fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/kthread.h>
  25 #include <linux/slab.h>
  26 #include <linux/ratelimit.h>
  27 #include "compat.h"
  28 #include "hash.h"
  29 #include "ctree.h"
  30 #include "disk-io.h"
  31 #include "print-tree.h"
  32 #include "transaction.h"
  33 #include "volumes.h"
  34 #include "locking.h"
  35 #include "free-space-cache.h"
  36
  37 /* control flags for do_chunk_alloc's force field
  38  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
  39  * if we really need one.
  40  *
  41  * CHUNK_ALLOC_FORCE means it must try to allocate one
  42  *
  43  * CHUNK_ALLOC_LIMITED means to only try and allocate one
  44  * if we have very few chunks already allocated.  This is
  45  * used as part of the clustering code to help make sure
  46  * we have a good pool of storage to cluster in, without
  47  * filling the FS with empty chunks
  48  *
  49  */
  50 enum {
  51         CHUNK_ALLOC_NO_FORCE = 0,
  52         CHUNK_ALLOC_FORCE = 1,
  53         CHUNK_ALLOC_LIMITED = 2,
  54 };
  55
  56 /*
  57  * Control how reservations are dealt with.
  58  *
  59  * RESERVE_FREE - freeing a reservation.
  60  * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
  61  *   ENOSPC accounting
  62  * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
  63  *   bytes_may_use as the ENOSPC accounting is done elsewhere
  64  */
  65 enum {
  66         RESERVE_FREE = 0,
  67         RESERVE_ALLOC = 1,
  68         RESERVE_ALLOC_NO_ACCOUNT = 2,
  69 };
  70
  71 static int update_block_group(struct btrfs_trans_handle *trans,
  72                               struct btrfs_root *root,
  73                               u64 bytenr, u64 num_bytes, int alloc);
  74 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  75                                 struct btrfs_root *root,
  76                                 u64 bytenr, u64 num_bytes, u64 parent,
  77                                 u64 root_objectid, u64 owner_objectid,
  78                                 u64 owner_offset, int refs_to_drop,
  79                                 struct btrfs_delayed_extent_op *extra_op);
  80 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  81                                     struct extent_buffer *leaf,
  82                                     struct btrfs_extent_item *ei);
  83 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  84                                       struct btrfs_root *root,
  85                                       u64 parent, u64 root_objectid,
  86                                       u64 flags, u64 owner, u64 offset,
  87                                       struct btrfs_key *ins, int ref_mod);
  88 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  89                                      struct btrfs_root *root,
  90                                      u64 parent, u64 root_objectid,
  91                                      u64 flags, struct btrfs_disk_key *key,
  92                                      int level, struct btrfs_key *ins);
  93 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  94                           struct btrfs_root *extent_root, u64 alloc_bytes,
  95                           u64 flags, int force);
  96 static int find_next_key(struct btrfs_path *path, int level,
  97                          struct btrfs_key *key);
  98 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  99                             int dump_block_groups);
 100 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
 101                                        u64 num_bytes, int reserve);
 102
 103 static noinline int
 104 block_group_cache_done(struct btrfs_block_group_cache *cache)
 105 {
 106         smp_mb();
 107         return cache->cached == BTRFS_CACHE_FINISHED;
 108 }
 109
 110 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 111 {
 112         return (cache->flags & bits) == bits;
 113 }
 114
 115 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
 116 {
 117         atomic_inc(&cache->count);
 118 }
 119
 120 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 121 {
 122         if (atomic_dec_and_test(&cache->count)) {
 123                 WARN_ON(cache->pinned > 0);
 124                 WARN_ON(cache->reserved > 0);
 125                 kfree(cache->free_space_ctl);
 126                 kfree(cache);
 127         }
 128 }
 129
 130 /*
 131  * this adds the block group to the fs_info rb tree for the block group
 132  * cache
 133  */
 134 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 135                                 struct btrfs_block_group_cache *block_group)
 136 {
 137         struct rb_node **p;
 138         struct rb_node *parent = NULL;
 139         struct btrfs_block_group_cache *cache;
 140
 141         spin_lock(&info->block_group_cache_lock);
 142         p = &info->block_group_cache_tree.rb_node;
 143
 144         while (*p) {
 145                 parent = *p;
 146                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 147                                  cache_node);
 148                 if (block_group->key.objectid < cache->key.objectid) {
 149                         p = &(*p)->rb_left;
 150                 } else if (block_group->key.objectid > cache->key.objectid) {
 151                         p = &(*p)->rb_right;
 152                 } else {
 153                         spin_unlock(&info->block_group_cache_lock);
 154                         return -EEXIST;
 155                 }
 156         }
 157
 158         rb_link_node(&block_group->cache_node, parent, p);
 159         rb_insert_color(&block_group->cache_node,
 160                         &info->block_group_cache_tree);
 161         spin_unlock(&info->block_group_cache_lock);
 162
 163         return 0;
 164 }
 165
 166 /*
 167  * This will return the block group at or after bytenr if contains is 0, else
 168  