed_ref(trans, root, path, objectid,
2685                                           offset, bytenr);
2686                 if (ret && ret != -ENOENT)
2687                         goto out;
2688
2689                 ret2 = check_delayed_ref(trans, root, path, objectid,
2690                                          offset, bytenr);
2691         } while (ret2 == -EAGAIN);
2692
2693         if (ret2 && ret2 != -ENOENT) {
2694                 ret = ret2;
2695                 goto out;
2696         }
2697
2698         if (ret != -ENOENT || ret2 != -ENOENT)
2699                 ret = 0;
2700 out:
2701         btrfs_free_path(path);
2702         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2703                 WARN_ON(ret > 0);
2704         return ret;
2705 }
2706
2707 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2708                            struct btrfs_root *root,
2709                            struct extent_buffer *buf,
2710                            int full_backref, int inc, int for_cow)
2711 {
2712         u64 bytenr;
2713         u64 num_bytes;
2714         u64 parent;
2715         u64 ref_root;
2716         u32 nritems;
2717         struct btrfs_key key;
2718         struct btrfs_file_extent_item *fi;
2719         int i;
2720         int level;
2721         int ret = 0;
2722         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2723                             u64, u64, u64, u64, u64, u64, int);
2724
2725         ref_root = btrfs_header_owner(buf);
2726         nritems = btrfs_header_nritems(buf);
2727         level = btrfs_header_level(buf);
2728
2729         if (!root->ref_cows && level == 0)
2730                 return 0;
2731
2732         if (inc)
2733                 process_func = btrfs_inc_extent_ref;
2734         else
2735                 process_func = btrfs_free_extent;
2736
2737         if (full_backref)
2738                 parent = buf->start;
2739         else
2740                 parent = 0;
2741
2742         for (i = 0; i < nritems; i++) {
2743                 if (level == 0) {
2744                         btrfs_item_key_to_cpu(buf, &key, i);
2745                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2746                                 continue;
2747                         fi = btrfs_item_ptr(buf, i,
2748                                             struct btrfs_file_extent_item);
2749                         if (btrfs_file_extent_type(buf, fi) ==
2750                             BTRFS_FILE_EXTENT_INLINE)
2751                                 continue;
2752                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2753                         if (bytenr == 0)
2754                                 continue;
2755
2756                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2757                         key.offset -= btrfs_file_extent_offset(buf, fi);
2758                         ret = process_func(trans, root, bytenr, num_bytes,
2759                                            parent, ref_root, key.objectid,
2760                                            key.offset, for_cow);
2761                         if (ret)
2762                                 goto fail;
2763                 } else {
2764                         bytenr = btrfs_node_blockptr(buf, i);
2765                         num_bytes = btrfs_level_size(root, level - 1);
2766                         ret = process_func(trans, root, bytenr, num_bytes,
2767                                            parent, ref_root, level - 1, 0,
2768                                            for_cow);
2769                         if (ret)
2770                                 goto fail;
2771                 }
2772         }
2773         return 0;
2774 fail:
2775         return ret;
2776 }
2777
2778 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2779                   struct extent_buffer *buf, int full_backref, int for_cow)
2780 {
2781         return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2782 }
2783
2784 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2785                   struct extent_buffer *buf, int full_backref, int for_cow)
2786 {
2787         return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2788 }
2789
2790 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2791                                  struct btrfs_root *root,
2792                                  struct btrfs_path *path,
2793                                  struct btrfs_block_group_cache *cache)
2794 {
2795         int ret;
2796         struct btrfs_root *extent_root = root->fs_info->extent_root;
2797         unsigned long bi;
2798         struct extent_buffer *leaf;
2799
2800         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2801         if (ret < 0)
2802                 goto fail;
2803         BUG_ON(ret); /* Corruption */
2804
2805         leaf = path->nodes[0];
2806         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2807         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2808         btrfs_mark_buffer_dirty(leaf);
2809         btrfs_release_path(path);
2810 fail:
2811         if (ret) {
2812                 btrfs_abort_transaction(trans, root, ret);
2813                 return ret;
2814         }
2815         return 0;
2816
2817 }
2818
2819 static struct btrfs_block_group_cache *
2820 next_block_group(struct btrfs_root *root,
2821                  struct btrfs_block_group_cache *cache)
2822 {
2823         struct rb_node *node;
2824         spin_lock(&root->fs_info->block_group_cache_lock);
2825         node = rb_next(&cache->cache_node);
2826         btrfs_put_block_group(cache);
2827         if (node) {
2828                 cache = rb_entry(node, struct btrfs_block_group_cache,
2829                                  cache_node);
2830                 btrfs_get_block_group(cache);
2831         } else
2832                 cache = NULL;
2833         spin_unlock(&root->fs_info->block_group_cache_lock);
2834         return cache;
2835 }
2836
2837 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2838                             struct btrfs_trans_handle *trans,
2839                             struct btrfs_path *path)
2840 {
2841         struct btrfs_root *root = block_group->fs_info->tree_root;
2842         struct inode *inode = NULL;
2843         u64 alloc_hint = 0;
2844         int dcs = BTRFS_DC_ERROR;
2845         int num_pages = 0;
2846         int retries = 0;
2847         int ret = 0;
2848
2849         /*
2850          * If this block group is smaller than 100 megs don't bother caching the
2851          * block group.
2852          */
2853         if (block_group->key.offset < (100 * 1024 * 1024)) {
2854                 spin_lock(&block_group->lock);
2855                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2856                 spin_unlock(&block_group->lock);
2857                 return 0;
2858         }
2859
2860 again:
2861         inode = lookup_free_space_inode(root, block_group, path);
2862         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2863                 ret = PTR_ERR(inode);
2864                 btrfs_release_path(path);
2865                 goto out;
2866         }
2867
2868         if (IS_ERR(inode)) {
2869                 BUG_ON(retries);
2870                 retries++;
2871
2872                 if (block_group->ro)
2873                         goto out_free;
2874
2875                 ret = create_free_space_inode(root, trans, block_group, path);
2876                 if (ret)
2877                         goto out_free;
2878                 goto again;
2879         }
2880
2881         /* We've already setup this transaction, go ahead and exit */
2882         if (block_group->cache_generation == trans->transid &&
2883             i_size_read(inode)) {
2884                 dcs = BTRFS_DC_SETUP;
2885                 goto out_put;
2886         }
2887
2888         /*
2889          * We want to set the generation to 0, that way if anything goes wrong
2890          * from here on out we know not to trust this cache when we load up next
2891          * time.
2892          */
2893         BTRFS_I(inode)->generation = 0;
2894         ret = btrfs_update_inode(trans, root, inode);
2895         WARN_ON(ret);
2896
2897         if (i_size_read(inode) > 0) {
2898                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2899                                                       inode);
2900                 if (ret)
2901                         goto out_put;
2902         }
2903
2904         spin_lock(&block_group->lock);
2905         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2906                 /* We're not cached, don't bother trying to write stuff out */
2907                 dcs = BTRFS_DC_WRITTEN;
2908                 spin_unlock(&block_group->lock);
2909                 goto out_put;
2910         }
2911         spin_unlock(&block_group->lock);
2912
2913         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2914         if (!num_pages)
2915                 num_pages = 1;
2916
2917         /*
2918          * Just to make absolutely sure we have enough space, we're going to
2919          * preallocate 12 pages worth of space for each block group.  In
2920          * practice we ought to use at most 8, but we need extra space so we can
2921          * add our header and have a terminator between the extents and the
2922          * bitmaps.
2923          */
2924         num_pages *= 16;
2925         num_pages *= PAGE_CACHE_SIZE;
2926
2927         ret = btrfs_check_data_free_space(inode, num_pages);
2928         if (ret)
2929                 goto out_put;
2930
2931         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2932                                               num_pages, num_pages,
2933                                               &alloc_hint);
2934         if (!ret)
2935                 dcs = BTRFS_DC_SETUP;
2936         btrfs_free_reserved_data_space(inode, num_pages);
2937
2938 out_put:
2939         iput(inode);
2940 out_free:
2941         btrfs_release_path(path);
2942 out:
2943         spin_lock(&block_group->lock);
2944         if (!ret && dcs == BTRFS_DC_SETUP)
2945                 block_group->cache_generation = trans->transid;
2946         block_group->disk_cache_state = dcs;
2947         spin_unlock(&block_group->lock);
2948
2949         return ret;
2950 }
2951
2952 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2953                                    struct btrfs_root *root)
2954 {
2955         struct btrfs_block_group_cache *cache;
2956         int err = 0;
2957         struct btrfs_path *path;
2958         u64 last = 0;
2959
2960         path = btrfs_alloc_path();
2961         if (!path)
2962                 return -ENOMEM;
2963
2964 again:
2965         while (1) {
2966                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2967                 while (cache) {
2968                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2969                                 break;
2970                         cache = next_block_group(root, cache);
2971                 }
2972                 if (!cache) {
2973                         if (last == 0)
2974                                 break;
2975                         last = 0;
2976                         continue;
2977                 }
2978                 err = cache_save_setup(cache, trans, path);
2979                 last = cache->key.objectid + cache->key.offset;
2980                 btrfs_put_block_group(cache);
2981         }
2982
2983         while (1) {
2984                 if (last == 0) {
2985                         err = btrfs_run_delayed_refs(trans, root,
2986                                                      (unsigned long)-1);
2987                         if (err) /* File system offline */
2988                                 goto out;
2989                 }
2990
2991                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2992                 while (cache) {
2993                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2994                                 btrfs_put_block_group(cache);
2995                                 goto again;
2996                         }
2997
2998                         if (cache->dirty)
2999                                 break;
3000                         cache = next_block_group(root, cache);
3001                 }
3002                 if (!cache) {
3003                         if (last == 0)
3004                                 break;
3005                         last = 0;
3006                         continue;
3007                 }
3008
3009                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3010                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3011                 cache->dirty = 0;
3012                 last = cache->key.objectid + cache->key.offset;
3013
3014                 err = write_one_cache_group(trans, root, path, cache);
3015                 if (err) /* File system offline */
3016                         goto out;
3017
3018                 btrfs_put_block_group(cache);
3019         }
3020
3021         while (1) {
3022                 /*
3023                  * I don't think this is needed since we're just marking our
3024                  * preallocated extent as written, but just in case it can't
3025                  * hurt.
3026                  */
3027                 if (last == 0) {
3028                         err = btrfs_run_delayed_refs(trans, root,
3029                                                      (unsigned long)-1);
3030                         if (err) /* File system offline */
3031                                 goto out;
3032                 }
3033
3034                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3035                 while (cache) {
3036                         /*
3037                          * Really this shouldn't happen, but it could if we
3038                          * couldn't write the entire preallocated extent and
3039                          * splitting the extent resulted in a new block.
3040                          */
3041                         if (cache->dirty) {
3042                                 btrfs_put_block_group(cache);
3043                                 goto again;
3044                         }
3045                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3046                                 break;
3047                         cache = next_block_group(root, cache);
3048                 }
3049                 if (!cache) {
3050                         if (last == 0)
3051                                 break;
3052                         last = 0;
3053                         continue;
3054                 }
3055
3056                 err = btrfs_write_out_cache(root, trans, cache, path);
3057
3058                 /*
3059                  * If we didn't have an error then the cache state is still
3060                  * NEED_WRITE, so we can set it to WRITTEN.
3061                  */
3062                 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3063                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
3064                 last = cache->key.objectid + cache->key.offset;
3065                 btrfs_put_block_group(cache);
3066         }
3067 out:
3068
3069         btrfs_free_path(path);
3070         return err;
3071 }
3072
3073 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3074 {
3075         struct btrfs_block_group_cache *block_group;
3076         int readonly = 0;
3077
3078         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3079         if (!block_group || block_group->ro)
3080                 readonly = 1;
3081         if (block_group)
3082                 btrfs_put_block_group(block_group);
3083         return readonly;
3084 }
3085
3086 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3087                              u64 total_bytes, u64 bytes_used,
3088                              struct btrfs_space_info **space_info)
3089 {
3090         struct btrfs_space_info *found;
3091         int i;
3092         int factor;
3093
3094         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3095                      BTRFS_BLOCK_GROUP_RAID10))
3096                 factor = 2;
3097         else
3098                 factor = 1;
3099
3100         found = __find_space_info(info, flags);
3101         if (found) {
3102                 spin_lock(&found->lock);
3103                 found->total_bytes += total_bytes;
3104                 found->disk_total += total_bytes * factor;
3105                 found->bytes_used += bytes_used;
3106                 found->disk_used += bytes_used * factor;
3107                 found->full = 0;
3108                 spin_unlock(&found->lock);
3109                 *space_info = found;
3110                 return 0;
3111         }
3112         found = kzalloc(sizeof(*found), GFP_NOFS);
3113         if (!found)
3114                 return -ENOMEM;
3115
3116         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3117                 INIT_LIST_HEAD(&found->block_groups[i]);
3118         init_rwsem(&found->groups_sem);
3119         spin_lock_init(&found->lock);
3120         found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3121         found->total_bytes = total_bytes;
3122         found->disk_total = total_bytes * factor;
3123         found->bytes_used = bytes_used;
3124         found->disk_used = bytes_used * factor;
3125         found->bytes_pinned = 0;
3126         found->bytes_reserved = 0;
3127         found->bytes_readonly = 0;
3128         found->bytes_may_use = 0;
3129         found->full = 0;
3130         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3131         found->chunk_alloc = 0;
3132         found->flush = 0;
3133         init_waitqueue_head(&found->wait);
3134         *space_info = found;
3135         list_add_rcu(&found->list, &info->space_info);
3136         return 0;
3137 }
3138
3139 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3140 {
3141         u64 extra_flags = chunk_to_extended(flags) &
3142                                 BTRFS_EXTENDED_PROFILE_MASK;
3143
3144         if (flags & BTRFS_BLOCK_GROUP_DATA)
3145                 fs_info->avail_data_alloc_bits |= extra_flags;
3146         if (flags & BTRFS_BLOCK_GROUP_METADATA)
3147                 fs_info->avail_metadata_alloc_bits |= extra_flags;
3148         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3149                 fs_info->avail_system_alloc_bits |= extra_flags;
3150 }
3151
3152 /*
3153  * returns target flags in extended format or 0 if restripe for this
3154  * chunk_type is not in progress
3155  *
3156  * should be called with either volume_mutex or balance_lock held
3157  */
3158 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3159 {
3160         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3161         u64 target = 0;
3162
3163         if (!bctl)
3164                 return 0;
3165
3166         if (flags & BTRFS_BLOCK_GROUP_DATA &&
3167             bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3168                 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3169         } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3170                    bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3171                 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3172         } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3173                    bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3174                 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3175         }
3176
3177         return target;
3178 }
3179
3180 /*
3181  * @flags: available profiles in extended format (see ctree.h)
3182  *
3183  * Returns reduced profile in chunk format.  If profile changing is in
3184  * progress (either running or paused) picks the target profile (if it's
3185  * already available), otherwise falls back to plain reducing.
3186  */
3187 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3188 {
3189         /*
3190          * we add in the count of missing devices because we want
3191          * to make sure that any RAID levels on a degraded FS
3192          * continue to be honored.
3193          */
3194         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3195                 root->fs_info->fs_devices->missing_devices;
3196         u64 target;
3197
3198         /*
3199          * see if restripe for this chunk_type is in progress, if so
3200          * try to reduce to the target profile
3201          */
3202         spin_lock(&root->fs_info->balance_lock);
3203         target = get_restripe_target(root->fs_info, flags);
3204         if (target) {
3205                 /* pick target profile only if it's already available */
3206                 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3207                         spin_unlock(&root->fs_info->balance_lock);
3208                         return extended_to_chunk(target);
3209                 }
3210         }
3211         spin_unlock(&root->fs_info->balance_lock);
3212
3213         if (num_devices == 1)
3214                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3215         if (num_devices < 4)
3216                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3217
3218         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3219             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3220                       BTRFS_BLOCK_GROUP_RAID10))) {
3221                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3222         }
3223
3224         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3225             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3226                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3227         }
3228
3229         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3230             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3231              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3232              (flags & BTRFS_BLOCK_GROUP_DUP))) {
3233                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3234         }
3235
3236         return extended_to_chunk(flags);
3237 }
3238
3239 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3240 {
3241         if (flags & BTRFS_BLOCK_GROUP_DATA)
3242                 flags |= root->fs_info->avail_data_alloc_bits;
3243         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3244                 flags |= root->fs_info->avail_system_alloc_bits;
3245         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3246                 flags |= root->fs_info->avail_metadata_alloc_bits;
3247
3248         return btrfs_reduce_alloc_profile(root, flags);
3249 }
3250
3251 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3252 {
3253         u64 flags;
3254
3255         if (data)
3256                 flags = BTRFS_BLOCK_GROUP_DATA;
3257         else if (root == root->fs_info->chunk_root)
3258                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3259         else
3260                 flags = BTRFS_BLOCK_GROUP_METADATA;
3261
3262         return get_alloc_profile(root, flags);
3263 }
3264
3265 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3266 {
3267         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3268                                                        BTRFS_BLOCK_GROUP_DATA);
3269 }
3270
3271 /*
3272  * This will check the space that the inode allocates from to make sure we have
3273  * enough space for bytes.
3274  */
3275 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3276 {
3277         struct btrfs_space_info *data_sinfo;
3278         struct btrfs_root *root = BTRFS_I(inode)->root;
3279         u64 used;
3280         int ret = 0, committed = 0, alloc_chunk = 1;
3281
3282         /* make sure bytes are sectorsize aligned */
3283         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3284
3285         if (root == root->fs_info->tree_root ||
3286             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3287                 alloc_chunk = 0;
3288                 committed = 1;
3289         }
3290
3291         data_sinfo = BTRFS_I(inode)->space_info;
3292         if (!data_sinfo)
3293                 goto alloc;
3294
3295 again:
3296         /* make sure we have enough space to handle the data first */
3297         spin_lock(&data_sinfo->lock);
3298         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3299                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3300                 data_sinfo->bytes_may_use;
3301
3302         if (used + bytes > data_sinfo->total_bytes) {
3303                 struct btrfs_trans_handle *trans;
3304
3305                 /*
3306                  * if we don't have enough free bytes in this space then we need
3307                  * to alloc a new chunk.
3308                  */
3309                 if (!data_sinfo->full && alloc_chunk) {
3310                         u64 alloc_target;
3311
3312                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3313                         spin_unlock(&data_sinfo->lock);
3314 alloc:
3315                         alloc_target = btrfs_get_alloc_profile(root, 1);
3316                         trans = btrfs_join_transaction(root);
3317                         if (IS_ERR(trans))
3318                                 return PTR_ERR(trans);
3319
3320                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3321                                              bytes + 2 * 1024 * 1024,
3322                                              alloc_target,
3323                                              CHUNK_ALLOC_NO_FORCE);
3324                         btrfs_end_transaction(trans, root);
3325                         if (ret < 0) {
3326                                 if (ret != -ENOSPC)
3327                                         return ret;
3328                                 else
3329                                         goto commit_trans;
3330                         }
3331
3332                         if (!data_sinfo) {
3333                                 btrfs_set_inode_space_info(root, inode);
3334                                 data_sinfo = BTRFS_I(inode)->space_info;
3335                         }
3336                         goto again;
3337                 }
3338
3339                 /*
3340                  * If we have less pinned bytes than we want to allocate then
3341                  * don't bother committing the transaction, it won't help us.
3342                  */
3343                 if (data_sinfo->bytes_pinned < bytes)
3344                         committed = 1;
3345                 spin_unlock(&data_sinfo->lock);
3346
3347                 /* commit the current transaction and try again */
3348 commit_trans:
3349                 if (!committed &&
3350                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3351                         committed = 1;
3352                         trans = btrfs_join_transaction(root);
3353                         if (IS_ERR(trans))
3354                                 return PTR_ERR(trans);
3355                         ret = btrfs_commit_transaction(trans, root);
3356                         if (ret)
3357                                 return ret;
3358                         goto again;
3359                 }
3360
3361                 return -ENOSPC;
3362         }
3363         data_sinfo->bytes_may_use += bytes;
3364         trace_btrfs_space_reservation(root->fs_info, "space_info",
3365                                       data_sinfo->flags, bytes, 1);
3366         spin_unlock(&data_sinfo->lock);
3367
3368         return 0;
3369 }
3370
3371 /*
3372  * Called if we need to clear a data reservation for this inode.
3373  */
3374 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3375 {
3376         struct btrfs_root *root = BTRFS_I(inode)->root;
3377         struct btrfs_space_info *data_sinfo;
3378
3379         /* make sure bytes are sectorsize aligned */
3380         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3381
3382         data_sinfo = BTRFS_I(inode)->space_info;
3383         spin_lock(&data_sinfo->lock);
3384         data_sinfo->bytes_may_use -= bytes;
3385         trace_btrfs_space_reservation(root->fs_info, "space_info",
3386                                       data_sinfo->flags, bytes, 0);
3387         spin_unlock(&data_sinfo->lock);
3388 }
3389
3390 static void force_metadata_allocation(struct btrfs_fs_info *info)
3391 {
3392         struct list_head *head = &info->space_info;
3393         struct btrfs_space_info *found;
3394
3395         rcu_read_lock();
3396         list_for_each_entry_rcu(found, head, list) {
3397                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3398                         found->force_alloc = CHUNK_ALLOC_FORCE;
3399         }
3400         rcu_read_unlock();
3401 }
3402
3403 static int should_alloc_chunk(struct btrfs_root *root,
3404                               struct btrfs_space_info *sinfo, u64 alloc_bytes,
3405                               int force)
3406 {
3407         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3408         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3409         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3410         u64 thresh;
3411
3412         if (force == CHUNK_ALLOC_FORCE)
3413                 return 1;
3414
3415         /*
3416          * We need to take into account the global rsv because for all intents
3417          * and purposes it's used space.  Don't worry about locking the
3418          * global_rsv, it doesn't change except when the transaction commits.
3419          */
3420         num_allocated += global_rsv->size;
3421
3422         /*
3423          * in limited mode, we want to have some free space up to
3424          * about 1% of the FS size.
3425          */
3426         if (force == CHUNK_ALLOC_LIMITED) {
3427                 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3428                 thresh = max_t(u64, 64 * 1024 * 1024,
3429                                div_factor_fine(thresh, 1));
3430
3431                 if (num_bytes - num_allocated < thresh)
3432                         return 1;
3433         }
3434         thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3435
3436         /* 256MB or 2% of the FS */
3437         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3438         /* system chunks need a much small threshold */
3439         if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM)
3440                 thresh = 32 * 1024 * 1024;
3441
3442         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3443                 return 0;
3444         return 1;
3445 }
3446
3447 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3448 {
3449         u64 num_dev;
3450
3451         if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3452             type & BTRFS_BLOCK_GROUP_RAID0)
3453                 num_dev = root->fs_info->fs_devices->rw_devices;
3454         else if (type & BTRFS_BLOCK_GROUP_RAID1)
3455                 num_dev = 2;
3456         else
3457                 num_dev = 1;    /* DUP or single */
3458
3459         /* metadata for updaing devices and chunk tree */
3460         return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3461 }
3462
3463 static void check_system_chunk(struct btrfs_trans_handle *trans,
3464                                struct btrfs_root *root, u64 type)
3465 {
3466         struct btrfs_space_info *info;
3467         u64 left;
3468         u64 thresh;
3469
3470         info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3471         spin_lock(&info->lock);
3472         left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3473                 info->bytes_reserved - info->bytes_readonly;
3474         spin_unlock(&info->lock);
3475
3476         thresh = get_system_chunk_thresh(root, type);
3477         if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3478                 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3479                        left, thresh, type);
3480                 dump_space_info(info, 0, 0);
3481         }
3482
3483         if (left < thresh) {
3484                 u64 flags;
3485
3486                 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3487                 btrfs_alloc_chunk(trans, root, flags);
3488         }
3489 }
3490
3491 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3492                           struct btrfs_root *extent_root, u64 alloc_bytes,
3493                           u64 flags, int force)
3494 {
3495         struct btrfs_space_info *space_info;
3496         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3497         int wait_for_alloc = 0;
3498         int ret = 0;
3499
3500         space_info = __find_space_info(extent_root->fs_info, flags);
3501         if (!space_info) {
3502                 ret = update_space_info(extent_root->fs_info, flags,
3503                                         0, 0, &space_info);
3504                 BUG_ON(ret); /* -ENOMEM */
3505         }
3506         BUG_ON(!space_info); /* Logic error */
3507
3508 again:
3509         spin_lock(&space_info->lock);
3510         if (force < space_info->force_alloc)
3511                 force = space_info->force_alloc;
3512         if (space_info->full) {
3513                 spin_unlock(&space_info->lock);
3514                 return 0;
3515         }
3516
3517         if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3518                 spin_unlock(&space_info->lock);
3519                 return 0;
3520         } else if (space_info->chunk_alloc) {
3521                 wait_for_alloc = 1;
3522         } else {
3523                 space_info->chunk_alloc = 1;
3524         }
3525
3526         spin_unlock(&space_info->lock);
3527
3528         mutex_lock(&fs_info->chunk_mutex);
3529
3530         /*
3531          * The chunk_mutex is held throughout the entirety of a chunk
3532          * allocation, so once we've acquired the chunk_mutex we know that the
3533          * other guy is done and we need to recheck and see if we should
3534          * allocate.
3535          */
3536         if (wait_for_alloc) {
3537                 mutex_unlock(&fs_info->chunk_mutex);
3538                 wait_for_alloc = 0;
3539                 goto again;
3540         }
3541
3542         /*
3543          * If we have mixed data/metadata chunks we want to make sure we keep
3544          * allocating mixed chunks instead of individual chunks.
3545          */
3546         if (btrfs_mixed_space_info(space_info))
3547                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3548
3549         /*
3550          * if we're doing a data chunk, go ahead and make sure that
3551          * we keep a reasonable number of metadata chunks allocated in the
3552          * FS as well.
3553          */
3554         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3555                 fs_info->data_chunk_allocations++;
3556                 if (!(fs_info->data_chunk_allocations %
3557                       fs_info->metadata_ratio))
3558                         force_metadata_allocation(fs_info);
3559         }
3560
3561         /*
3562          * Check if we have enough space in SYSTEM chunk because we may need
3563          * to update devices.
3564          */
3565         check_system_chunk(trans, extent_root, flags);
3566
3567         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3568         if (ret < 0 && ret != -ENOSPC)
3569                 goto out;
3570
3571         spin_lock(&space_info->lock);
3572         if (ret)
3573                 space_info->full = 1;
3574         else
3575                 ret = 1;
3576
3577         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3578         space_info->chunk_alloc = 0;
3579         spin_unlock(&space_info->lock);
3580 out:
3581         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3582         return ret;
3583 }
3584
3585 /*
3586  * shrink metadata reservation for delalloc
3587  */
3588 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3589                            bool wait_ordered)
3590 {
3591         struct btrfs_block_rsv *block_rsv;
3592         struct btrfs_space_info *space_info;
3593         struct btrfs_trans_handle *trans;
3594         u64 reserved;
3595         u64 max_reclaim;
3596         u64 reclaimed = 0;
3597         long time_left;
3598         unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3599         int loops = 0;
3600         unsigned long progress;
3601
3602         trans = (struct btrfs_trans_handle *)current->journal_info;
3603         block_rsv = &root->fs_info->delalloc_block_rsv;
3604         space_info = block_rsv->space_info;
3605
3606         smp_mb();
3607         reserved = space_info->bytes_may_use;
3608         progress = space_info->reservation_progress;
3609
3610         if (reserved == 0)
3611                 return 0;
3612
3613         smp_mb();
3614         if (root->fs_info->delalloc_bytes == 0) {
3615                 if (trans)
3616                         return 0;
3617                 btrfs_wait_ordered_extents(root, 0, 0);
3618                 return 0;
3619         }
3620
3621         max_reclaim = min(reserved, to_reclaim);
3622         nr_pages = max_t(unsigned long, nr_pages,
3623                          max_reclaim >> PAGE_CACHE_SHIFT);
3624         while (loops < 1024) {
3625                 /* have the flusher threads jump in and do some IO */
3626                 smp_mb();
3627                 nr_pages = min_t(unsigned long, nr_pages,
3628                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3629                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3630                                                 WB_REASON_FS_FREE_SPACE);
3631
3632                 spin_lock(&space_info->lock);
3633                 if (reserved > space_info->bytes_may_use)
3634                         reclaimed += reserved - space_info->bytes_may_use;
3635                 reserved = space_info->bytes_may_use;
3636                 spin_unlock(&space_info->lock);
3637
3638                 loops++;
3639
3640                 if (reserved == 0 || reclaimed >= max_reclaim)
3641                         break;
3642
3643                 if (trans && trans->transaction->blocked)
3644                         return -EAGAIN;
3645
3646                 if (wait_ordered && !trans) {
3647                         btrfs_wait_ordered_extents(root, 0, 0);
3648                 } else {
3649                         time_left = schedule_timeout_interruptible(1);
3650
3651                         /* We were interrupted, exit */
3652                         if (time_left)
3653                                 break;
3654                 }
3655
3656                 /* we've kicked the IO a few times, if anything has been freed,
3657                  * exit.  There is no sense in looping here for a long time
3658                  * when we really need to commit the transaction, or there are
3659                  * just too many writers without enough free space
3660                  */
3661
3662                 if (loops > 3) {
3663                         smp_mb();
3664                         if (progress != space_info->reservation_progress)
3665                                 break;
3666                 }
3667
3668         }
3669
3670         return reclaimed >= to_reclaim;
3671 }
3672
3673 /**
3674  * maybe_commit_transaction - possibly commit the transaction if its ok to
3675  * @root - the root we're allocating for
3676  * @bytes - the number of bytes we want to reserve
3677  * @force - force the commit
3678  *
3679  * This will check to make sure that committing the transaction will actually
3680  * get us somewhere and then commit the transaction if it does.  Otherwise it
3681  * will return -ENOSPC.
3682  */
3683 static int may_commit_transaction(struct btrfs_root *root,
3684                                   struct btrfs_space_info *space_info,
3685                                   u64 bytes, int force)
3686 {
3687         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3688         struct btrfs_trans_handle *trans;
3689
3690         trans = (struct btrfs_trans_handle *)current->journal_info;
3691         if (trans)
3692                 return -EAGAIN;
3693
3694         if (force)
3695                 goto commit;
3696
3697         /* See if there is enough pinned space to make this reservation */
3698         spin_lock(&space_info->lock);
3699         if (space_info->bytes_pinned >= bytes) {
3700                 spin_unlock(&space_info->lock);
3701                 goto commit;
3702         }
3703         spin_unlock(&space_info->lock);
3704
3705         /*
3706          * See if there is some space in the delayed insertion reservation for
3707          * this reservation.
3708          */
3709         if (space_info != delayed_rsv->space_info)
3710                 return -ENOSPC;
3711
3712         spin_lock(&space_info->lock);
3713         spin_lock(&delayed_rsv->lock);
3714         if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3715                 spin_unlock(&delayed_rsv->lock);
3716                 spin_unlock(&space_info->lock);
3717                 return -ENOSPC;
3718         }
3719         spin_unlock(&delayed_rsv->lock);
3720         spin_unlock(&space_info->lock);
3721
3722 commit:
3723         trans = btrfs_join_transaction(root);
3724         if (IS_ERR(trans))
3725                 return -ENOSPC;
3726
3727         return btrfs_commit_transaction(trans, root);
3728 }
3729
3730 /**
3731  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3732  * @root - the root we're allocating for
3733  * @block_rsv - the block_rsv we're allocating for
3734  * @orig_bytes - the number of bytes we want
3735  * @flush - wether or not we can flush to make our reservation
3736  *
3737  * This will reserve orgi_bytes number of bytes from the space info associated
3738  * with the block_rsv.  If there is not enough space it will make an attempt to
3739  * flush out space to make room.  It will do this by flushing delalloc if
3740  * possible or committing the transaction.  If flush is 0 then no attempts to
3741  * regain reservations will be made and this will fail if there is not enough
3742  * space already.
3743  */
3744 static int reserve_metadata_bytes(struct btrfs_root *root,
3745                                   struct btrfs_block_rsv *block_rsv,
3746                                   u64 orig_bytes, int flush)
3747 {
3748         struct btrfs_space_info *space_info = block_rsv->space_info;
3749         u64 used;
3750         u64 num_bytes = orig_bytes;
3751         int retries = 0;
3752         int ret = 0;
3753         bool committed = false;
3754         bool flushing = false;
3755         bool wait_ordered = false;
3756
3757 again:
3758         ret = 0;
3759         spin_lock(&space_info->lock);
3760         /*
3761          * We only want to wait if somebody other than us is flushing and we are
3762          * actually alloed to flush.
3763          */
3764         while (flush && !flushing && space_info->flush) {
3765                 spin_unlock(&space_info->lock);
3766                 /*
3767                  * If we have a trans handle we can't wait because the flusher
3768                  * may have to commit the transaction, which would mean we would
3769                  * deadlock since we are waiting for the flusher to finish, but
3770                  * hold the current transaction open.
3771                  */
3772                 if (current->journal_info)
3773                         return -EAGAIN;
3774                 ret = wait_event_killable(space_info->wait, !space_info->flush);
3775                 /* Must have been killed, return */
3776                 if (ret)
3777                         return -EINTR;
3778
3779                 spin_lock(&space_info->lock);
3780         }
3781
3782         ret = -ENOSPC;
3783         used = space_info->bytes_used + space_info->bytes_reserved +
3784                 space_info->bytes_pinned + space_info->bytes_readonly +
3785                 space_info->bytes_may_use;
3786
3787         /*
3788          * The idea here is that we've not already over-reserved the block group
3789          * then we can go ahead and save our reservation first and then start
3790          * flushing if we need to.  Otherwise if we've already overcommitted
3791          * lets start flushing stuff first and then come back and try to make
3792          * our reservation.
3793          */
3794         if (used <= space_info->total_bytes) {
3795                 if (used + orig_bytes <= space_info->total_bytes) {
3796                         space_info->bytes_may_use += orig_bytes;
3797                         trace_btrfs_space_reservation(root->fs_info,
3798                                 "space_info", space_info->flags, orig_bytes, 1);
3799                         ret = 0;
3800                 } else {
3801                         /*
3802                          * Ok set num_bytes to orig_bytes since we aren't
3803                          * overocmmitted, this way we only try and reclaim what
3804                          * we need.
3805                          */
3806                         num_bytes = orig_bytes;
3807                 }
3808         } else {
3809                 /*
3810                  * Ok we're over committed, set num_bytes to the overcommitted
3811                  * amount plus the amount of bytes that we need for this
3812                  * reservation.
3813                  */
3814                 wait_ordered = true;
3815                 num_bytes = used - space_info->total_bytes +
3816                         (orig_bytes * (retries + 1));
3817         }
3818
3819         if (ret) {
3820                 u64 profile = btrfs_get_alloc_profile(root, 0);
3821                 u64 avail;
3822
3823                 /*
3824                  * If we have a lot of space that's pinned, don't bother doing
3825                  * the overcommit dance yet and just commit the transaction.
3826                  */
3827                 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3828                 do_div(avail, 10);
3829                 if (space_info->bytes_pinned >= avail && flush && !committed) {
3830                         space_info->flush = 1;
3831                         flushing = true;
3832                         spin_unlock(&space_info->lock);
3833                         ret = may_commit_transaction(root, space_info,
3834                                                      orig_bytes, 1);
3835                         if (ret)
3836                                 goto out;
3837                         committed = true;
3838                         goto again;
3839                 }
3840
3841                 spin_lock(&root->fs_info->free_chunk_lock);
3842                 avail = root->fs_info->free_chunk_space;
3843
3844                 /*
3845                  * If we have dup, raid1 or raid10 then only half of the free
3846                  * space is actually useable.
3847                  */
3848                 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3849                                BTRFS_BLOCK_GROUP_RAID1 |
3850                                BTRFS_BLOCK_GROUP_RAID10))
3851                         avail >>= 1;
3852
3853                 /*
3854                  * If we aren't flushing don't let us overcommit too much, say
3855                  * 1/8th of the space.  If we can flush, let it overcommit up to
3856                  * 1/2 of the space.
3857                  */
3858                 if (flush)
3859                         avail >>= 3;
3860                 else
3861                         avail >>= 1;
3862                  spin_unlock(&root->fs_info->free_chunk_lock);
3863
3864                 if (used + num_bytes < space_info->total_bytes + avail) {
3865                         space_info->bytes_may_use += orig_bytes;
3866                         trace_btrfs_space_reservation(root->fs_info,
3867                                 "space_info", space_info->flags, orig_bytes, 1);
3868                         ret = 0;
3869                 } else {
3870                         wait_ordered = true;
3871                 }
3872         }
3873
3874         /*
3875          * Couldn't make our reservation, save our place so while we're trying
3876          * to reclaim space we can actually use it instead of somebody else
3877          * stealing it from us.
3878          */
3879         if (ret && flush) {
3880                 flushing = true;
3881                 space_info->flush = 1;
3882         }
3883
3884         spin_unlock(&space_info->lock);
3885
3886         if (!ret || !flush)
3887                 goto out;
3888
3889         /*
3890          * We do synchronous shrinking since we don't actually unreserve
3891          * metadata until after the IO is completed.
3892          */
3893         ret = shrink_delalloc(root, num_bytes, wait_ordered);
3894         if (ret < 0)
3895                 goto out;
3896
3897         ret = 0;
3898
3899         /*
3900          * So if we were overcommitted it's possible that somebody else flushed
3901          * out enough space and we simply didn't have enough space to reclaim,
3902          * so go back around and try again.
3903          */
3904         if (retries < 2) {
3905                 wait_ordered = true;
3906                 retries++;
3907                 goto again;
3908         }
3909
3910         ret = -ENOSPC;
3911         if (committed)
3912                 goto out;
3913
3914         ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3915         if (!ret) {
3916                 committed = true;
3917                 goto again;
3918         }
3919
3920 out:
3921         if (flushing) {
3922                 spin_lock(&space_info->lock);
3923                 space_info->flush = 0;
3924                 wake_up_all(&space_info->wait);
3925                 spin_unlock(&space_info->lock);
3926         }
3927         return ret;
3928 }
3929
3930 static struct btrfs_block_rsv *get_block_rsv(
3931                                         const struct btrfs_trans_handle *trans,
3932                                         const struct btrfs_root *root)
3933 {
3934         struct btrfs_block_rsv *block_rsv = NULL;
3935
3936         if (root->ref_cows || root == root->fs_info->csum_root)
3937                 block_rsv = trans->block_rsv;
3938
3939         if (!block_rsv)
3940                 block_rsv = root->block_rsv;
3941
3942         if (!block_rsv)
3943                 block_rsv = &root->fs_info->empty_block_rsv;
3944
3945         return block_rsv;
3946 }
3947
3948 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3949                                u64 num_bytes)
3950 {
3951         int ret = -ENOSPC;
3952         spin_lock(&block_rsv->lock);
3953         if (block_rsv->reserved >= num_bytes) {
3954                 block_rsv->reserved -= num_bytes;
3955                 if (block_rsv->reserved < block_rsv->size)
3956                         block_rsv->full = 0;
3957                 ret = 0;
3958         }
3959         spin_unlock(&block_rsv->lock);
3960         return ret;
3961 }
3962
3963 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3964                                 u64 num_bytes, int update_size)
3965 {
3966         spin_lock(&block_rsv->lock);
3967         block_rsv->reserved += num_bytes;
3968         if (update_size)
3969                 block_rsv->size += num_bytes;
3970         else if (block_rsv->reserved >= block_rsv->size)
3971                 block_rsv->full = 1;
3972         spin_unlock(&block_rsv->lock);
3973 }
3974
3975 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
3976                                     struct btrfs_block_rsv *block_rsv,
3977                                     struct btrfs_block_rsv *dest, u64 num_bytes)
3978 {
3979         struct btrfs_space_info *space_info = block_rsv->space_info;
3980
3981         spin_lock(&block_rsv->lock);
3982         if (num_bytes == (u64)-1)
3983                 num_bytes = block_rsv->size;
3984         block_rsv->size -= num_bytes;
3985         if (block_rsv->reserved >= block_rsv->size) {
3986                 num_bytes = block_rsv->reserved - block_rsv->size;
3987                 block_rsv->reserved = block_rsv->size;
3988                 block_rsv->full = 1;
3989         } else {
3990                 num_bytes = 0;
3991         }
3992         spin_unlock(&block_rsv->lock);
3993
3994         if (num_bytes > 0) {
3995                 if (dest) {
3996                         spin_lock(&dest->lock);
3997                         if (!dest->full) {
3998                                 u64 bytes_to_add;
3999
4000                                 bytes_to_add = dest->size - dest->reserved;
4001                                 bytes_to_add = min(num_bytes, bytes_to_add);
4002                                 dest->reserved += bytes_to_add;
4003                                 if (dest->reserved >= dest->size)
4004                                         dest->full = 1;
4005                                 num_bytes -= bytes_to_add;
4006                         }
4007                         spin_unlock(&dest->lock);
4008                 }
4009                 if (num_bytes) {
4010                         spin_lock(&space_info->lock);
4011                         space_info->bytes_may_use -= num_bytes;
4012                         trace_btrfs_space_reservation(fs_info, "space_info",
4013                                         space_info->flags, num_bytes, 0);
4014                         space_info->reservation_progress++;
4015                         spin_unlock(&space_info->lock);
4016                 }
4017         }
4018 }
4019
4020 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4021                                    struct btrfs_block_rsv *dst, u64 num_bytes)
4022 {
4023         int ret;
4024
4025         ret = block_rsv_use_bytes(src, num_bytes);
4026         if (ret)
4027                 return ret;
4028
4029         block_rsv_add_bytes(dst, num_bytes, 1);
4030         return 0;
4031 }
4032
4033 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
4034 {
4035         memset(rsv, 0, sizeof(*rsv));
4036         spin_lock_init(&rsv->lock);
4037 }
4038
4039 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
4040 {
4041         struct btrfs_block_rsv *block_rsv;
4042         struct btrfs_fs_info *fs_info = root->fs_info;
4043
4044         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4045         if (!block_rsv)
4046                 return NULL;
4047
4048         btrfs_init_block_rsv(block_rsv);
4049         block_rsv->space_info = __find_space_info(fs_info,
4050                                                   BTRFS_BLOCK_GROUP_METADATA);
4051         return block_rsv;
4052 }
4053
4054 void btrfs_free_block_rsv(struct btrfs_root *root,
4055                           struct btrfs_block_rsv *rsv)
4056 {
4057         btrfs_block_rsv_release(root, rsv, (u64)-1);
4058         kfree(rsv);
4059 }
4060
4061 static inline int __block_rsv_add(struct btrfs_root *root,
4062                                   struct btrfs_block_rsv *block_rsv,
4063                                   u64 num_bytes, int flush)
4064 {
4065         int ret;
4066
4067         if (num_bytes == 0)
4068                 return 0;
4069
4070         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4071         if (!ret) {
4072                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4073                 return 0;
4074         }
4075
4076         return ret;
4077 }
4078
4079 int btrfs_block_rsv_add(struct btrfs_root *root,
4080                         struct btrfs_block_rsv *block_rsv,
4081                         u64 num_bytes)
4082 {
4083         return __block_rsv_add(root, block_rsv, num_bytes, 1);
4084 }
4085
4086 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
4087                                 struct btrfs_block_rsv *block_rsv,
4088                                 u64 num_bytes)
4089 {
4090         return __block_rsv_add(root, block_rsv, num_bytes, 0);
4091 }
4092
4093 int btrfs_block_rsv_check(struct btrfs_root *root,
4094                           struct btrfs_block_rsv *block_rsv, int min_factor)
4095 {
4096         u64 num_bytes = 0;
4097         int ret = -ENOSPC;
4098
4099         if (!block_rsv)
4100                 return 0;
4101
4102         spin_lock(&block_rsv->lock);
4103         num_bytes = div_factor(block_rsv->size, min_factor);
4104         if (block_rsv->reserved >= num_bytes)
4105                 ret = 0;
4106         spin_unlock(&block_rsv->lock);
4107
4108         return ret;
4109 }
4110
4111 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4112                                            struct btrfs_block_rsv *block_rsv,
4113                                            u64 min_reserved, int flush)
4114 {
4115         u64 num_bytes = 0;
4116         int ret = -ENOSPC;
4117
4118         if (!block_rsv)
4119                 return 0;
4120
4121         spin_lock(&block_rsv->lock);
4122         num_bytes = min_reserved;
4123         if (block_rsv->reserved >= num_bytes)
4124                 ret = 0;
4125         else
4126                 num_bytes -= block_rsv->reserved;
4127         spin_unlock(&block_rsv->lock);
4128
4129         if (!ret)
4130                 return 0;
4131
4132         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4133         if (!ret) {
4134                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4135                 return 0;
4136         }
4137
4138         return ret;
4139 }
4140
4141 int btrfs_block_rsv_refill(struct btrfs_root *root,
4142                            struct btrfs_block_rsv *block_rsv,
4143                            u64 min_reserved)
4144 {
4145         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4146 }
4147
4148 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4149                                    struct btrfs_block_rsv *block_rsv,
4150                                    u64 min_reserved)
4151 {
4152         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4153 }
4154
4155 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4156                             struct btrfs_block_rsv *dst_rsv,
4157                             u64 num_bytes)
4158 {
4159         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4160 }
4161
4162 void btrfs_block_rsv_release(struct btrfs_root *root,
4163                              struct btrfs_block_rsv *block_rsv,
4164                              u64 num_bytes)
4165 {
4166         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4167         if (global_rsv->full || global_rsv == block_rsv ||
4168             block_rsv->space_info != global_rsv->space_info)
4169                 global_rsv = NULL;
4170         block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4171                                 num_bytes);
4172 }
4173
4174 /*
4175  * helper to calculate size of global block reservation.
4176  * the desired value is sum of space used by extent tree,
4177  * checksum tree and root tree
4178  */
4179 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4180 {
4181         struct btrfs_space_info *sinfo;
4182         u64 num_bytes;
4183         u64 meta_used;
4184         u64 data_used;
4185         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4186
4187         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4188         spin_lock(&sinfo->lock);
4189         data_used = sinfo->bytes_used;
4190         spin_unlock(&sinfo->lock);
4191
4192         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4193         spin_lock(&sinfo->lock);
4194         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4195                 data_used = 0;
4196         meta_used = sinfo->bytes_used;
4197         spin_unlock(&sinfo->lock);
4198
4199         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4200                     csum_size * 2;
4201         num_bytes += div64_u64(data_used + meta_used, 50);
4202
4203         if (num_bytes * 3 > meta_used)
4204                 num_bytes = div64_u64(meta_used, 3);
4205
4206         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4207 }
4208
4209 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4210 {
4211         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4212         struct btrfs_space_info *sinfo = block_rsv->space_info;
4213         u64 num_bytes;
4214
4215         num_bytes = calc_global_metadata_size(fs_info);
4216
4217         spin_lock(&sinfo->lock);
4218         spin_lock(&block_rsv->lock);
4219
4220         block_rsv->size = num_bytes;
4221
4222         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4223                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4224                     sinfo->bytes_may_use;
4225
4226         if (sinfo->total_bytes > num_bytes) {
4227                 num_bytes = sinfo->total_bytes - num_bytes;
4228                 block_rsv->reserved += num_bytes;
4229                 sinfo->bytes_may_use += num_bytes;
4230                 trace_btrfs_space_reservation(fs_info, "space_info",
4231                                       sinfo->flags, num_bytes, 1);
4232         }
4233
4234         if (block_rsv->reserved >= block_rsv->size) {
4235                 num_bytes = block_rsv->reserved - block_rsv->size;
4236                 sinfo->bytes_may_use -= num_bytes;
4237                 trace_btrfs_space_reservation(fs_info, "space_info",
4238                                       sinfo->flags, num_bytes, 0);
4239                 sinfo->reservation_progress++;
4240                 block_rsv->reserved = block_rsv->size;
4241                 block_rsv->full = 1;
4242         }
4243
4244         spin_unlock(&block_rsv->lock);
4245         spin_unlock(&sinfo->lock);
4246 }
4247
4248 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4249 {
4250         struct btrfs_space_info *space_info;
4251
4252         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4253         fs_info->chunk_block_rsv.space_info = space_info;
4254
4255         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4256         fs_info->global_block_rsv.space_info = space_info;
4257         fs_info->delalloc_block_rsv.space_info = space_info;
4258         fs_info->trans_block_rsv.space_info = space_info;
4259         fs_info->empty_block_rsv.space_info = space_info;
4260         fs_info->delayed_block_rsv.space_info = space_info;
4261
4262         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4263         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4264         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4265         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4266         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4267
4268         update_global_block_rsv(fs_info);
4269 }
4270
4271 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4272 {
4273         block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4274                                 (u64)-1);
4275         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4276         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4277         WARN_ON(fs_info->trans_block_rsv.size > 0);
4278         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4279         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4280         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4281         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4282         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4283 }
4284
4285 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4286                                   struct btrfs_root *root)
4287 {
4288         if (!trans->bytes_reserved)
4289                 return;
4290
4291         trace_btrfs_space_reservation(root->fs_info, "transaction",
4292                                       trans->transid, trans->bytes_reserved, 0);
4293         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4294         trans->bytes_reserved = 0;
4295 }
4296
4297 /* Can only return 0 or -ENOSPC */
4298 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4299                                   struct inode *inode)
4300 {
4301         struct btrfs_root *root = BTRFS_I(inode)->root;
4302         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4303         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4304
4305         /*
4306          * We need to hold space in order to delete our orphan item once we've
4307          * added it, so this takes the reservation so we can release it later
4308          * when we are truly done with the orphan item.
4309          */
4310         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4311         trace_btrfs_space_reservation(root->fs_info, "orphan",
4312                                       btrfs_ino(inode), num_bytes, 1);
4313         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4314 }
4315
4316 void btrfs_orphan_release_metadata(struct inode *inode)
4317 {
4318         struct btrfs_root *root = BTRFS_I(inode)->root;
4319         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4320         trace_btrfs_space_reservation(root->fs_info, "orphan",
4321                                       btrfs_ino(inode), num_bytes, 0);
4322         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4323 }
4324
4325 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4326                                 struct btrfs_pending_snapshot *pending)
4327 {
4328         struct btrfs_root *root = pending->root;
4329         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4330         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4331         /*
4332          * two for root back/forward refs, two for directory entries
4333          * and one for root of the snapshot.
4334          */
4335         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4336         dst_rsv->space_info = src_rsv->space_info;
4337         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4338 }
4339
4340 /**
4341  * drop_outstanding_extent - drop an outstanding extent
4342  * @inode: the inode we're dropping the extent for
4343  *
4344  * This is called when we are freeing up an outstanding extent, either called
4345  * after an error or after an extent is written.  This will return the number of
4346  * reserved extents that need to be freed.  This must be called with
4347  * BTRFS_I(inode)->lock held.
4348  */
4349 static unsigned drop_outstanding_extent(struct inode *inode)
4350 {
4351         unsigned drop_inode_space = 0;
4352         unsigned dropped_extents = 0;
4353
4354         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4355         BTRFS_I(inode)->outstanding_extents--;
4356
4357         if (BTRFS_I(inode)->outstanding_extents == 0 &&
4358             BTRFS_I(inode)->delalloc_meta_reserved) {
4359                 drop_inode_space = 1;
4360                 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4361         }
4362
4363         /*
4364          * If we have more or the same amount of outsanding extents than we have
4365          * reserved then we need to leave the reserved extents count alone.
4366          */
4367         if (BTRFS_I(inode)->outstanding_extents >=
4368             BTRFS_I(inode)->reserved_extents)
4369                 return drop_inode_space;
4370
4371         dropped_extents = BTRFS_I(inode)->reserved_extents -
4372                 BTRFS_I(inode)->outstanding_extents;
4373         BTRFS_I(inode)->reserved_extents -= dropped_extents;
4374         return dropped_extents + drop_inode_space;
4375 }
4376
4377 /**
4378  * calc_csum_metadata_size - return the amount of metada space that must be
4379  *      reserved/free'd for the given bytes.
4380  * @inode: the inode we're manipulating
4381  * @num_bytes: the number of bytes in question
4382  * @reserve: 1 if we are reserving space, 0 if we are freeing space
4383  *
4384  * This adjusts the number of csum_bytes in the inode and then returns the
4385  * correct amount of metadata that must either be reserved or freed.  We
4386  * calculate how many checksums we can fit into one leaf and then divide the
4387  * number of bytes that will need to be checksumed by this value to figure out
4388  * how many checksums will be required.  If we are adding bytes then the number
4389  * may go up and we will return the number of additional bytes that must be
4390  * reserved.  If it is going down we will return the number of bytes that must
4391  * be freed.
4392  *
4393  * This must be called with BTRFS_I(inode)->lock held.
4394  */
4395 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4396                                    int reserve)
4397 {
4398         struct btrfs_root *root = BTRFS_I(inode)->root;
4399         u64 csum_size;
4400         int num_csums_per_leaf;
4401         int num_csums;
4402         int old_csums;
4403
4404         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4405             BTRFS_I(inode)->csum_bytes == 0)
4406                 return 0;
4407
4408         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4409         if (reserve)
4410                 BTRFS_I(inode)->csum_bytes += num_bytes;
4411         else
4412                 BTRFS_I(inode)->csum_bytes -= num_bytes;
4413         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4414         num_csums_per_leaf = (int)div64_u64(csum_size,
4415                                             sizeof(struct btrfs_csum_item) +
4416                                             sizeof(struct btrfs_disk_key));
4417         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4418         num_csums = num_csums + num_csums_per_leaf - 1;
4419         num_csums = num_csums / num_csums_per_leaf;
4420
4421         old_csums = old_csums + num_csums_per_leaf - 1;
4422         old_csums = old_csums / num_csums_per_leaf;
4423
4424         /* No change, no need to reserve more */
4425         if (old_csums == num_csums)
4426                 return 0;
4427
4428         if (reserve)
4429                 return btrfs_calc_trans_metadata_size(root,
4430                                                       num_csums - old_csums);
4431
4432         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4433 }
4434
4435 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4436 {
4437         struct btrfs_root *root = BTRFS_I(inode)->root;
4438         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4439         u64 to_reserve = 0;
4440         u64 csum_bytes;
4441         unsigned nr_extents = 0;
4442         int extra_reserve = 0;
4443         int flush = 1;
4444         int ret;
4445
4446         /* Need to be holding the i_mutex here if we aren't free space cache */
4447         if (btrfs_is_free_space_inode(root, inode))
4448                 flush = 0;
4449
4450         if (flush && btrfs_transaction_in_commit(root->fs_info))
4451                 schedule_timeout(1);
4452
4453         mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4454         num_bytes = ALIGN(num_bytes, root->sectorsize);
4455
4456         spin_lock(&BTRFS_I(inode)->lock);
4457         BTRFS_I(inode)->outstanding_extents++;
4458
4459         if (BTRFS_I(inode)->outstanding_extents >
4460             BTRFS_I(inode)->reserved_extents)
4461                 nr_extents = BTRFS_I(inode)->outstanding_extents -
4462                         BTRFS_I(inode)->reserved_extents;
4463
4464         /*
4465          * Add an item to reserve for updating the inode when we complete the
4466          * delalloc io.
4467          */
4468         if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4469                 nr_extents++;
4470                 extra_reserve = 1;
4471         }
4472
4473         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4474         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4475         csum_bytes = BTRFS_I(inode)->csum_bytes;
4476         spin_unlock(&BTRFS_I(inode)->lock);
4477
4478         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4479         if (ret) {
4480                 u64 to_free = 0;
4481                 unsigned dropped;
4482
4483                 spin_lock(&BTRFS_I(inode)->lock);
4484                 dropped = drop_outstanding_extent(inode);
4485                 /*
4486                  * If the inodes csum_bytes is the same as the original
4487                  * csum_bytes then we know we haven't raced with any free()ers
4488                  * so we can just reduce our inodes csum bytes and carry on.
4489                  * Otherwise we have to do the normal free thing to account for
4490                  * the case that the free side didn't free up its reserve
4491                  * because of this outstanding reservation.
4492                  */
4493                 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4494                         calc_csum_metadata_size(inode, num_bytes, 0);
4495                 else
4496                         to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4497                 spin_unlock(&BTRFS_I(inode)->lock);
4498                 if (dropped)
4499                         to_free += btrfs_calc_trans_metadata_size(root, dropped);
4500
4501                 if (to_free) {
4502                         btrfs_block_rsv_release(root, block_rsv, to_free);
4503                         trace_btrfs_space_reservation(root->fs_info,
4504                                                       "delalloc",
4505                                                       btrfs_ino(inode),
4506                                                       to_free, 0);
4507                 }
4508                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4509                 return ret;
4510         }
4511
4512         spin_lock(&BTRFS_I(inode)->lock);
4513         if (extra_reserve) {
4514                 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4515                 nr_extents--;
4516         }
4517         BTRFS_I(inode)->reserved_extents += nr_extents;
4518         spin_unlock(&BTRFS_I(inode)->lock);
4519         mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4520
4521         if (to_reserve)
4522                 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4523                                               btrfs_ino(inode), to_reserve, 1);
4524         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4525
4526         return 0;
4527 }
4528
4529 /**
4530  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4531  * @inode: the inode to release the reservation for
4532  * @num_bytes: the number of bytes we're releasing
4533  *
4534  * This will release the metadata reservation for an inode.  This can be called
4535  * once we complete IO for a given set of bytes to release their metadata
4536  * reservations.
4537  */
4538 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4539 {
4540         struct btrfs_root *root = BTRFS_I(inode)->root;
4541         u64 to_free = 0;
4542         unsigned dropped;
4543
4544         num_bytes = ALIGN(num_bytes, root->sectorsize);
4545         spin_lock(&BTRFS_I(inode)->lock);
4546         dropped = drop_outstanding_extent(inode);
4547
4548         to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4549         spin_unlock(&BTRFS_I(inode)->lock);
4550         if (dropped > 0)
4551                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4552
4553         trace_btrfs_space_reservation(root->fs_info, "delalloc",
4554                                       btrfs_ino(inode), to_free, 0);
4555         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4556                                 to_free);
4557 }
4558
4559 /**
4560  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4561  * @inode: inode we're writing to
4562  * @num_bytes: the number of bytes we want to allocate
4563  *
4564  * This will do the following things
4565  *
4566  * o reserve space in the data space info for num_bytes
4567  * o reserve space in the metadata space info based on number of outstanding
4568  *   extents and how much csums will be needed
4569  * o add to the inodes ->delalloc_bytes
4570  * o add it to the fs_info's delalloc inodes list.
4571  *
4572  * This will return 0 for success and -ENOSPC if there is no space left.
4573  */
4574 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4575 {
4576         int ret;
4577
4578         ret = btrfs_check_data_free_space(inode, num_bytes);
4579         if (ret)
4580                 return ret;
4581
4582         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4583         if (ret) {
4584                 btrfs_free_reserved_data_space(inode, num_bytes);
4585                 return ret;
4586         }
4587
4588         return 0;
4589 }
4590
4591 /**
4592  * btrfs_delalloc_release_space - release data and metadata space for delalloc
4593  * @inode: inode we're releasing space for
4594  * @num_bytes: the number of bytes we want to free up
4595  *
4596  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
4597  * called in the case that we don't need the metadata AND data reservations
4598  * anymore.  So if there is an error or we insert an inline extent.
4599  *
4600  * This function will release the metadata space that was not used and will
4601  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4602  * list if there are no delalloc bytes left.
4603  */
4604 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4605 {
4606         btrfs_delalloc_release_metadata(inode, num_bytes);
4607         btrfs_free_reserved_data_space(inode, num_bytes);
4608 }
4609
4610 static int update_block_group(struct btrfs_trans_handle *trans,
4611                               struct btrfs_root *root,
4612                               u64 bytenr, u64 num_bytes, int alloc)
4613 {
4614         struct btrfs_block_group_cache *cache = NULL;
4615         struct btrfs_fs_info *info = root->fs_info;
4616         u64 total = num_bytes;
4617         u64 old_val;
4618         u64 byte_in_group;
4619         int factor;
4620
4621         /* block accounting for super block */
4622         spin_lock(&info->delalloc_lock);
4623         old_val = btrfs_super_bytes_used(info->super_copy);
4624         if (alloc)
4625                 old_val += num_bytes;
4626         else
4627                 old_val -= num_bytes;
4628         btrfs_set_super_bytes_used(info->super_copy, old_val);
4629         spin_unlock(&info->delalloc_lock);
4630
4631         while (total) {
4632                 cache = btrfs_lookup_block_group(info, bytenr);
4633                 if (!cache)
4634                         return -ENOENT;
4635                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4636                                     BTRFS_BLOCK_GROUP_RAID1 |
4637                                     BTRFS_BLOCK_GROUP_RAID10))
4638                         factor = 2;
4639                 else
4640                         factor = 1;
4641                 /*
4642                  * If this block group has free space cache written out, we
4643                  * need to make sure to load it if we are removing space.  This
4644                  * is because we need the unpinning stage to actually add the
4645                  * space back to the block group, otherwise we will leak space.
4646                  */
4647                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4648                         cache_block_group(cache, trans, NULL, 1);
4649
4650                 byte_in_group = bytenr - cache->key.objectid;
4651                 WARN_ON(byte_in_group > cache->key.offset);
4652
4653                 spin_lock(&cache->space_info->lock);
4654                 spin_lock(&cache->lock);
4655
4656                 if (btrfs_test_opt(root, SPACE_CACHE) &&
4657                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4658                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4659
4660                 cache->dirty = 1;
4661                 old_val = btrfs_block_group_used(&cache->item);
4662                 num_bytes = min(total, cache->key.offset - byte_in_group);
4663                 if (alloc) {
4664                         old_val += num_bytes;
4665                         btrfs_set_block_group_used(&cache->item, old_val);
4666                         cache->reserved -= num_bytes;
4667                         cache->space_info->bytes_reserved -= num_bytes;
4668                         cache->space_info->bytes_used += num_bytes;
4669                         cache->space_info->disk_used += num_bytes * factor;
4670                         spin_unlock(&cache->lock);
4671                         spin_unlock(&cache->space_info->lock);
4672                 } else {
4673                         old_val -= num_bytes;
4674                         btrfs_set_block_group_used(&cache->item, old_val);
4675                         cache->pinned += num_bytes;
4676                         cache->space_info->bytes_pinned += num_bytes;
4677                         cache->space_info->bytes_used -= num_bytes;
4678                         cache->space_info->disk_used -= num_bytes * factor;
4679                         spin_unlock(&cache->lock);
4680                         spin_unlock(&cache->space_info->lock);
4681
4682                         set_extent_dirty(info->pinned_extents,
4683                                          bytenr, bytenr + num_bytes - 1,
4684                                          GFP_NOFS | __GFP_NOFAIL);
4685                 }
4686                 btrfs_put_block_group(cache);
4687                 total -= num_bytes;
4688                 bytenr += num_bytes;
4689         }
4690         return 0;
4691 }
4692
4693 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4694 {
4695         struct btrfs_block_group_cache *cache;
4696         u64 bytenr;
4697
4698         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4699         if (!cache)
4700                 return 0;
4701
4702         bytenr = cache->key.objectid;
4703         btrfs_put_block_group(cache);
4704
4705         return bytenr;
4706 }
4707
4708 static int pin_down_extent(struct btrfs_root *root,
4709                            struct btrfs_block_group_cache *cache,
4710                            u64 bytenr, u64 num_bytes, int reserved)
4711 {
4712         spin_lock(&cache->space_info->lock);
4713         spin_lock(&cache->lock);
4714         cache->pinned += num_bytes;
4715         cache->space_info->bytes_pinned += num_bytes;
4716         if (reserved) {
4717                 cache->reserved -= num_bytes;
4718                 cache->space_info->bytes_reserved -= num_bytes;
4719         }
4720         spin_unlock(&cache->lock);
4721         spin_unlock(&cache->space_info->lock);
4722
4723         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4724                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4725         return 0;
4726 }
4727
4728 /*
4729  * this function must be called within transaction
4730  */
4731 int btrfs_pin_extent(struct btrfs_root *root,
4732                      u64 bytenr, u64 num_bytes, int reserved)
4733 {
4734         struct btrfs_block_group_cache *cache;
4735
4736         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4737         BUG_ON(!cache); /* Logic error */
4738
4739         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4740
4741         btrfs_put_block_group(cache);
4742         return 0;
4743 }
4744
4745 /*
4746  * this function must be called within transaction
4747  */
4748 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4749                                     struct btrfs_root *root,
4750                                     u64 bytenr, u64 num_bytes)
4751 {
4752         struct btrfs_block_group_cache *cache;
4753
4754         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4755         BUG_ON(!cache); /* Logic error */
4756
4757         /*
4758          * pull in the free space cache (if any) so that our pin
4759          * removes the free space from the cache.  We have load_only set
4760          * to one because the slow code to read in the free extents does check
4761          * the pinned extents.
4762          */
4763         cache_block_group(cache, trans, root, 1);
4764
4765         pin_down_extent(root, cache, bytenr, num_bytes, 0);
4766
4767         /* remove us from the free space cache (if we're there at all) */
4768         btrfs_remove_free_space(cache, bytenr, num_bytes);
4769         btrfs_put_block_group(cache);
4770         return 0;
4771 }
4772
4773 /**
4774  * btrfs_update_reserved_bytes - update the block_group and space info counters
4775  * @cache:      The cache we are manipulating
4776  * @num_bytes:  The number of bytes in question
4777  * @reserve:    One of the reservation enums
4778  *
4779  * This is called by the allocator when it reserves space, or by somebody who is
4780  * freeing space that was never actually used on disk.  For example if you
4781  * reserve some space for a new leaf in transaction A and before transaction A
4782  * commits you free that leaf, you call this with reserve set to 0 in order to
4783  * clear the reservation.
4784  *
4785  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4786  * ENOSPC accounting.  For data we handle the reservation through clearing the
4787  * delalloc bits in the io_tree.  We have to do this since we could end up
4788  * allocating less disk space for the amount of data we have reserved in the
4789  * case of compression.
4790  *
4791  * If this is a reservation and the block group has become read only we cannot
4792  * make the reservation and return -EAGAIN, otherwise this function always
4793  * succeeds.
4794  */
4795 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4796                                        u64 num_bytes, int reserve)
4797 {
4798         struct btrfs_space_info *space_info = cache->space_info;
4799         int ret = 0;
4800
4801         spin_lock(&space_info->lock);
4802         spin_lock(&cache->lock);
4803         if (reserve != RESERVE_FREE) {
4804                 if (cache->ro) {
4805                         ret = -EAGAIN;
4806                 } else {
4807                         cache->reserved += num_bytes;
4808                         space_info->bytes_reserved += num_bytes;
4809                         if (reserve == RESERVE_ALLOC) {
4810                                 trace_btrfs_space_reservation(cache->fs_info,
4811                                                 "space_info", space_info->flags,
4812                                                 num_bytes, 0);
4813                                 space_info->bytes_may_use -= num_bytes;
4814                         }
4815                 }
4816         } else {
4817                 if (cache->ro)
4818                         space_info->bytes_readonly += num_bytes;
4819                 cache->reserved -= num_bytes;
4820                 space_info->bytes_reserved -= num_bytes;
4821                 space_info->reservation_progress++;
4822         }
4823         spin_unlock(&cache->lock);
4824         spin_unlock(&space_info->lock);
4825         return ret;
4826 }
4827
4828 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4829                                 struct btrfs_root *root)
4830 {
4831         struct btrfs_fs_info *fs_info = root->fs_info;
4832         struct btrfs_caching_control *next;
4833         struct btrfs_caching_control *caching_ctl;
4834         struct btrfs_block_group_cache *cache;
4835
4836         down_write(&fs_info->extent_commit_sem);
4837
4838         list_for_each_entry_safe(caching_ctl, next,
4839                                  &fs_info->caching_block_groups, list) {
4840                 cache = caching_ctl->block_group;
4841                 if (block_group_cache_done(cache)) {
4842                         cache->last_byte_to_unpin = (u64)-1;
4843                         list_del_init(&caching_ctl->list);
4844                         put_caching_control(caching_ctl);
4845                 } else {
4846                         cache->last_byte_to_unpin = caching_ctl->progress;
4847                 }
4848         }
4849
4850         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4851                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4852         else
4853                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4854
4855         up_write(&fs_info->extent_commit_sem);
4856
4857         update_global_block_rsv(fs_info);
4858 }
4859
4860 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4861 {
4862         struct btrfs_fs_info *fs_info = root->fs_info;
4863         struct btrfs_block_group_cache *cache = NULL;
4864         u64 len;
4865
4866         while (start <= end) {
4867                 if (!cache ||
4868                     start >= cache->key.objectid + cache->key.offset) {
4869                         if (cache)
4870                                 btrfs_put_block_group(cache);
4871                         cache = btrfs_lookup_block_group(fs_info, start);
4872                         BUG_ON(!cache); /* Logic error */
4873                 }
4874
4875                 len = cache->key.objectid + cache->key.offset - start;
4876                 len = min(len, end + 1 - start);
4877
4878                 if (start < cache->last_byte_to_unpin) {
4879                         len = min(len, cache->last_byte_to_unpin - start);
4880                         btrfs_add_free_space(cache, start, len);
4881                 }
4882
4883                 start += len;
4884
4885                 spin_lock(&cache->space_info->lock);
4886                 spin_lock(&cache->lock);
4887                 cache->pinned -= len;
4888                 cache->space_info->bytes_pinned -= len;
4889                 if (cache->ro)
4890                         cache->space_info->bytes_readonly += len;
4891                 spin_unlock(&cache->lock);
4892                 spin_unlock(&cache->space_info->lock);
4893         }
4894
4895         if (cache)
4896                 btrfs_put_block_group(cache);
4897         return 0;
4898 }
4899
4900 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4901                                struct btrfs_root *root)
4902 {
4903         struct btrfs_fs_info *fs_info = root->fs_info;
4904         struct extent_io_tree *unpin;
4905         u64 start;
4906         u64 end;
4907         int ret;
4908
4909         if (trans->aborted)
4910                 return 0;
4911
4912         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4913                 unpin = &fs_info->freed_extents[1];
4914         else
4915                 unpin = &fs_info->freed_extents[0];
4916
4917         while (1) {
4918                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4919                                             EXTENT_DIRTY);
4920                 if (ret)
4921                         break;
4922
4923                 if (btrfs_test_opt(root, DISCARD))
4924                         ret = btrfs_discard_extent(root, start,
4925                                                    end + 1 - start, NULL);
4926
4927                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4928                 unpin_extent_range(root, start, end);
4929                 cond_resched();
4930         }
4931
4932         return 0;
4933 }
4934
4935 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4936                                 struct btrfs_root *root,
4937                                 u64 bytenr, u64 num_bytes, u64 parent,
4938                                 u64 root_objectid, u64 owner_objectid,
4939                                 u64 owner_offset, int refs_to_drop,
4940                                 struct btrfs_delayed_extent_op *extent_op)
4941 {
4942         struct btrfs_key key;
4943         struct btrfs_path *path;
4944         struct btrfs_fs_info *info = root->fs_info;
4945         struct btrfs_root *extent_root = info->extent_root;
4946         struct extent_buffer *leaf;
4947         struct btrfs_extent_item *ei;
4948         struct btrfs_extent_inline_ref *iref;
4949         int ret;
4950         int is_data;
4951         int extent_slot = 0;
4952         int found_extent = 0;
4953         int num_to_del = 1;
4954         u32 item_size;
4955         u64 refs;
4956
4957         path = btrfs_alloc_path();
4958         if (!path)
4959                 return -ENOMEM;
4960
4961         path->reada = 1;
4962         path->leave_spinning = 1;
4963
4964         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4965         BUG_ON(!is_data && refs_to_drop != 1);
4966
4967         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4968                                     bytenr, num_bytes, parent,
4969                                     root_objectid, owner_objectid,
4970                                     owner_offset);
4971         if (ret == 0) {
4972                 extent_slot = path->slots[0];
4973                 while (extent_slot >= 0) {
4974                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4975                                               extent_slot);
4976                         if (key.objectid != bytenr)
4977                                 break;
4978                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4979                             key.offset == num_bytes) {
4980                                 found_extent = 1;
4981                                 break;
4982                         }
4983                         if (path->slots[0] - extent_slot > 5)
4984                                 break;
4985                         extent_slot--;
4986                 }
4987 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4988                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4989                 if (found_extent && item_size < sizeof(*ei))
4990                         found_extent = 0;
4991 #endif
4992                 if (!found_extent) {
4993                         BUG_ON(iref);
4994                         ret = remove_extent_backref(trans, extent_root, path,
4995                                                     NULL, refs_to_drop,
4996                                                     is_data);
4997                         if (ret)
4998                                 goto abort;
4999                         btrfs_release_path(path);
5000                         path->leave_spinning = 1;
5001
5002                         key.objectid = bytenr;
5003                         key.type = BTRFS_EXTENT_ITEM_KEY;
5004                         key.offset = num_bytes;
5005
5006                         ret = btrfs_search_slot(trans, extent_root,
5007                                                 &key, path, -1, 1);
5008                         if (ret) {
5009                                 printk(KERN_ERR "umm, got %d back from search"
5010                                        ", was looking for %llu\n", ret,
5011                                        (unsigned long long)bytenr);
5012                                 if (ret > 0)
5013                                         btrfs_print_leaf(extent_root,
5014                                                          path->nodes[0]);
5015                         }
5016                         if (ret < 0)
5017                                 goto abort;
5018                         extent_slot = path->slots[0];
5019                 }
5020         } else if (ret == -ENOENT) {
5021                 btrfs_print_leaf(extent_root, path->nodes[0]);
5022                 WARN_ON(1);
5023                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5024                        "parent %llu root %llu  owner %llu offset %llu\n",
5025                        (unsigned long long)bytenr,
5026                        (unsigned long long)parent,
5027                        (unsigned long long)root_objectid,
5028                        (unsigned long long)owner_objectid,
5029                        (unsigned long long)owner_offset);
5030         } else {
5031                 goto abort;
5032         }
5033
5034         leaf = path->nodes[0];
5035         item_size = btrfs_item_size_nr(leaf, extent_slot);
5036 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5037         if (item_size < sizeof(*ei)) {
5038                 BUG_ON(found_extent || extent_slot != path->slots[0]);
5039                 ret = convert_extent_item_v0(trans, extent_root, path,
5040                                              owner_objectid, 0);
5041                 if (ret < 0)
5042                         goto abort;
5043
5044                 btrfs_release_path(path);
5045                 path->leave_spinning = 1;
5046
5047                 key.objectid = bytenr;
5048                 key.type = BTRFS_EXTENT_ITEM_KEY;
5049                 key.offset = num_bytes;
5050
5051                 ret = btrfs_search_slot(trans, extent_root, &key, path,
5052                                         -1, 1);
5053                 if (ret) {
5054                         printk(KERN_ERR "umm, got %d back from search"
5055                                ", was looking for %llu\n", ret,
5056                                (unsigned long long)bytenr);
5057                         btrfs_print_leaf(extent_root, path->nodes[0]);
5058                 }
5059                 if (ret < 0)
5060                         goto abort;
5061                 extent_slot = path->slots[0];
5062                 leaf = path->nodes[0];
5063                 item_size = btrfs_item_size_nr(leaf, extent_slot);
5064         }
5065 #endif
5066         BUG_ON(item_size < sizeof(*ei));
5067         ei = btrfs_item_ptr(leaf, extent_slot,
5068                             struct btrfs_extent_item);
5069         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5070                 struct btrfs_tree_block_info *bi;
5071                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5072                 bi = (struct btrfs_tree_block_info *)(ei + 1);
5073                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5074         }
5075
5076         refs = btrfs_extent_refs(leaf, ei);
5077         BUG_ON(refs < refs_to_drop);
5078         refs -= refs_to_drop;
5079
5080         if (refs > 0) {
5081                 if (extent_op)
5082                         __run_delayed_extent_op(extent_op, leaf, ei);
5083                 /*
5084                  * In the case of inline back ref, reference count will
5085                  * be updated by remove_extent_backref
5086                  */
5087                 if (iref) {
5088                         BUG_ON(!found_extent);
5089                 } else {
5090                         btrfs_set_extent_refs(leaf, ei, refs);
5091                         btrfs_mark_buffer_dirty(leaf);
5092                 }
5093                 if (found_extent) {
5094                         ret = remove_extent_backref(trans, extent_root, path,
5095                                                     iref, refs_to_drop,
5096                                                     is_data);
5097                         if (ret)
5098                                 goto abort;
5099                 }
5100         } else {
5101                 if (found_extent) {
5102                         BUG_ON(is_data && refs_to_drop !=
5103                                extent_data_ref_count(root, path, iref));
5104                         if (iref) {
5105                                 BUG_ON(path->slots[0] != extent_slot);
5106                         } else {
5107                                 BUG_ON(path->slots[0] != extent_slot + 1);
5108                                 path->slots[0] = extent_slot;
5109                                 num_to_del = 2;
5110                         }
5111                 }
5112
5113                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5114                                       num_to_del);
5115                 if (ret)
5116                         goto abort;
5117                 btrfs_release_path(path);
5118
5119                 if (is_data) {
5120                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5121                         if (ret)
5122                                 goto abort;
5123                 }
5124
5125                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5126                 if (ret)
5127                         goto abort;
5128         }
5129 out:
5130         btrfs_free_path(path);
5131         return ret;
5132
5133 abort:
5134         btrfs_abort_transaction(trans, extent_root, ret);
5135         goto out;
5136 }
5137
5138 /*
5139  * when we free an block, it is possible (and likely) that we free the last
5140  * delayed ref for that extent as well.  This searches the delayed ref tree for
5141  * a given extent, and if there are no other delayed refs to be processed, it
5142  * removes it from the tree.
5143  */
5144 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5145                                       struct btrfs_root *root, u64 bytenr)
5146 {
5147         struct btrfs_delayed_ref_head *head;
5148         struct btrfs_delayed_ref_root *delayed_refs;
5149         struct btrfs_delayed_ref_node *ref;
5150         struct rb_node *node;
5151         int ret = 0;
5152
5153         delayed_refs = &trans->transaction->delayed_refs;
5154         spin_lock(&delayed_refs->lock);
5155         head = btrfs_find_delayed_ref_head(trans, bytenr);
5156         if (!head)
5157                 goto out;
5158
5159         node = rb_prev(&head->node.rb_node);
5160         if (!node)
5161                 goto out;
5162
5163         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5164
5165         /* there are still entries for this ref, we can't drop it */
5166         if (ref->bytenr == bytenr)
5167                 goto out;
5168
5169         if (head->extent_op) {
5170                 if (!head->must_insert_reserved)
5171                         goto out;
5172                 kfree(head->extent_op);
5173                 head->extent_op = NULL;
5174         }
5175
5176         /*
5177          * waiting for the lock here would deadlock.  If someone else has it
5178          * locked they are already in the process of dropping it anyway
5179          */
5180         if (!mutex_trylock(&head->mutex))
5181                 goto out;
5182
5183         /*
5184          * at this point we have a head with no other entries.  Go
5185          * ahead and process it.
5186          */
5187         head->node.in_tree = 0;
5188         rb_erase(&head->node.rb_node, &delayed_refs->root);
5189
5190         delayed_refs->num_entries--;
5191         if (waitqueue_active(&delayed_refs->seq_wait))
5192                 wake_up(&delayed_refs->seq_wait);
5193
5194         /*
5195          * we don't take a ref on the node because we're removing it from the
5196          * tree, so we just steal the ref the tree was holding.
5197          */
5198         delayed_refs->num_heads--;
5199         if (list_empty(&head->cluster))
5200                 delayed_refs->num_heads_ready--;
5201
5202         list_del_init(&head->cluster);
5203         spin_unlock(&delayed_refs->lock);
5204
5205         BUG_ON(head->extent_op);
5206         if (head->must_insert_reserved)
5207                 ret = 1;
5208
5209         mutex_unlock(&head->mutex);
5210         btrfs_put_delayed_ref(&head->node);
5211         return ret;
5212 out:
5213         spin_unlock(&delayed_refs->lock);
5214         return 0;
5215 }
5216
5217 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5218                            struct btrfs_root *root,
5219                            struct extent_buffer *buf,
5220                            u64 parent, int last_ref, int for_cow)
5221 {
5222         struct btrfs_block_group_cache *cache = NULL;
5223         int ret;
5224
5225         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5226                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5227                                         buf->start, buf->len,
5228                                         parent, root->root_key.objectid,
5229                                         btrfs_header_level(buf),
5230                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5231                 BUG_ON(ret); /* -ENOMEM */
5232         }
5233
5234         if (!last_ref)
5235                 return;
5236
5237         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5238
5239         if (btrfs_header_generation(buf) == trans->transid) {
5240                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5241                         ret = check_ref_cleanup(trans, root, buf->start);
5242                         if (!ret)
5243                                 goto out;
5244                 }
5245
5246                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5247                         pin_down_extent(root, cache, buf->start, buf->len, 1);
5248                         goto out;
5249                 }
5250
5251                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5252
5253                 btrfs_add_free_space(cache, buf->start, buf->len);
5254                 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5255         }
5256 out:
5257         /*
5258          * Deleting the buffer, clear the corrupt flag since it doesn't matter
5259          * anymore.
5260          */
5261         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5262         btrfs_put_block_group(cache);
5263 }
5264
5265 /* Can return -ENOMEM */
5266 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5267                       u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5268                       u64 owner, u64 offset, int for_cow)
5269 {
5270         int ret;
5271         struct btrfs_fs_info *fs_info = root->fs_info;
5272
5273         /*
5274          * tree log blocks never actually go into the extent allocation
5275          * tree, just update pinning info and exit early.
5276          */
5277         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5278                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5279                 /* unlocks the pinned mutex */
5280                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5281                 ret = 0;
5282         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5283                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5284                                         num_bytes,
5285                                         parent, root_objectid, (int)owner,
5286                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5287         } else {
5288                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5289                                                 num_bytes,
5290                                                 parent, root_objectid, owner,
5291                                                 offset, BTRFS_DROP_DELAYED_REF,
5292                                                 NULL, for_cow);
5293         }
5294         return ret;
5295 }
5296
5297 static u64 stripe_align(struct btrfs_root *root, u64 val)
5298 {
5299         u64 mask = ((u64)root->stripesize - 1);
5300         u64 ret = (val + mask) & ~mask;
5301         return ret;
5302 }
5303
5304 /*
5305  * when we wait for progress in the block group caching, its because
5306  * our allocation attempt failed at least once.  So, we must sleep
5307  * and let some progress happen before we try again.
5308  *
5309  * This function will sleep at least once waiting for new free space to
5310  * show up, and then it will check the block group free space numbers
5311  * for our min num_bytes.  Another option is to have it go ahead
5312  * and look in the rbtree for a free extent of a given size, but this
5313  * is a good start.
5314  */
5315 static noinline int
5316 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5317                                 u64 num_bytes)
5318 {
5319         struct btrfs_caching_control *caching_ctl;
5320         DEFINE_WAIT(wait);
5321
5322         caching_ctl = get_caching_control(cache);
5323         if (!caching_ctl)
5324                 return 0;
5325
5326         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5327                    (cache->free_space_ctl->free_space >= num_bytes));
5328
5329         put_caching_control(caching_ctl);
5330         return 0;
5331 }
5332
5333 static noinline int
5334 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5335 {
5336         struct btrfs_caching_control *caching_ctl;
5337         DEFINE_WAIT(wait);
5338
5339         caching_ctl = get_caching_control(cache);
5340         if (!caching_ctl)
5341                 return 0;
5342
5343         wait_event(caching_ctl->wait, block_group_cache_done(cache));
5344
5345         put_caching_control(caching_ctl);
5346         return 0;
5347 }
5348
5349 static int __get_block_group_index(u64 flags)
5350 {
5351         int index;
5352
5353         if (flags & BTRFS_BLOCK_GROUP_RAID10)
5354                 index = 0;
5355         else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5356                 index = 1;
5357         else if (flags & BTRFS_BLOCK_GROUP_DUP)
5358                 index = 2;
5359         else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5360                 index = 3;
5361         else
5362                 index = 4;
5363
5364         return index;
5365 }
5366
5367 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5368 {
5369         return __get_block_group_index(cache->flags);
5370 }
5371
5372 enum btrfs_loop_type {
5373         LOOP_CACHING_NOWAIT = 0,
5374         LOOP_CACHING_WAIT = 1,
5375         LOOP_ALLOC_CHUNK = 2,
5376         LOOP_NO_EMPTY_SIZE = 3,
5377 };
5378
5379 /*
5380  * walks the btree of allocated extents and find a hole of a given size.
5381  * The key ins is changed to record the hole:
5382  * ins->objectid == block start
5383  * ins->flags = BTRFS_EXTENT_ITEM_KEY
5384  * ins->offset == number of blocks
5385  * Any available blocks before search_start are skipped.
5386  */
5387 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5388                                      struct btrfs_root *orig_root,
5389                                      u64 num_bytes, u64 empty_size,
5390                                      u64 hint_byte, struct btrfs_key *ins,
5391                                      u64 data)
5392 {
5393         int ret = 0;
5394         struct btrfs_root *root = orig_root->fs_info->extent_root;
5395         struct btrfs_free_cluster *last_ptr = NULL;
5396         struct btrfs_block_group_cache *block_group = NULL;
5397         struct btrfs_block_group_cache *used_block_group;
5398         u64 search_start = 0;
5399         int empty_cluster = 2 * 1024 * 1024;
5400         int allowed_chunk_alloc = 0;
5401         int done_chunk_alloc = 0;
5402         struct btrfs_space_info *space_info;
5403         int loop = 0;
5404         int index = 0;
5405         int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5406                 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5407         bool found_uncached_bg = false;
5408         bool failed_cluster_refill = false;
5409         bool failed_alloc = false;
5410         bool use_cluster = true;
5411         bool have_caching_bg = false;
5412
5413         WARN_ON(num_bytes < root->sectorsize);
5414         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5415         ins->objectid = 0;
5416         ins->offset = 0;
5417
5418         trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5419
5420         space_info = __find_space_info(root->fs_info, data);
5421         if (!space_info) {
5422                 printk(KERN_ERR "No space info for %llu\n", data);
5423                 return -ENOSPC;
5424         }
5425
5426         /*
5427          * If the space info is for both data and metadata it means we have a
5428          * small filesystem and we can't use the clustering stuff.
5429          */
5430         if (btrfs_mixed_space_info(space_info))
5431                 use_cluster = false;
5432
5433         if (orig_root->ref_cows || empty_size)
5434                 allowed_chunk_alloc = 1;
5435
5436         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5437                 last_ptr = &root->fs_info->meta_alloc_cluster;
5438                 if (!btrfs_test_opt(root, SSD))
5439                         empty_cluster = 64 * 1024;
5440         }
5441
5442         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5443             btrfs_test_opt(root, SSD)) {
5444                 last_ptr = &root->fs_info->data_alloc_cluster;
5445         }
5446
5447         if (last_ptr) {
5448                 spin_lock(&last_ptr->lock);
5449                 if (last_ptr->block_group)
5450                         hint_byte = last_ptr->window_start;
5451                 spin_unlock(&last_ptr->lock);
5452         }
5453
5454         search_start = max(search_start, first_logical_byte(root, 0));
5455         search_start = max(search_start, hint_byte);
5456
5457         if (!last_ptr)
5458                 empty_cluster = 0;
5459
5460         if (search_start == hint_byte) {
5461                 block_group = btrfs_lookup_block_group(root->fs_info,
5462                                                        search_start);
5463                 used_block_group = block_group;
5464                 /*
5465                  * we don't want to use the block group if it doesn't match our
5466                  * allocation bits, or if its not cached.
5467                  *
5468                  * However if we are re-searching with an ideal block group
5469                  * picked out then we don't care that the block group is cached.
5470                  */
5471                 if (block_group && block_group_bits(block_group, data) &&
5472                     block_group->cached != BTRFS_CACHE_NO) {
5473                         down_read(&space_info->groups_sem);
5474                         if (list_empty(&block_group->list) ||
5475                             block_group->ro) {
5476                                 /*
5477                                  * someone is removing this block group,
5478                                  * we can't jump into the have_block_group
5479                                  * target because our list pointers are not
5480                                  * valid
5481                                  */
5482                                 btrfs_put_block_group(block_group);
5483                                 up_read(&space_info->groups_sem);
5484                         } else {
5485                                 index = get_block_group_index(block_group);
5486                                 goto have_block_group;
5487                         }
5488                 } else if (block_group) {
5489                         btrfs_put_block_group(block_group);
5490                 }
5491         }
5492 search:
5493         have_caching_bg = false;
5494         down_read(&space_info->groups_sem);
5495         list_for_each_entry(block_group, &space_info->block_groups[index],
5496                             list) {
5497                 u64 offset;
5498                 int cached;
5499
5500                 used_block_group = block_group;
5501                 btrfs_get_block_group(block_group);
5502                 search_start = block_group->key.objectid;
5503
5504                 /*
5505                  * this can happen if we end up cycling through all the
5506                  * raid types, but we want to make sure we only allocate
5507                  * for the proper type.
5508                  */
5509                 if (!block_group_bits(block_group, data)) {
5510                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
5511                                 BTRFS_BLOCK_GROUP_RAID1 |
5512                                 BTRFS_BLOCK_GROUP_RAID10;
5513
5514                         /*
5515                          * if they asked for extra copies and this block group
5516                          * doesn't provide them, bail.  This does allow us to
5517                          * fill raid0 from raid1.
5518                          */
5519                         if ((data & extra) && !(block_group->flags & extra))
5520                                 goto loop;
5521                 }
5522
5523 have_block_group:
5524                 cached = block_group_cache_done(block_group);
5525                 if (unlikely(!cached)) {
5526                         found_uncached_bg = true;
5527                         ret = cache_block_group(block_group, trans,
5528                                                 orig_root, 0);
5529                         BUG_ON(ret < 0);
5530                         ret = 0;
5531                 }
5532
5533                 if (unlikely(block_group->ro))
5534                         goto loop;
5535
5536                 /*
5537                  * Ok we want to try and use the cluster allocator, so
5538                  * lets look there
5539                  */
5540                 if (last_ptr) {
5541                         /*
5542                          * the refill lock keeps out other
5543                          * people trying to start a new cluster
5544                          */
5545                         spin_lock(&last_ptr->refill_lock);
5546                         used_block_group = last_ptr->block_group;
5547                         if (used_block_group != block_group &&
5548                             (!used_block_group ||
5549                              used_block_group->ro ||
5550                              !block_group_bits(used_block_group, data))) {
5551                                 used_block_group = block_group;
5552                                 goto refill_cluster;
5553                         }
5554
5555                         if (used_block_group != block_group)
5556                                 btrfs_get_block_group(used_block_group);
5557
5558                         offset = btrfs_alloc_from_cluster(used_block_group,
5559                           last_ptr, num_bytes, used_block_group->key.objectid);
5560                         if (offset) {
5561                                 /* we have a block, we're done */
5562                                 spin_unlock(&last_ptr->refill_lock);
5563                                 trace_btrfs_reserve_extent_cluster(root,
5564                                         block_group, search_start, num_bytes);
5565                                 goto checks;
5566                         }
5567
5568                         WARN_ON(last_ptr->block_group != used_block_group);
5569                         if (used_block_group != block_group) {
5570                                 btrfs_put_block_group(used_block_group);
5571                                 used_block_group = block_group;
5572                         }
5573 refill_cluster:
5574                         BUG_ON(used_block_group != block_group);
5575                         /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5576                          * set up a new clusters, so lets just skip it
5577                          * and let the allocator find whatever block
5578                          * it can find.  If we reach this point, we
5579                          * will have tried the cluster allocator
5580                          * plenty of times and not have found
5581                          * anything, so we are likely way too
5582                          * fragmented for the clustering stuff to find
5583                          * anything.
5584                          *
5585                          * However, if the cluster is taken from the
5586                          * current block group, release the cluster
5587                          * first, so that we stand a better chance of
5588                          * succeeding in the unclustered
5589                          * allocation.  */
5590                         if (loop >= LOOP_NO_EMPTY_SIZE &&
5591                             last_ptr->block_group != block_group) {
5592                                 spin_unlock(&last_ptr->refill_lock);
5593                                 goto unclustered_alloc;
5594                         }
5595
5596                         /*
5597                          * this cluster didn't work out, free it and
5598                          * start over
5599                          */
5600                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5601
5602                         if (loop >= LOOP_NO_EMPTY_SIZE) {
5603                                 spin_unlock(&last_ptr->refill_lock);
5604                                 goto unclustered_alloc;
5605                         }
5606
5607                         /* allocate a cluster in this block group */
5608                         ret = btrfs_find_space_cluster(trans, root,
5609                                                block_group, last_ptr,
5610                                                search_start, num_bytes,
5611                                                empty_cluster + empty_size);
5612                         if (ret == 0) {
5613                                 /*
5614                                  * now pull our allocation out of this
5615                                  * cluster
5616                                  */
5617                                 offset = btrfs_alloc_from_cluster(block_group,
5618                                                   last_ptr, num_bytes,
5619                                                   search_start);
5620                                 if (offset) {
5621                                         /* we found one, proceed */
5622                                         spin_unlock(&last_ptr->refill_lock);
5623                                         trace_btrfs_reserve_extent_cluster(root,
5624                                                 block_group, search_start,
5625                                                 num_bytes);
5626                                         goto checks;
5627                                 }
5628                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5629                                    && !failed_cluster_refill) {
5630                                 spin_unlock(&last_ptr->refill_lock);
5631
5632                                 failed_cluster_refill = true;
5633                                 wait_block_group_cache_progress(block_group,
5634                                        num_bytes + empty_cluster + empty_size);
5635                                 goto have_block_group;
5636                         }
5637
5638                         /*
5639                          * at this point we either didn't find a cluster
5640                          * or we weren't able to allocate a block from our
5641                          * cluster.  Free the cluster we've been trying
5642                          * to use, and go to the next block group
5643                          */
5644                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5645                         spin_unlock(&last_ptr->refill_lock);
5646                         goto loop;
5647                 }
5648
5649 unclustered_alloc:
5650                 spin_lock(&block_group->free_space_ctl->tree_lock);
5651                 if (cached &&
5652                     block_group->free_space_ctl->free_space <
5653                     num_bytes + empty_cluster + empty_size) {
5654                         spin_unlock(&block_group->free_space_ctl->tree_lock);
5655                         goto loop;
5656                 }
5657                 spin_unlock(&block_group->free_space_ctl->tree_lock);
5658
5659                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5660                                                     num_bytes, empty_size);
5661                 /*
5662                  * If we didn't find a chunk, and we haven't failed on this
5663                  * block group before, and this block group is in the middle of
5664                  * caching and we are ok with waiting, then go ahead and wait
5665                  * for progress to be made, and set failed_alloc to true.
5666                  *
5667                  * If failed_alloc is true then we've already waited on this
5668                  * block group once and should move on to the next block group.
5669                  */
5670                 if (!offset && !failed_alloc && !cached &&
5671                     loop > LOOP_CACHING_NOWAIT) {
5672                         wait_block_group_cache_progress(block_group,
5673                                                 num_bytes + empty_size);
5674                         failed_alloc = true;
5675                         goto have_block_group;
5676                 } else if (!offset) {
5677                         if (!cached)
5678                                 have_caching_bg = true;
5679                         goto loop;
5680                 }
5681 checks:
5682                 search_start = stripe_align(root, offset);
5683
5684                 /* move on to the next group */
5685                 if (search_start + num_bytes >
5686                     used_block_group->key.objectid + used_block_group->key.offset) {
5687                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5688                         goto loop;
5689                 }
5690
5691                 if (offset < search_start)
5692                         btrfs_add_free_space(used_block_group, offset,
5693                                              search_start - offset);
5694                 BUG_ON(offset > search_start);
5695
5696                 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5697                                                   alloc_type);
5698                 if (ret == -EAGAIN) {
5699                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5700                         goto loop;
5701                 }
5702
5703                 /* we are all good, lets return */
5704                 ins->objectid = search_start;
5705                 ins->offset = num_bytes;
5706
5707                 trace_btrfs_reserve_extent(orig_root, block_group,
5708                                            search_start, num_bytes);
5709                 if (offset < search_start)
5710                         btrfs_add_free_space(used_block_group, offset,
5711                                              search_start - offset);
5712                 BUG_ON(offset > search_start);
5713                 if (used_block_group != block_group)
5714                         btrfs_put_block_group(used_block_group);
5715                 btrfs_put_block_group(block_group);
5716                 break;
5717 loop:
5718                 failed_cluster_refill = false;
5719                 failed_alloc = false;
5720                 BUG_ON(index != get_block_group_index(block_group));
5721                 if (used_block_group != block_group)
5722                         btrfs_put_block_group(used_block_group);
5723                 btrfs_put_block_group(block_group);
5724         }
5725         up_read(&space_info->groups_sem);
5726
5727         if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5728                 goto search;
5729
5730         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5731                 goto search;
5732
5733         /*
5734          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5735          *                      caching kthreads as we move along
5736          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5737          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5738          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5739          *                      again
5740          */
5741         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5742                 index = 0;
5743                 loop++;
5744                 if (loop == LOOP_ALLOC_CHUNK) {
5745                        if (allowed_chunk_alloc) {
5746                                 ret = do_chunk_alloc(trans, root, num_bytes +
5747                                                      2 * 1024 * 1024, data,
5748                                                      CHUNK_ALLOC_LIMITED);
5749                                 if (ret < 0) {
5750                                         btrfs_abort_transaction(trans,
5751                                                                 root, ret);
5752                                         goto out;
5753                                 }
5754                                 allowed_chunk_alloc = 0;
5755                                 if (ret == 1)
5756                                         done_chunk_alloc = 1;
5757                         } else if (!done_chunk_alloc &&
5758                                    space_info->force_alloc ==
5759                                    CHUNK_ALLOC_NO_FORCE) {
5760                                 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5761                         }
5762
5763                        /*
5764                         * We didn't allocate a chunk, go ahead and drop the
5765                         * empty size and loop again.
5766                         */
5767                        if (!done_chunk_alloc)
5768                                loop = LOOP_NO_EMPTY_SIZE;
5769                 }
5770
5771                 if (loop == LOOP_NO_EMPTY_SIZE) {
5772                         empty_size = 0;
5773                         empty_cluster = 0;
5774                 }
5775
5776                 goto search;
5777         } else if (!ins->objectid) {
5778                 ret = -ENOSPC;
5779         } else if (ins->objectid) {
5780                 ret = 0;
5781         }
5782 out:
5783
5784         return ret;
5785 }
5786
5787 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5788                             int dump_block_groups)
5789 {
5790         struct btrfs_block_group_cache *cache;
5791         int index = 0;
5792
5793         spin_lock(&info->lock);
5794         printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5795                (unsigned long long)info->flags,
5796                (unsigned long long)(info->total_bytes - info->bytes_used -
5797                                     info->bytes_pinned - info->bytes_reserved -
5798                                     info->bytes_readonly),
5799                (info->full) ? "" : "not ");
5800         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5801                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5802                (unsigned long long)info->total_bytes,
5803                (unsigned long long)info->bytes_used,
5804                (unsigned long long)info->bytes_pinned,
5805                (unsigned long long)info->bytes_reserved,
5806                (unsigned long long)info->bytes_may_use,
5807                (unsigned long long)info->bytes_readonly);
5808         spin_unlock(&info->lock);
5809
5810         if (!dump_block_groups)
5811                 return;
5812
5813         down_read(&info->groups_sem);
5814 again:
5815         list_for_each_entry(cache, &info->block_groups[index], list) {
5816                 spin_lock(&cache->lock);
5817                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5818                        "%llu pinned %llu reserved\n",
5819                        (unsigned long long)cache->key.objectid,
5820                        (unsigned long long)cache->key.offset,
5821                        (unsigned long long)btrfs_block_group_used(&cache->item),
5822                        (unsigned long long)cache->pinned,
5823                        (unsigned long long)cache->reserved);
5824                 btrfs_dump_free_space(cache, bytes);
5825                 spin_unlock(&cache->lock);
5826         }
5827         if (++index < BTRFS_NR_RAID_TYPES)
5828                 goto again;
5829         up_read(&info->groups_sem);
5830 }
5831
5832 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5833                          struct btrfs_root *root,
5834                          u64 num_bytes, u64 min_alloc_size,
5835                          u64 empty_size, u64 hint_byte,
5836                          struct btrfs_key *ins, u64 data)
5837 {
5838         bool final_tried = false;
5839         int ret;
5840
5841         data = btrfs_get_alloc_profile(root, data);
5842 again:
5843         /*
5844          * the only place that sets empty_size is btrfs_realloc_node, which
5845          * is not called recursively on allocations
5846          */
5847         if (empty_size || root->ref_cows) {
5848                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5849                                      num_bytes + 2 * 1024 * 1024, data,
5850                                      CHUNK_ALLOC_NO_FORCE);
5851                 if (ret < 0 && ret != -ENOSPC) {
5852                         btrfs_abort_transaction(trans, root, ret);
5853                         return ret;
5854                 }
5855         }
5856
5857         WARN_ON(num_bytes < root->sectorsize);
5858         ret = find_free_extent(trans, root, num_bytes, empty_size,
5859                                hint_byte, ins, data);
5860
5861         if (ret == -ENOSPC) {
5862                 if (!final_tried) {
5863                         num_bytes = num_bytes >> 1;
5864                         num_bytes = num_bytes & ~(root->sectorsize - 1);
5865                         num_bytes = max(num_bytes, min_alloc_size);
5866                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5867                                        num_bytes, data, CHUNK_ALLOC_FORCE);
5868                         if (ret < 0 && ret != -ENOSPC) {
5869                                 btrfs_abort_transaction(trans, root, ret);
5870                                 return ret;
5871                         }
5872                         if (num_bytes == min_alloc_size)
5873                                 final_tried = true;
5874                         goto again;
5875                 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5876                         struct btrfs_space_info *sinfo;
5877
5878                         sinfo = __find_space_info(root->fs_info, data);
5879                         printk(KERN_ERR "btrfs allocation failed flags %llu, "
5880                                "wanted %llu\n", (unsigned long long)data,
5881                                (unsigned long long)num_bytes);
5882                         if (sinfo)
5883                                 dump_space_info(sinfo, num_bytes, 1);
5884                 }
5885         }
5886
5887         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5888
5889         return ret;
5890 }
5891
5892 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5893                                         u64 start, u64 len, int pin)
5894 {
5895         struct btrfs_block_group_cache *cache;
5896         int ret = 0;
5897
5898         cache = btrfs_lookup_block_group(root->fs_info, start);
5899         if (!cache) {
5900                 printk(KERN_ERR "Unable to find block group for %llu\n",
5901                        (unsigned long long)start);
5902                 return -ENOSPC;
5903         }
5904
5905         if (btrfs_test_opt(root, DISCARD))
5906                 ret = btrfs_discard_extent(root, start, len, NULL);
5907
5908         if (pin)
5909                 pin_down_extent(root, cache, start, len, 1);
5910         else {
5911                 btrfs_add_free_space(cache, start, len);
5912                 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5913         }
5914         btrfs_put_block_group(cache);
5915
5916         trace_btrfs_reserved_extent_free(root, start, len);
5917
5918         return ret;
5919 }
5920
5921 int btrfs_free_reserved_extent(struct btrfs_root *root,
5922                                         u64 start, u64 len)
5923 {
5924         return __btrfs_free_reserved_extent(root, start, len, 0);
5925 }
5926
5927 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5928                                        u64 start, u64 len)
5929 {
5930         return __btrfs_free_reserved_extent(root, start, len, 1);
5931 }
5932
5933 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5934                                       struct btrfs_root *root,
5935                                       u64 parent, u64 root_objectid,
5936                                       u64 flags, u64 owner, u64 offset,
5937                                       struct btrfs_key *ins, int ref_mod)
5938 {
5939         int ret;
5940         struct btrfs_fs_info *fs_info = root->fs_info;
5941         struct btrfs_extent_item *extent_item;
5942         struct btrfs_extent_inline_ref *iref;
5943         struct btrfs_path *path;
5944         struct extent_buffer *leaf;
5945         int type;
5946         u32 size;
5947
5948         if (parent > 0)
5949                 type = BTRFS_SHARED_DATA_REF_KEY;
5950         else
5951                 type = BTRFS_EXTENT_DATA_REF_KEY;
5952
5953         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5954
5955         path = btrfs_alloc_path();
5956         if (!path)
5957                 return -ENOMEM;
5958
5959         path->leave_spinning = 1;
5960         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5961                                       ins, size);
5962         if (ret) {
5963                 btrfs_free_path(path);
5964                 return ret;
5965         }
5966
5967         leaf = path->nodes[0];
5968         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5969                                      struct btrfs_extent_item);
5970         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5971         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5972         btrfs_set_extent_flags(leaf, extent_item,
5973                                flags | BTRFS_EXTENT_FLAG_DATA);
5974
5975         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5976         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5977         if (parent > 0) {
5978                 struct btrfs_shared_data_ref *ref;
5979                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5980                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5981                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5982         } else {
5983                 struct btrfs_extent_data_ref *ref;
5984                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5985                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5986                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5987                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5988                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5989         }
5990
5991         btrfs_mark_buffer_dirty(path->nodes[0]);
5992         btrfs_free_path(path);
5993
5994         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5995         if (ret) { /* -ENOENT, logic error */
5996                 printk(KERN_ERR "btrfs update block group failed for %llu "
5997                        "%llu\n", (unsigned long long)ins->objectid,
5998                        (unsigned long long)ins->offset);
5999                 BUG();
6000         }
6001         return ret;
6002 }
6003
6004 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6005                                      struct btrfs_root *root,
6006                                      u64 parent, u64 root_objectid,
6007                                      u64 flags, struct btrfs_disk_key *key,
6008                                      int level, struct btrfs_key *ins)
6009 {
6010         int ret;
6011         struct btrfs_fs_info *fs_info = root->fs_info;
6012         struct btrfs_extent_item *extent_item;
6013         struct btrfs_tree_block_info *block_info;
6014         struct btrfs_extent_inline_ref *iref;
6015         struct btrfs_path *path;
6016         struct extent_buffer *leaf;
6017         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6018
6019         path = btrfs_alloc_path();
6020         if (!path)
6021                 return -ENOMEM;
6022
6023         path->leave_spinning = 1;
6024         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6025                                       ins, size);
6026         if (ret) {
6027                 btrfs_free_path(path);
6028                 return ret;
6029         }
6030
6031         leaf = path->nodes[0];
6032         extent_item = btrfs_item_ptr(leaf, path->slots[0],
6033                                      struct btrfs_extent_item);
6034         btrfs_set_extent_refs(leaf, extent_item, 1);
6035         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6036         btrfs_set_extent_flags(leaf, extent_item,
6037                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6038         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6039
6040         btrfs_set_tree_block_key(leaf, block_info, key);
6041         btrfs_set_tree_block_level(leaf, block_info, level);
6042
6043         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6044         if (parent > 0) {
6045                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6046                 btrfs_set_extent_inline_ref_type(leaf, iref,
6047                                                  BTRFS_SHARED_BLOCK_REF_KEY);
6048                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6049         } else {
6050                 btrfs_set_extent_inline_ref_type(leaf, iref,
6051                                                  BTRFS_TREE_BLOCK_REF_KEY);
6052                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6053         }
6054
6055         btrfs_mark_buffer_dirty(leaf);
6056         btrfs_free_path(path);
6057
6058         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6059         if (ret) { /* -ENOENT, logic error */
6060                 printk(KERN_ERR "btrfs update block group failed for %llu "
6061                        "%llu\n", (unsigned long long)ins->objectid,
6062                        (unsigned long long)ins->offset);
6063                 BUG();
6064         }
6065         return ret;
6066 }
6067
6068 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6069                                      struct btrfs_root *root,
6070                                      u64 root_objectid, u64 owner,
6071                                      u64 offset, struct btrfs_key *ins)
6072 {
6073         int ret;
6074
6075         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6076
6077         ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6078                                          ins->offset, 0,
6079                                          root_objectid, owner, offset,
6080                                          BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6081         return ret;
6082 }
6083
6084 /*
6085  * this is used by the tree logging recovery code.  It records that
6086  * an extent has been allocated and makes sure to clear the free
6087  * space cache bits as well
6088  */
6089 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6090                                    struct btrfs_root *root,
6091                                    u64 root_objectid, u64 owner, u64 offset,
6092                                    struct btrfs_key *ins)
6093 {
6094         int ret;
6095         struct btrfs_block_group_cache *block_group;
6096         struct btrfs_caching_control *caching_ctl;
6097         u64 start = ins->objectid;
6098         u64 num_bytes = ins->offset;
6099
6100         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6101         cache_block_group(block_group, trans, NULL, 0);
6102         caching_ctl = get_caching_control(block_group);
6103
6104         if (!caching_ctl) {
6105                 BUG_ON(!block_group_cache_done(block_group));
6106                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6107                 BUG_ON(ret); /* -ENOMEM */
6108         } else {
6109                 mutex_lock(&caching_ctl->mutex);
6110
6111                 if (start >= caching_ctl->progress) {
6112                         ret = add_excluded_extent(root, start, num_bytes);
6113                         BUG_ON(ret); /* -ENOMEM */
6114                 } else if (start + num_bytes <= caching_ctl->progress) {
6115                         ret = btrfs_remove_free_space(block_group,
6116                                                       start, num_bytes);
6117                         BUG_ON(ret); /* -ENOMEM */
6118                 } else {
6119                         num_bytes = caching_ctl->progress - start;
6120                         ret = btrfs_remove_free_space(block_group,
6121                                                       start, num_bytes);
6122                         BUG_ON(ret); /* -ENOMEM */
6123
6124                         start = caching_ctl->progress;
6125                         num_bytes = ins->objectid + ins->offset -
6126                                     caching_ctl->progress;
6127                         ret = add_excluded_extent(root, start, num_bytes);
6128                         BUG_ON(ret); /* -ENOMEM */
6129                 }
6130
6131                 mutex_unlock(&caching_ctl->mutex);
6132                 put_caching_control(caching_ctl);
6133         }
6134
6135         ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6136                                           RESERVE_ALLOC_NO_ACCOUNT);
6137         BUG_ON(ret); /* logic error */
6138         btrfs_put_block_group(block_group);
6139         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6140                                          0, owner, offset, ins, 1);
6141         return ret;
6142 }
6143
6144 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6145                                             struct btrfs_root *root,
6146                                             u64 bytenr, u32 blocksize,
6147                                             int level)
6148 {
6149         struct extent_buffer *buf;
6150
6151         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6152         if (!buf)
6153                 return ERR_PTR(-ENOMEM);
6154         btrfs_set_header_generation(buf, trans->transid);
6155         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6156         btrfs_tree_lock(buf);
6157         clean_tree_block(trans, root, buf);
6158         clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6159
6160         btrfs_set_lock_blocking(buf);
6161         btrfs_set_buffer_uptodate(buf);
6162
6163         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6164                 /*
6165                  * we allow two log transactions at a time, use different
6166                  * EXENT bit to differentiate dirty pages.
6167                  */
6168                 if (root->log_transid % 2 == 0)
6169                         set_extent_dirty(&root->dirty_log_pages, buf->start,
6170                                         buf->start + buf->len - 1, GFP_NOFS);
6171                 else
6172                         set_extent_new(&root->dirty_log_pages, buf->start,
6173                                         buf->start + buf->len - 1, GFP_NOFS);
6174         } else {
6175                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6176                          buf->start + buf->len - 1, GFP_NOFS);
6177         }
6178         trans->blocks_used++;
6179         /* this returns a buffer locked for blocking */
6180         return buf;
6181 }
6182
6183 static struct btrfs_block_rsv *
6184 use_block_rsv(struct btrfs_trans_handle *trans,
6185               struct btrfs_root *root, u32 blocksize)
6186 {
6187         struct btrfs_block_rsv *block_rsv;
6188         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6189         int ret;
6190
6191         block_rsv = get_block_rsv(trans, root);
6192
6193         if (block_rsv->size == 0) {
6194                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6195                 /*
6196                  * If we couldn't reserve metadata bytes try and use some from
6197                  * the global reserve.
6198                  */
6199                 if (ret && block_rsv != global_rsv) {
6200                         ret = block_rsv_use_bytes(global_rsv, blocksize);
6201                         if (!ret)
6202                                 return global_rsv;
6203                         return ERR_PTR(ret);
6204                 } else if (ret) {
6205                         return ERR_PTR(ret);
6206                 }
6207                 return block_rsv;
6208         }
6209
6210         ret = block_rsv_use_bytes(block_rsv, blocksize);
6211         if (!ret)
6212                 return block_rsv;
6213         if (ret) {
6214                 static DEFINE_RATELIMIT_STATE(_rs,
6215                                 DEFAULT_RATELIMIT_INTERVAL,
6216                                 /*DEFAULT_RATELIMIT_BURST*/ 2);
6217                 if (__ratelimit(&_rs)) {
6218                         printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6219                         WARN_ON(1);
6220                 }
6221                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6222                 if (!ret) {
6223                         return block_rsv;
6224                 } else if (ret && block_rsv != global_rsv) {
6225                         ret = block_rsv_use_bytes(global_rsv, blocksize);
6226                         if (!ret)
6227                                 return global_rsv;
6228                 }
6229         }
6230
6231         return ERR_PTR(-ENOSPC);
6232 }
6233
6234 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6235                             struct btrfs_block_rsv *block_rsv, u32 blocksize)
6236 {
6237         block_rsv_add_bytes(block_rsv, blocksize, 0);
6238         block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6239 }
6240
6241 /*
6242  * finds a free extent and does all the dirty work required for allocation
6243  * returns the key for the extent through ins, and a tree buffer for
6244  * the first block of the extent through buf.
6245  *
6246  * returns the tree buffer or NULL.
6247  */
6248 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6249                                         struct btrfs_root *root, u32 blocksize,
6250                                         u64 parent, u64 root_objectid,
6251                                         struct btrfs_disk_key *key, int level,
6252                                         u64 hint, u64 empty_size, int for_cow)
6253 {
6254         struct btrfs_key ins;
6255         struct btrfs_block_rsv *block_rsv;
6256         struct extent_buffer *buf;
6257         u64 flags = 0;
6258         int ret;
6259
6260
6261         block_rsv = use_block_rsv(trans, root, blocksize);
6262         if (IS_ERR(block_rsv))
6263                 return ERR_CAST(block_rsv);
6264
6265         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6266                                    empty_size, hint, &ins, 0);
6267         if (ret) {
6268                 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6269                 return ERR_PTR(ret);
6270         }
6271
6272         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6273                                     blocksize, level);
6274         BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6275
6276         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6277                 if (parent == 0)
6278                         parent = ins.objectid;
6279                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6280         } else
6281                 BUG_ON(parent > 0);
6282
6283         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6284                 struct btrfs_delayed_extent_op *extent_op;
6285                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6286                 BUG_ON(!extent_op); /* -ENOMEM */
6287                 if (key)
6288                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
6289                 else
6290                         memset(&extent_op->key, 0, sizeof(extent_op->key));
6291                 extent_op->flags_to_set = flags;
6292                 extent_op->update_key = 1;
6293                 extent_op->update_flags = 1;
6294                 extent_op->is_data = 0;
6295
6296                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6297                                         ins.objectid,
6298                                         ins.offset, parent, root_objectid,
6299                                         level, BTRFS_ADD_DELAYED_EXTENT,
6300                                         extent_op, for_cow);
6301                 BUG_ON(ret); /* -ENOMEM */
6302         }
6303         return buf;
6304 }
6305
6306 struct walk_control {
6307         u64 refs[BTRFS_MAX_LEVEL];
6308         u64 flags[BTRFS_MAX_LEVEL];
6309         struct btrfs_key update_progress;
6310         int stage;
6311         int level;
6312         int shared_level;
6313         int update_ref;
6314         int keep_locks;
6315         int reada_slot;
6316         int reada_count;
6317         int for_reloc;
6318 };
6319
6320 #define DROP_REFERENCE  1
6321 #define UPDATE_BACKREF  2
6322
6323 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6324                                      struct btrfs_root *root,
6325                                      struct walk_control *wc,
6326                                      struct btrfs_path *path)
6327 {
6328         u64 bytenr;
6329         u64 generation;
6330         u64 refs;
6331         u64 flags;
6332         u32 nritems;
6333         u32 blocksize;
6334         struct btrfs_key key;
6335         struct extent_buffer *eb;
6336         int ret;
6337         int slot;
6338         int nread = 0;
6339
6340         if (path->slots[wc->level] < wc->reada_slot) {
6341                 wc->reada_count = wc->reada_count * 2 / 3;
6342                 wc->reada_count = max(wc->reada_count, 2);
6343         } else {
6344                 wc->reada_count = wc->reada_count * 3 / 2;
6345                 wc->reada_count = min_t(int, wc->reada_count,
6346                                         BTRFS_NODEPTRS_PER_BLOCK(root));
6347         }
6348
6349         eb = path->nodes[wc->level];
6350         nritems = btrfs_header_nritems(eb);
6351         blocksize = btrfs_level_size(root, wc->level - 1);
6352
6353         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6354                 if (nread >= wc->reada_count)
6355                         break;
6356
6357                 cond_resched();
6358                 bytenr = btrfs_node_blockptr(eb, slot);
6359                 generation = btrfs_node_ptr_generation(eb, slot);
6360
6361                 if (slot == path->slots[wc->level])
6362                         goto reada;
6363
6364                 if (wc->stage == UPDATE_BACKREF &&
6365                     generation <= root->root_key.offset)
6366                         continue;
6367
6368                 /* We don't lock the tree block, it's OK to be racy here */
6369                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6370                                                &refs, &flags);
6371                 /* We don't care about errors in readahead. */
6372                 if (ret < 0)
6373                         continue;
6374                 BUG_ON(refs == 0);
6375
6376                 if (wc->stage == DROP_REFERENCE) {
6377                         if (refs == 1)
6378                                 goto reada;
6379
6380                         if (wc->level == 1 &&
6381                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6382                                 continue;
6383                         if (!wc->update_ref ||
6384                             generation <= root->root_key.offset)
6385                                 continue;
6386                         btrfs_node_key_to_cpu(eb, &key, slot);
6387                         ret = btrfs_comp_cpu_keys(&key,
6388                                                   &wc->update_progress);
6389                         if (ret < 0)
6390                                 continue;
6391                 } else {
6392                         if (wc->level == 1 &&
6393                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6394                                 continue;
6395                 }
6396 reada:
6397                 ret = readahead_tree_block(root, bytenr, blocksize,
6398                                            generation);
6399                 if (ret)
6400                         break;
6401                 nread++;
6402         }
6403         wc->reada_slot = slot;
6404 }
6405
6406 /*
6407  * hepler to process tree block while walking down the tree.
6408  *
6409  * when wc->stage == UPDATE_BACKREF, this function updates
6410  * back refs for pointers in the block.
6411  *
6412  * NOTE: return value 1 means we should stop walking down.
6413  */
6414 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6415                                    struct btrfs_root *root,
6416                                    struct btrfs_path *path,
6417                                    struct walk_control *wc, int lookup_info)
6418 {
6419         int level = wc->level;
6420         struct extent_buffer *eb = path->nodes[level];
6421         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6422         int ret;
6423
6424         if (wc->stage == UPDATE_BACKREF &&
6425             btrfs_header_owner(eb) != root->root_key.objectid)
6426                 return 1;
6427
6428         /*
6429          * when reference count of tree block is 1, it won't increase
6430          * again. once full backref flag is set, we never clear it.
6431          */
6432         if (lookup_info &&
6433             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6434              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6435                 BUG_ON(!path->locks[level]);
6436                 ret = btrfs_lookup_extent_info(trans, root,
6437                                                eb->start, eb->len,
6438                                                &wc->refs[level],
6439                                                &wc->flags[level]);
6440                 BUG_ON(ret == -ENOMEM);
6441                 if (ret)
6442                         return ret;
6443                 BUG_ON(wc->refs[level] == 0);
6444         }
6445
6446         if (wc->stage == DROP_REFERENCE) {
6447                 if (wc->refs[level] > 1)
6448                         return 1;
6449
6450                 if (path->locks[level] && !wc->keep_locks) {
6451                         btrfs_tree_unlock_rw(eb, path->locks[level]);
6452                         path->locks[level] = 0;
6453                 }
6454                 return 0;
6455         }
6456
6457         /* wc->stage == UPDATE_BACKREF */
6458         if (!(wc->flags[level] & flag)) {
6459                 BUG_ON(!path->locks[level]);
6460                 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6461                 BUG_ON(ret); /* -ENOMEM */
6462                 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6463                 BUG_ON(ret); /* -ENOMEM */
6464                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6465                                                   eb->len, flag, 0);
6466                 BUG_ON(ret); /* -ENOMEM */
6467                 wc->flags[level] |= flag;
6468         }
6469
6470         /*
6471          * the block is shared by multiple trees, so it's not good to
6472          * keep the tree lock
6473          */
6474         if (path->locks[level] && level > 0) {
6475                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6476                 path->locks[level] = 0;
6477         }
6478         return 0;
6479 }
6480
6481 /*
6482  * hepler to process tree block pointer.
6483  *
6484  * when wc->stage == DROP_REFERENCE, this function checks
6485  * reference count of the block pointed to. if the block
6486  * is shared and we need update back refs for the subtree
6487  * rooted at the block, this function changes wc->stage to
6488  * UPDATE_BACKREF. if the block is shared and there is no
6489  * need to update back, this function drops the reference
6490  * to the block.
6491  *
6492  * NOTE: return value 1 means we should stop walking down.
6493  */
6494 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6495                                  struct btrfs_root *root,
6496                                  struct btrfs_path *path,
6497                                  struct walk_control *wc, int *lookup_info)
6498 {
6499         u64 bytenr;
6500         u64 generation;
6501         u64 parent;
6502         u32 blocksize;
6503         struct btrfs_key key;
6504         struct extent_buffer *next;
6505         int level = wc->level;
6506         int reada = 0;
6507         int ret = 0;
6508
6509         generation = btrfs_node_ptr_generation(path->nodes[level],
6510                                                path->slots[level]);
6511         /*
6512          * if the lower level block was created before the snapshot
6513          * was created, we know there is no need to update back refs
6514          * for the subtree
6515          */
6516         if (wc->stage == UPDATE_BACKREF &&
6517             generation <= root->root_key.offset) {
6518                 *lookup_info = 1;
6519                 return 1;
6520         }
6521
6522         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6523         blocksize = btrfs_level_size(root, level - 1);
6524
6525         next = btrfs_find_tree_block(root, bytenr, blocksize);
6526         if (!next) {
6527                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6528                 if (!next)
6529                         return -ENOMEM;
6530                 reada = 1;
6531         }
6532         btrfs_tree_lock(next);
6533         btrfs_set_lock_blocking(next);
6534
6535         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6536                                        &wc->refs[level - 1],
6537                                        &wc->flags[level - 1]);
6538         if (ret < 0) {
6539                 btrfs_tree_unlock(next);
6540                 return ret;
6541         }
6542
6543         BUG_ON(wc->refs[level - 1] == 0);
6544         *lookup_info = 0;
6545
6546         if (wc->stage == DROP_REFERENCE) {
6547                 if (wc->refs[level - 1] > 1) {
6548                         if (level == 1 &&
6549                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6550                                 goto skip;
6551
6552                         if (!wc->update_ref ||
6553                             generation <= root->root_key.offset)
6554                                 goto skip;
6555
6556                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6557                                               path->slots[level]);
6558                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6559                         if (ret < 0)
6560                                 goto skip;
6561
6562                         wc->stage = UPDATE_BACKREF;
6563                         wc->shared_level = level - 1;
6564                 }
6565         } else {
6566                 if (level == 1 &&
6567                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6568                         goto skip;
6569         }
6570
6571         if (!btrfs_buffer_uptodate(next, generation)) {
6572                 btrfs_tree_unlock(next);
6573                 free_extent_buffer(next);
6574                 next = NULL;
6575                 *lookup_info = 1;
6576         }
6577
6578         if (!next) {
6579                 if (reada && level == 1)
6580                         reada_walk_down(trans, root, wc, path);
6581                 next = read_tree_block(root, bytenr, blocksize, generation);
6582                 if (!next)
6583                         return -EIO;
6584                 btrfs_tree_lock(next);
6585                 btrfs_set_lock_blocking(next);
6586         }
6587
6588         level--;
6589         BUG_ON(level != btrfs_header_level(next));
6590         path->nodes[level] = next;
6591         path->slots[level] = 0;
6592         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6593         wc->level = level;
6594         if (wc->level == 1)
6595                 wc->reada_slot = 0;
6596         return 0;
6597 skip:
6598         wc->refs[level - 1] = 0;
6599         wc->flags[level - 1] = 0;
6600         if (wc->stage == DROP_REFERENCE) {
6601                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6602                         parent = path->nodes[level]->start;
6603                 } else {
6604                         BUG_ON(root->root_key.objectid !=
6605                                btrfs_header_owner(path->nodes[level]));
6606                         parent = 0;
6607                 }
6608
6609                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6610                                 root->root_key.objectid, level - 1, 0, 0);
6611                 BUG_ON(ret); /* -ENOMEM */
6612         }
6613         btrfs_tree_unlock(next);
6614         free_extent_buffer(next);
6615         *lookup_info = 1;
6616         return 1;
6617 }
6618
6619 /*
6620  * hepler to process tree block while walking up the tree.
6621  *
6622  * when wc->stage == DROP_REFERENCE, this function drops
6623  * reference count on the block.
6624  *
6625  * when wc->stage == UPDATE_BACKREF, this function changes
6626  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6627  * to UPDATE_BACKREF previously while processing the block.
6628  *
6629  * NOTE: return value 1 means we should stop walking up.
6630  */
6631 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6632                                  struct btrfs_root *root,
6633                                  struct btrfs_path *path,
6634                                  struct walk_control *wc)
6635 {
6636         int ret;
6637         int level = wc->level;
6638         struct extent_buffer *eb = path->nodes[level];
6639         u64 parent = 0;
6640
6641         if (wc->stage == UPDATE_BACKREF) {
6642                 BUG_ON(wc->shared_level < level);
6643                 if (level < wc->shared_level)
6644                         goto out;
6645
6646                 ret = find_next_key(path, level + 1, &wc->update_progress);
6647                 if (ret > 0)
6648                         wc->update_ref = 0;
6649
6650                 wc->stage = DROP_REFERENCE;
6651                 wc->shared_level = -1;
6652                 path->slots[level] = 0;
6653
6654                 /*
6655                  * check reference count again if the block isn't locked.
6656                  * we should start walking down the tree again if reference
6657                  * count is one.
6658                  */
6659                 if (!path->locks[level]) {
6660                         BUG_ON(level == 0);
6661                         btrfs_tree_lock(eb);
6662                         btrfs_set_lock_blocking(eb);
6663                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6664
6665                         ret = btrfs_lookup_extent_info(trans, root,
6666                                                        eb->start, eb->len,
6667                                                        &wc->refs[level],
6668                                                        &wc->flags[level]);
6669                         if (ret < 0) {
6670                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6671                                 return ret;
6672                         }
6673                         BUG_ON(wc->refs[level] == 0);
6674                         if (wc->refs[level] == 1) {
6675                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6676                                 return 1;
6677                         }
6678                 }
6679         }
6680
6681         /* wc->stage == DROP_REFERENCE */
6682         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6683
6684         if (wc->refs[level] == 1) {
6685                 if (level == 0) {
6686                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6687                                 ret = btrfs_dec_ref(trans, root, eb, 1,
6688                                                     wc->for_reloc);
6689                         else
6690                                 ret = btrfs_dec_ref(trans, root, eb, 0,
6691                                                     wc->for_reloc);
6692                         BUG_ON(ret); /* -ENOMEM */
6693                 }
6694                 /* make block locked assertion in clean_tree_block happy */
6695                 if (!path->locks[level] &&
6696                     btrfs_header_generation(eb) == trans->transid) {
6697                         btrfs_tree_lock(eb);
6698                         btrfs_set_lock_blocking(eb);
6699                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6700                 }
6701                 clean_tree_block(trans, root, eb);
6702         }
6703
6704         if (eb == root->node) {
6705                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6706                         parent = eb->start;
6707                 else
6708                         BUG_ON(root->root_key.objectid !=
6709                                btrfs_header_owner(eb));
6710         } else {
6711                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6712                         parent = path->nodes[level + 1]->start;
6713                 else
6714                         BUG_ON(root->root_key.objectid !=
6715                                btrfs_header_owner(path->nodes[level + 1]));
6716         }
6717
6718         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0);
6719 out:
6720         wc->refs[level] = 0;
6721         wc->flags[level] = 0;
6722         return 0;
6723 }
6724
6725 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6726                                    struct btrfs_root *root,
6727                                    struct btrfs_path *path,
6728                                    struct walk_control *wc)
6729 {
6730         int level = wc->level;
6731         int lookup_info = 1;
6732         int ret;
6733
6734         while (level >= 0) {
6735                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6736                 if (ret > 0)
6737                         break;
6738
6739                 if (level == 0)
6740                         break;
6741
6742                 if (path->slots[level] >=
6743                     btrfs_header_nritems(path->nodes[level]))
6744                         break;
6745
6746                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6747                 if (ret > 0) {
6748                         path->slots[level]++;
6749                         continue;
6750                 } else if (ret < 0)
6751                         return ret;
6752                 level = wc->level;
6753         }
6754         return 0;
6755 }
6756
6757 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6758                                  struct btrfs_root *root,
6759                                  struct btrfs_path *path,
6760                                  struct walk_control *wc, int max_level)
6761 {
6762         int level = wc->level;
6763         int ret;
6764
6765         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6766         while (level < max_level && path->nodes[level]) {
6767                 wc->level = level;
6768                 if (path->slots[level] + 1 <
6769                     btrfs_header_nritems(path->nodes[level])) {
6770                         path->slots[level]++;
6771                         return 0;
6772                 } else {
6773                         ret = walk_up_proc(trans, root, path, wc);
6774                         if (ret > 0)
6775                                 return 0;
6776
6777                         if (path->locks[level]) {
6778                                 btrfs_tree_unlock_rw(path->nodes[level],
6779                                                      path->locks[level]);
6780                                 path->locks[level] = 0;
6781                         }
6782                         free_extent_buffer(path->nodes[level]);
6783                         path->nodes[level] = NULL;
6784                         level++;
6785                 }
6786         }
6787         return 1;
6788 }
6789
6790 /*
6791  * drop a subvolume tree.
6792  *
6793  * this function traverses the tree freeing any blocks that only
6794  * referenced by the tree.
6795  *
6796  * when a shared tree block is found. this function decreases its
6797  * reference count by one. if update_ref is true, this function
6798  * also make sure backrefs for the shared block and all lower level
6799  * blocks are properly updated.
6800  */
6801 int btrfs_drop_snapshot(struct btrfs_root *root,
6802                          struct btrfs_block_rsv *block_rsv, int update_ref,
6803                          int for_reloc)
6804 {
6805         struct btrfs_path *path;
6806         struct btrfs_trans_handle *trans;
6807         struct btrfs_root *tree_root = root->fs_info->tree_root;
6808         struct btrfs_root_item *root_item = &root->root_item;
6809         struct walk_control *wc;
6810         struct btrfs_key key;
6811         int err = 0;
6812         int ret;
6813         int level;
6814
6815         path = btrfs_alloc_path();
6816         if (!path) {
6817                 err = -ENOMEM;
6818                 goto out;
6819         }
6820
6821         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6822         if (!wc) {
6823                 btrfs_free_path(path);
6824                 err = -ENOMEM;
6825                 goto out;
6826         }
6827
6828         trans = btrfs_start_transaction(tree_root, 0);
6829         if (IS_ERR(trans)) {
6830                 err = PTR_ERR(trans);
6831                 goto out_free;
6832         }
6833
6834         if (block_rsv)
6835                 trans->block_rsv = block_rsv;
6836
6837         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6838                 level = btrfs_header_level(root->node);
6839                 path->nodes[level] = btrfs_lock_root_node(root);
6840                 btrfs_set_lock_blocking(path->nodes[level]);
6841                 path->slots[level] = 0;
6842                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6843                 memset(&wc->update_progress, 0,
6844                        sizeof(wc->update_progress));
6845         } else {
6846                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6847                 memcpy(&wc->update_progress, &key,
6848                        sizeof(wc->update_progress));
6849
6850                 level = root_item->drop_level;
6851                 BUG_ON(level == 0);
6852                 path->lowest_level = level;
6853                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6854                 path->lowest_level = 0;
6855                 if (ret < 0) {
6856                         err = ret;
6857                         goto out_end_trans;
6858                 }
6859                 WARN_ON(ret > 0);
6860
6861                 /*
6862                  * unlock our path, this is safe because only this
6863                  * function is allowed to delete this snapshot
6864                  */
6865                 btrfs_unlock_up_safe(path, 0);
6866
6867                 level = btrfs_header_level(root->node);
6868                 while (1) {
6869                         btrfs_tree_lock(path->nodes[level]);
6870                         btrfs_set_lock_blocking(path->nodes[level]);
6871
6872                         ret = btrfs_lookup_extent_info(trans, root,
6873                                                 path->nodes[level]->start,
6874                                                 path->nodes[level]->len,
6875                                                 &wc->refs[level],
6876                                                 &wc->flags[level]);
6877                         if (ret < 0) {
6878                                 err = ret;
6879                                 goto out_end_trans;
6880                         }
6881                         BUG_ON(wc->refs[level] == 0);
6882
6883                         if (level == root_item->drop_level)
6884                                 break;
6885
6886                         btrfs_tree_unlock(path->nodes[level]);
6887                         WARN_ON(wc->refs[level] != 1);
6888                         level--;
6889                 }
6890         }
6891
6892         wc->level = level;
6893         wc->shared_level = -1;
6894         wc->stage = DROP_REFERENCE;
6895         wc->update_ref = update_ref;
6896         wc->keep_locks = 0;
6897         wc->for_reloc = for_reloc;
6898         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6899
6900         while (1) {
6901                 ret = walk_down_tree(trans, root, path, wc);
6902                 if (ret < 0) {
6903                         err = ret;
6904                         break;
6905                 }
6906
6907                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6908                 if (ret < 0) {
6909                         err = ret;
6910                         break;
6911                 }
6912
6913                 if (ret > 0) {
6914                         BUG_ON(wc->stage != DROP_REFERENCE);
6915                         break;
6916                 }
6917
6918                 if (wc->stage == DROP_REFERENCE) {
6919                         level = wc->level;
6920                         btrfs_node_key(path->nodes[level],
6921                                        &root_item->drop_progress,
6922                                        path->slots[level]);
6923                         root_item->drop_level = level;
6924                 }
6925
6926                 BUG_ON(wc->level == 0);
6927                 if (btrfs_should_end_transaction(trans, tree_root)) {
6928                         ret = btrfs_update_root(trans, tree_root,
6929                                                 &root->root_key,
6930                                                 root_item);
6931                         if (ret) {
6932                                 btrfs_abort_transaction(trans, tree_root, ret);
6933                                 err = ret;
6934                                 goto out_end_trans;
6935                         }
6936
6937                         btrfs_end_transaction_throttle(trans, tree_root);
6938                         trans = btrfs_start_transaction(tree_root, 0);
6939                         if (IS_ERR(trans)) {
6940                                 err = PTR_ERR(trans);
6941                                 goto out_free;
6942                         }
6943                         if (block_rsv)
6944                                 trans->block_rsv = block_rsv;
6945                 }
6946         }
6947         btrfs_release_path(path);
6948         if (err)
6949                 goto out_end_trans;
6950
6951         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6952         if (ret) {
6953                 btrfs_abort_transaction(trans, tree_root, ret);
6954                 goto out_end_trans;
6955         }
6956
6957         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6958                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6959                                            NULL, NULL);
6960                 if (ret < 0) {
6961                         btrfs_abort_transaction(trans, tree_root, ret);
6962                         err = ret;
6963                         goto out_end_trans;
6964                 } else if (ret > 0) {
6965                         /* if we fail to delete the orphan item this time
6966                          * around, it'll get picked up the next time.
6967                          *
6968                          * The most common failure here is just -ENOENT.
6969                          */
6970                         btrfs_del_orphan_item(trans, tree_root,
6971                                               root->root_key.objectid);
6972                 }
6973         }
6974
6975         if (root->in_radix) {
6976                 btrfs_free_fs_root(tree_root->fs_info, root);
6977         } else {
6978                 free_extent_buffer(root->node);
6979                 free_extent_buffer(root->commit_root);
6980                 kfree(root);
6981         }
6982 out_end_trans:
6983         btrfs_end_transaction_throttle(trans, tree_root);
6984 out_free:
6985         kfree(wc);
6986         btrfs_free_path(path);
6987 out:
6988         if (err)
6989                 btrfs_std_error(root->fs_info, err);
6990         return err;
6991 }
6992
6993 /*
6994  * drop subtree rooted at tree block 'node'.
6995  *
6996  * NOTE: this function will unlock and release tree block 'node'
6997  * only used by relocation code
6998  */
6999 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7000                         struct btrfs_root *root,
7001                         struct extent_buffer *node,
7002                         struct extent_buffer *parent)
7003 {
7004         struct btrfs_path *path;
7005         struct walk_control *wc;
7006         int level;
7007         int parent_level;
7008         int ret = 0;
7009         int wret;
7010
7011         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7012
7013         path = btrfs_alloc_path();
7014         if (!path)
7015                 return -ENOMEM;
7016
7017         wc = kzalloc(sizeof(*wc), GFP_NOFS);
7018         if (!wc) {
7019                 btrfs_free_path(path);
7020                 return -ENOMEM;
7021         }
7022
7023         btrfs_assert_tree_locked(parent);
7024         parent_level = btrfs_header_level(parent);
7025         extent_buffer_get(parent);
7026         path->nodes[parent_level] = parent;
7027         path->slots[parent_level] = btrfs_header_nritems(parent);
7028
7029         btrfs_assert_tree_locked(node);
7030         level = btrfs_header_level(node);
7031         path->nodes[level] = node;
7032         path->slots[level] = 0;
7033         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7034
7035         wc->refs[parent_level] = 1;
7036         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7037         wc->level = level;
7038         wc->shared_level = -1;
7039         wc->stage = DROP_REFERENCE;
7040         wc->update_ref = 0;
7041         wc->keep_locks = 1;
7042         wc->for_reloc = 1;
7043         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7044
7045         while (1) {
7046                 wret = walk_down_tree(trans, root, path, wc);
7047                 if (wret < 0) {
7048                         ret = wret;
7049                         break;
7050                 }
7051
7052                 wret = walk_up_tree(trans, root, path, wc, parent_level);
7053                 if (wret < 0)
7054                         ret = wret;
7055                 if (wret != 0)
7056                         break;
7057         }
7058
7059         kfree(wc);
7060         btrfs_free_path(path);
7061         return ret;
7062 }
7063
7064 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7065 {
7066         u64 num_devices;
7067         u64 stripped;
7068
7069         /*
7070          * if restripe for this chunk_type is on pick target profile and
7071          * return, otherwise do the usual balance
7072          */
7073         stripped = get_restripe_target(root->fs_info, flags);
7074         if (stripped)
7075                 return extended_to_chunk(stripped);
7076
7077         /*
7078          * we add in the count of missing devices because we want
7079          * to make sure that any RAID levels on a degraded FS
7080          * continue to be honored.
7081          */
7082         num_devices = root->fs_info->fs_devices->rw_devices +
7083                 root->fs_info->fs_devices->missing_devices;
7084
7085         stripped = BTRFS_BLOCK_GROUP_RAID0 |
7086                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7087
7088         if (num_devices == 1) {
7089                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7090                 stripped = flags & ~stripped;
7091
7092                 /* turn raid0 into single device chunks */
7093                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7094                         return stripped;
7095
7096                 /* turn mirroring into duplication */
7097                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7098                              BTRFS_BLOCK_GROUP_RAID10))
7099                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7100         } else {
7101                 /* they already had raid on here, just return */
7102                 if (flags & stripped)
7103                         return flags;
7104
7105                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7106                 stripped = flags & ~stripped;
7107
7108                 /* switch duplicated blocks with raid1 */
7109                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7110                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7111
7112                 /* this is drive concat, leave it alone */
7113         }
7114
7115         return flags;
7116 }
7117
7118 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7119 {
7120         struct btrfs_space_info *sinfo = cache->space_info;
7121         u64 num_bytes;
7122         u64 min_allocable_bytes;
7123         int ret = -ENOSPC;
7124
7125
7126         /*
7127          * We need some metadata space and system metadata space for
7128          * allocating chunks in some corner cases until we force to set
7129          * it to be readonly.
7130          */
7131         if ((sinfo->flags &
7132              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7133             !force)
7134                 min_allocable_bytes = 1 * 1024 * 1024;
7135         else
7136                 min_allocable_bytes = 0;
7137
7138         spin_lock(&sinfo->lock);
7139         spin_lock(&cache->lock);
7140
7141         if (cache->ro) {
7142                 ret = 0;
7143                 goto out;
7144         }
7145
7146         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7147                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7148
7149         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7150             sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7151             min_allocable_bytes <= sinfo->total_bytes) {
7152                 sinfo->bytes_readonly += num_bytes;
7153                 cache->ro = 1;
7154                 ret = 0;
7155         }
7156 out:
7157         spin_unlock(&cache->lock);
7158         spin_unlock(&sinfo->lock);
7159         return ret;
7160 }
7161
7162 int btrfs_set_block_group_ro(struct btrfs_root *root,
7163                              struct btrfs_block_group_cache *cache)
7164
7165 {
7166         struct btrfs_trans_handle *trans;
7167         u64 alloc_flags;
7168         int ret;
7169
7170         BUG_ON(cache->ro);
7171
7172         trans = btrfs_join_transaction(root);
7173         if (IS_ERR(trans))
7174                 return PTR_ERR(trans);
7175
7176         alloc_flags = update_block_group_flags(root, cache->flags);
7177         if (alloc_flags != cache->flags) {
7178                 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7179                                      CHUNK_ALLOC_FORCE);
7180                 if (ret < 0)
7181                         goto out;
7182         }
7183
7184         ret = set_block_group_ro(cache, 0);
7185         if (!ret)
7186                 goto out;
7187         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7188         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7189                              CHUNK_ALLOC_FORCE);
7190         if (ret < 0)
7191                 goto out;
7192         ret = set_block_group_ro(cache, 0);
7193 out:
7194         btrfs_end_transaction(trans, root);
7195         return ret;
7196 }
7197
7198 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7199                             struct btrfs_root *root, u64 type)
7200 {
7201         u64 alloc_flags = get_alloc_profile(root, type);
7202         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7203                               CHUNK_ALLOC_FORCE);
7204 }
7205
7206 /*
7207  * helper to account the unused space of all the readonly block group in the
7208  * list. takes mirrors into account.
7209  */
7210 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7211 {
7212         struct btrfs_block_group_cache *block_group;
7213         u64 free_bytes = 0;
7214         int factor;
7215
7216         list_for_each_entry(block_group, groups_list, list) {
7217                 spin_lock(&block_group->lock);
7218
7219                 if (!block_group->ro) {
7220                         spin_unlock(&block_group->lock);
7221                         continue;
7222                 }
7223
7224                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7225                                           BTRFS_BLOCK_GROUP_RAID10 |
7226                                           BTRFS_BLOCK_GROUP_DUP))
7227                         factor = 2;
7228                 else
7229                         factor = 1;
7230
7231                 free_bytes += (block_group->key.offset -
7232                                btrfs_block_group_used(&block_group->item)) *
7233                                factor;
7234
7235                 spin_unlock(&block_group->lock);
7236         }
7237
7238         return free_bytes;
7239 }
7240
7241 /*
7242  * helper to account the unused space of all the readonly block group in the
7243  * space_info. takes mirrors into account.
7244  */
7245 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7246 {
7247         int i;
7248         u64 free_bytes = 0;
7249
7250         spin_lock(&sinfo->lock);
7251
7252         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7253                 if (!list_empty(&sinfo->block_groups[i]))
7254                         free_bytes += __btrfs_get_ro_block_group_free_space(
7255                                                 &sinfo->block_groups[i]);
7256
7257         spin_unlock(&sinfo->lock);
7258
7259         return free_bytes;
7260 }
7261
7262 void btrfs_set_block_group_rw(struct btrfs_root *root,
7263                               struct btrfs_block_group_cache *cache)
7264 {
7265         struct btrfs_space_info *sinfo = cache->space_info;
7266         u64 num_bytes;
7267
7268         BUG_ON(!cache->ro);
7269
7270         spin_lock(&sinfo->lock);
7271         spin_lock(&cache->lock);
7272         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7273                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7274         sinfo->bytes_readonly -= num_bytes;
7275         cache->ro = 0;
7276         spin_unlock(&cache->lock);
7277         spin_unlock(&sinfo->lock);
7278 }
7279
7280 /*
7281  * checks to see if its even possible to relocate this block group.
7282  *
7283  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7284  * ok to go ahead and try.
7285  */
7286 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7287 {
7288         struct btrfs_block_group_cache *block_group;
7289         struct btrfs_space_info *space_info;
7290         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7291         struct btrfs_device *device;
7292         u64 min_free;
7293         u64 dev_min = 1;
7294         u64 dev_nr = 0;
7295         u64 target;
7296         int index;
7297         int full = 0;
7298         int ret = 0;
7299
7300         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7301
7302         /* odd, couldn't find the block group, leave it alone */
7303         if (!block_group)
7304                 return -1;
7305
7306         min_free = btrfs_block_group_used(&block_group->item);
7307
7308         /* no bytes used, we're good */
7309         if (!min_free)
7310                 goto out;
7311
7312         space_info = block_group->space_info;
7313         spin_lock(&space_info->lock);
7314
7315         full = space_info->full;
7316
7317         /*
7318          * if this is the last block group we have in this space, we can't
7319          * relocate it unless we're able to allocate a new chunk below.
7320          *
7321          * Otherwise, we need to make sure we have room in the space to handle
7322          * all of the extents from this block group.  If we can, we're good
7323          */
7324         if ((space_info->total_bytes != block_group->key.offset) &&
7325             (space_info->bytes_used + space_info->bytes_reserved +
7326              space_info->bytes_pinned + space_info->bytes_readonly +
7327              min_free < space_info->total_bytes)) {
7328                 spin_unlock(&space_info->lock);
7329                 goto out;
7330         }
7331         spin_unlock(&space_info->lock);
7332
7333         /*
7334          * ok we don't have enough space, but maybe we have free space on our
7335          * devices to allocate new chunks for relocation, so loop through our
7336          * alloc devices and guess if we have enough space.  if this block
7337          * group is going to be restriped, run checks against the target
7338          * profile instead of the current one.
7339          */
7340         ret = -1;
7341
7342         /*
7343          * index:
7344          *      0: raid10
7345          *      1: raid1
7346          *      2: dup
7347          *      3: raid0
7348          *      4: single
7349          */
7350         target = get_restripe_target(root->fs_info, block_group->flags);
7351         if (target) {
7352                 index = __get_block_group_index(extended_to_chunk(target));
7353         } else {
7354                 /*
7355                  * this is just a balance, so if we were marked as full
7356                  * we know there is no space for a new chunk
7357                  */
7358                 if (full)
7359                         goto out;
7360
7361                 index = get_block_group_index(block_group);
7362         }
7363
7364         if (index == 0) {
7365                 dev_min = 4;
7366                 /* Divide by 2 */
7367                 min_free >>= 1;
7368         } else if (index == 1) {
7369                 dev_min = 2;
7370         } else if (index == 2) {
7371                 /* Multiply by 2 */
7372                 min_free <<= 1;
7373         } else if (index == 3) {
7374                 dev_min = fs_devices->rw_devices;
7375                 do_div(min_free, dev_min);
7376         }
7377
7378         mutex_lock(&root->fs_info->chunk_mutex);
7379         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7380                 u64 dev_offset;
7381
7382                 /*
7383                  * check to make sure we can actually find a chunk with enough
7384                  * space to fit our block group in.
7385                  */
7386                 if (device->total_bytes > device->bytes_used + min_free) {
7387                         ret = find_free_dev_extent(device, min_free,
7388                                                    &dev_offset, NULL);
7389                         if (!ret)
7390                                 dev_nr++;
7391
7392                         if (dev_nr >= dev_min)
7393                                 break;
7394
7395                         ret = -1;
7396                 }
7397         }
7398         mutex_unlock(&root->fs_info->chunk_mutex);
7399 out:
7400         btrfs_put_block_group(block_group);
7401         return ret;
7402 }
7403
7404 static int find_first_block_group(struct btrfs_root *root,
7405                 struct btrfs_path *path, struct btrfs_key *key)
7406 {
7407         int ret = 0;
7408         struct btrfs_key found_key;
7409         struct extent_buffer *leaf;
7410         int slot;
7411
7412         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7413         if (ret < 0)
7414                 goto out;
7415
7416         while (1) {
7417                 slot = path->slots[0];
7418                 leaf = path->nodes[0];
7419                 if (slot >= btrfs_header_nritems(leaf)) {
7420                         ret = btrfs_next_leaf(root, path);
7421                         if (ret == 0)
7422                                 continue;
7423                         if (ret < 0)
7424                                 goto out;
7425                         break;
7426                 }
7427                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7428
7429                 if (found_key.objectid >= key->objectid &&
7430                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7431                         ret = 0;
7432                         goto out;
7433                 }
7434                 path->slots[0]++;
7435         }
7436 out:
7437         return ret;
7438 }
7439
7440 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7441 {
7442         struct btrfs_block_group_cache *block_group;
7443         u64 last = 0;
7444
7445         while (1) {
7446                 struct inode *inode;
7447
7448                 block_group = btrfs_lookup_first_block_group(info, last);
7449                 while (block_group) {
7450                         spin_lock(&block_group->lock);
7451                         if (block_group->iref)
7452                                 break;
7453                         spin_unlock(&block_group->lock);
7454                         block_group = next_block_group(info->tree_root,
7455                                                        block_group);
7456                 }
7457                 if (!block_group) {
7458                         if (last == 0)
7459                                 break;
7460                         last = 0;
7461                         continue;
7462                 }
7463
7464                 inode = block_group->inode;
7465                 block_group->iref = 0;
7466                 block_group->inode = NULL;
7467                 spin_unlock(&block_group->lock);
7468                 iput(inode);
7469                 last = block_group->key.objectid + block_group->key.offset;
7470                 btrfs_put_block_group(block_group);
7471         }
7472 }
7473
7474 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7475 {
7476         struct btrfs_block_group_cache *block_group;
7477         struct btrfs_space_info *space_info;
7478         struct btrfs_caching_control *caching_ctl;
7479         struct rb_node *n;
7480
7481         down_write(&info->extent_commit_sem);
7482         while (!list_empty(&info->caching_block_groups)) {
7483                 caching_ctl = list_entry(info->caching_block_groups.next,
7484                                          struct btrfs_caching_control, list);
7485                 list_del(&caching_ctl->list);
7486                 put_caching_control(caching_ctl);
7487         }
7488         up_write(&info->extent_commit_sem);
7489
7490         spin_lock(&info->block_group_cache_lock);
7491         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7492                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7493                                        cache_node);
7494                 rb_erase(&block_group->cache_node,
7495                          &info->block_group_cache_tree);
7496                 spin_unlock(&info->block_group_cache_lock);
7497
7498                 down_write(&block_group->space_info->groups_sem);
7499                 list_del(&block_group->list);
7500                 up_write(&block_group->space_info->groups_sem);
7501
7502                 if (block_group->cached == BTRFS_CACHE_STARTED)
7503                         wait_block_group_cache_done(block_group);
7504
7505                 /*
7506                  * We haven't cached this block group, which means we could
7507                  * possibly have excluded extents on this block group.
7508                  */
7509                 if (block_group->cached == BTRFS_CACHE_NO)
7510                         free_excluded_extents(info->extent_root, block_group);
7511
7512                 btrfs_remove_free_space_cache(block_group);
7513                 btrfs_put_block_group(block_group);
7514
7515                 spin_lock(&info->block_group_cache_lock);
7516         }
7517         spin_unlock(&info->block_group_cache_lock);
7518
7519         /* now that all the block groups are freed, go through and
7520          * free all the space_info structs.  This is only called during
7521          * the final stages of unmount, and so we know nobody is
7522          * using them.  We call synchronize_rcu() once before we start,
7523          * just to be on the safe side.
7524          */
7525         synchronize_rcu();
7526
7527         release_global_block_rsv(info);
7528
7529         while(!list_empty(&info->space_info)) {
7530                 space_info = list_entry(info->space_info.next,
7531                                         struct btrfs_space_info,
7532                                         list);
7533                 if (space_info->bytes_pinned > 0 ||
7534                     space_info->bytes_reserved > 0 ||
7535                     space_info->bytes_may_use > 0) {
7536                         WARN_ON(1);
7537                         dump_space_info(space_info, 0, 0);
7538                 }
7539                 list_del(&space_info->list);
7540                 kfree(space_info);
7541         }
7542         return 0;
7543 }
7544
7545 static void __link_block_group(struct btrfs_space_info *space_info,
7546                                struct btrfs_block_group_cache *cache)
7547 {
7548         int index = get_block_group_index(cache);
7549
7550         down_write(&space_info->groups_sem);
7551         list_add_tail(&cache->list, &space_info->block_groups[index]);
7552         up_write(&space_info->groups_sem);
7553 }
7554
7555 int btrfs_read_block_groups(struct btrfs_root *root)
7556 {
7557         struct btrfs_path *path;
7558         int ret;
7559         struct btrfs_block_group_cache *cache;
7560         struct btrfs_fs_info *info = root->fs_info;
7561         struct btrfs_space_info *space_info;
7562         struct btrfs_key key;
7563         struct btrfs_key found_key;
7564         struct extent_buffer *leaf;
7565         int need_clear = 0;
7566         u64 cache_gen;
7567
7568         root = info->extent_root;
7569         key.objectid = 0;
7570         key.offset = 0;
7571         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7572         path = btrfs_alloc_path();
7573         if (!path)
7574                 return -ENOMEM;
7575         path->reada = 1;
7576
7577         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7578         if (btrfs_test_opt(root, SPACE_CACHE) &&
7579             btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7580                 need_clear = 1;
7581         if (btrfs_test_opt(root, CLEAR_CACHE))
7582                 need_clear = 1;
7583
7584         while (1) {
7585                 ret = find_first_block_group(root, path, &key);
7586                 if (ret > 0)
7587                         break;
7588                 if (ret != 0)
7589                         goto error;
7590                 leaf = path->nodes[0];
7591                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7592                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7593                 if (!cache) {
7594                         ret = -ENOMEM;
7595                         goto error;
7596                 }
7597                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7598                                                 GFP_NOFS);
7599                 if (!cache->free_space_ctl) {
7600                         kfree(cache);
7601                         ret = -ENOMEM;
7602                         goto error;
7603                 }
7604
7605                 atomic_set(&cache->count, 1);
7606                 spin_lock_init(&cache->lock);
7607                 cache->fs_info = info;
7608                 INIT_LIST_HEAD(&cache->list);
7609                 INIT_LIST_HEAD(&cache->cluster_list);
7610
7611                 if (need_clear)
7612                         cache->disk_cache_state = BTRFS_DC_CLEAR;
7613
7614                 read_extent_buffer(leaf, &cache->item,
7615                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7616                                    sizeof(cache->item));
7617                 memcpy(&cache->key, &found_key, sizeof(found_key));
7618
7619                 key.objectid = found_key.objectid + found_key.offset;
7620                 btrfs_release_path(path);
7621                 cache->flags = btrfs_block_group_flags(&cache->item);
7622                 cache->sectorsize = root->sectorsize;
7623
7624                 btrfs_init_free_space_ctl(cache);
7625
7626                 /*
7627                  * We need to exclude the super stripes now so that the space
7628                  * info has super bytes accounted for, otherwise we'll think
7629                  * we have more space than we actually do.
7630                  */
7631                 exclude_super_stripes(root, cache);
7632
7633                 /*
7634                  * check for two cases, either we are full, and therefore
7635                  * don't need to bother with the caching work since we won't
7636                  * find any space, or we are empty, and we can just add all
7637                  * the space in and be done with it.  This saves us _alot_ of
7638                  * time, particularly in the full case.
7639                  */
7640                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7641                         cache->last_byte_to_unpin = (u64)-1;
7642                         cache->cached = BTRFS_CACHE_FINISHED;
7643                         free_excluded_extents(root, cache);
7644                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7645                         cache->last_byte_to_unpin = (u64)-1;
7646                         cache->cached = BTRFS_CACHE_FINISHED;
7647                         add_new_free_space(cache, root->fs_info,
7648                                            found_key.objectid,
7649                                            found_key.objectid +
7650                                            found_key.offset);
7651                         free_excluded_extents(root, cache);
7652                 }
7653
7654                 ret = update_space_info(info, cache->flags, found_key.offset,
7655                                         btrfs_block_group_used(&cache->item),
7656                                         &space_info);
7657                 BUG_ON(ret); /* -ENOMEM */
7658                 cache->space_info = space_info;
7659                 spin_lock(&cache->space_info->lock);
7660                 cache->space_info->bytes_readonly += cache->bytes_super;
7661                 spin_unlock(&cache->space_info->lock);
7662
7663                 __link_block_group(space_info, cache);
7664
7665                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7666                 BUG_ON(ret); /* Logic error */
7667
7668                 set_avail_alloc_bits(root->fs_info, cache->flags);
7669                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7670                         set_block_group_ro(cache, 1);
7671         }
7672
7673         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7674                 if (!(get_alloc_profile(root, space_info->flags) &
7675                       (BTRFS_BLOCK_GROUP_RAID10 |
7676                        BTRFS_BLOCK_GROUP_RAID1 |
7677                        BTRFS_BLOCK_GROUP_DUP)))
7678                         continue;
7679                 /*
7680                  * avoid allocating from un-mirrored block group if there are
7681                  * mirrored block groups.
7682                  */
7683                 list_for_each_entry(cache, &space_info->block_groups[3], list)
7684                         set_block_group_ro(cache, 1);
7685                 list_for_each_entry(cache, &space_info->block_groups[4], list)
7686                         set_block_group_ro(cache, 1);
7687         }
7688
7689         init_global_block_rsv(info);
7690         ret = 0;
7691 error:
7692         btrfs_free_path(path);
7693         return ret;
7694 }
7695
7696 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7697                            struct btrfs_root *root, u64 bytes_used,
7698                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7699                            u64 size)
7700 {
7701         int ret;
7702         struct btrfs_root *extent_root;
7703         struct btrfs_block_group_cache *cache;
7704
7705         extent_root = root->fs_info->extent_root;
7706
7707         root->fs_info->last_trans_log_full_commit = trans->transid;
7708
7709         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7710         if (!cache)
7711                 return -ENOMEM;
7712         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7713                                         GFP_NOFS);
7714         if (!cache->free_space_ctl) {
7715                 kfree(cache);
7716                 return -ENOMEM;
7717         }
7718
7719         cache->key.objectid = chunk_offset;
7720         cache->key.offset = size;
7721         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7722         cache->sectorsize = root->sectorsize;
7723         cache->fs_info = root->fs_info;
7724
7725         atomic_set(&cache->count, 1);
7726         spin_lock_init(&cache->lock);
7727         INIT_LIST_HEAD(&cache->list);
7728         INIT_LIST_HEAD(&cache->cluster_list);
7729
7730         btrfs_init_free_space_ctl(cache);
7731
7732         btrfs_set_block_group_used(&cache->item, bytes_used);
7733         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7734         cache->flags = type;
7735         btrfs_set_block_group_flags(&cache->item, type);
7736
7737         cache->last_byte_to_unpin = (u64)-1;
7738         cache->cached = BTRFS_CACHE_FINISHED;
7739         exclude_super_stripes(root, cache);
7740
7741         add_new_free_space(cache, root->fs_info, chunk_offset,
7742                            chunk_offset + size);
7743
7744         free_excluded_extents(root, cache);
7745
7746         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7747                                 &cache->space_info);
7748         BUG_ON(ret); /* -ENOMEM */
7749         update_global_block_rsv(root->fs_info);
7750
7751         spin_lock(&cache->space_info->lock);
7752         cache->space_info->bytes_readonly += cache->bytes_super;
7753         spin_unlock(&cache->space_info->lock);
7754
7755         __link_block_group(cache->space_info, cache);
7756
7757         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7758         BUG_ON(ret); /* Logic error */
7759
7760         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7761                                 sizeof(cache->item));
7762         if (ret) {
7763                 btrfs_abort_transaction(trans, extent_root, ret);
7764                 return ret;
7765         }
7766
7767         set_avail_alloc_bits(extent_root->fs_info, type);
7768
7769         return 0;
7770 }
7771
7772 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7773 {
7774         u64 extra_flags = chunk_to_extended(flags) &
7775                                 BTRFS_EXTENDED_PROFILE_MASK;
7776
7777         if (flags & BTRFS_BLOCK_GROUP_DATA)
7778                 fs_info->avail_data_alloc_bits &= ~extra_flags;
7779         if (flags & BTRFS_BLOCK_GROUP_METADATA)
7780                 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7781         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7782                 fs_info->avail_system_alloc_bits &= ~extra_flags;
7783 }
7784
7785 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7786                              struct btrfs_root *root, u64 group_start)
7787 {
7788         struct btrfs_path *path;
7789         struct btrfs_block_group_cache *block_group;
7790         struct btrfs_free_cluster *cluster;
7791         struct btrfs_root *tree_root = root->fs_info->tree_root;
7792         struct btrfs_key key;
7793         struct inode *inode;
7794         int ret;
7795         int index;
7796         int factor;
7797
7798         root = root->fs_info->extent_root;
7799
7800         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7801         BUG_ON(!block_group);
7802         BUG_ON(!block_group->ro);
7803
7804         /*
7805          * Free the reserved super bytes from this block group before
7806          * remove it.
7807          */
7808         free_excluded_extents(root, block_group);
7809
7810         memcpy(&key, &block_group->key, sizeof(key));
7811         index = get_block_group_index(block_group);
7812         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7813                                   BTRFS_BLOCK_GROUP_RAID1 |
7814                                   BTRFS_BLOCK_GROUP_RAID10))
7815                 factor = 2;
7816         else
7817                 factor = 1;
7818
7819         /* make sure this block group isn't part of an allocation cluster */
7820         cluster = &root->fs_info->data_alloc_cluster;
7821         spin_lock(&cluster->refill_lock);
7822         btrfs_return_cluster_to_free_space(block_group, cluster);
7823         spin_unlock(&cluster->refill_lock);
7824
7825         /*
7826          * make sure this block group isn't part of a metadata
7827          * allocation cluster
7828          */
7829         cluster = &root->fs_info->meta_alloc_cluster;
7830         spin_lock(&cluster->refill_lock);
7831         btrfs_return_cluster_to_free_space(block_group, cluster);
7832         spin_unlock(&cluster->refill_lock);
7833
7834         path = btrfs_alloc_path();
7835         if (!path) {
7836                 ret = -ENOMEM;
7837                 goto out;
7838         }
7839
7840         inode = lookup_free_space_inode(tree_root, block_group, path);
7841         if (!IS_ERR(inode)) {
7842                 ret = btrfs_orphan_add(trans, inode);
7843                 if (ret) {
7844                         btrfs_add_delayed_iput(inode);
7845                         goto out;
7846                 }
7847                 clear_nlink(inode);
7848                 /* One for the block groups ref */
7849                 spin_lock(&block_group->lock);
7850                 if (block_group->iref) {
7851                         block_group->iref = 0;
7852                         block_group->inode = NULL;
7853                         spin_unlock(&block_group->lock);
7854                         iput(inode);
7855                 } else {
7856                         spin_unlock(&block_group->lock);
7857                 }
7858                 /* One for our lookup ref */
7859                 btrfs_add_delayed_iput(inode);
7860         }
7861
7862         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7863         key.offset = block_group->key.objectid;
7864         key.type = 0;
7865
7866         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7867         if (ret < 0)
7868                 goto out;
7869         if (ret > 0)
7870                 btrfs_release_path(path);
7871         if (ret == 0) {
7872                 ret = btrfs_del_item(trans, tree_root, path);
7873                 if (ret)
7874                         goto out;
7875                 btrfs_release_path(path);
7876         }
7877
7878         spin_lock(&root->fs_info->block_group_cache_lock);
7879         rb_erase(&block_group->cache_node,
7880                  &root->fs_info->block_group_cache_tree);
7881         spin_unlock(&root->fs_info->block_group_cache_lock);
7882
7883         down_write(&block_group->space_info->groups_sem);
7884         /*
7885          * we must use list_del_init so people can check to see if they
7886          * are still on the list after taking the semaphore
7887          */
7888         list_del_init(&block_group->list);
7889         if (list_empty(&block_group->space_info->block_groups[index]))
7890                 clear_avail_alloc_bits(root->fs_info, block_group->flags);
7891         up_write(&block_group->space_info->groups_sem);
7892
7893         if (block_group->cached == BTRFS_CACHE_STARTED)
7894                 wait_block_group_cache_done(block_group);
7895
7896         btrfs_remove_free_space_cache(block_group);
7897
7898         spin_lock(&block_group->space_info->lock);
7899         block_group->space_info->total_bytes -= block_group->key.offset;
7900         block_group->space_info->bytes_readonly -= block_group->key.offset;
7901         block_group->space_info->disk_total -= block_group->key.offset * factor;
7902         spin_unlock(&block_group->space_info->lock);
7903
7904         memcpy(&key, &block_group->key, sizeof(key));
7905
7906         btrfs_clear_space_info_full(root->fs_info);
7907
7908         btrfs_put_block_group(block_group);
7909         btrfs_put_block_group(block_group);
7910
7911         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7912         if (ret > 0)
7913                 ret = -EIO;
7914         if (ret < 0)
7915                 goto out;
7916
7917         ret = btrfs_del_item(trans, root, path);
7918 out:
7919         btrfs_free_path(path);
7920         return ret;
7921 }
7922
7923 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7924 {
7925         struct btrfs_space_info *space_info;
7926         struct btrfs_super_block *disk_super;
7927         u64 features;
7928         u64 flags;
7929         int mixed = 0;
7930         int ret;
7931
7932         disk_super = fs_info->super_copy;
7933         if (!btrfs_super_root(disk_super))
7934                 return 1;
7935
7936         features = btrfs_super_incompat_flags(disk_super);
7937         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7938                 mixed = 1;
7939
7940         flags = BTRFS_BLOCK_GROUP_SYSTEM;
7941         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7942         if (ret)
7943                 goto out;
7944
7945         if (mixed) {
7946                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7947                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7948         } else {
7949                 flags = BTRFS_BLOCK_GROUP_METADATA;
7950                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7951                 if (ret)
7952                         goto out;
7953
7954                 flags = BTRFS_BLOCK_GROUP_DATA;
7955                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7956         }
7957 out:
7958         return ret;
7959 }
7960
7961 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7962 {
7963         return unpin_extent_range(root, start, end);
7964 }
7965
7966 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7967                                u64 num_bytes, u64 *actual_bytes)
7968 {
7969         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7970 }
7971
7972 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7973 {
7974         struct btrfs_fs_info *fs_info = root->fs_info;
7975         struct btrfs_block_group_cache *cache = NULL;
7976         u64 group_trimmed;
7977         u64 start;
7978         u64 end;
7979         u64 trimmed = 0;
7980         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
7981         int ret = 0;
7982
7983         /*
7984          * try to trim all FS space, our block group may start from non-zero.
7985          */
7986         if (range->len == total_bytes)
7987                 cache = btrfs_lookup_first_block_group(fs_info, range->start);
7988         else
7989                 cache = btrfs_lookup_block_group(fs_info, range->start);
7990
7991         while (cache) {
7992                 if (cache->key.objectid >= (range->start + range->len)) {
7993                         btrfs_put_block_group(cache);
7994                         break;
7995                 }
7996
7997                 start = max(range->start, cache->key.objectid);
7998                 end = min(range->start + range->len,
7999                                 cache->key.objectid + cache->key.offset);
8000
8001                 if (end - start >= range->minlen) {
8002                         if (!block_group_cache_done(cache)) {
8003                                 ret = cache_block_group(cache, NULL, root, 0);
8004                                 if (!ret)
8005                                         wait_block_group_cache_done(cache);
8006                         }
8007                         ret = btrfs_trim_block_group(cache,
8008                                                      &group_trimmed,
8009                                                      start,
8010                                                      end,
8011                                                      range->minlen);
8012
8013                         trimmed += group_trimmed;
8014                         if (ret) {
8015                                 btrfs_put_block_group(cache);
8016                                 break;
8017                         }
8018                 }
8019
8020                 cache = next_block_group(fs_info->tree_root, cache);
8021         }
8022
8023         range->len = trimmed;
8024         return ret;
8025 }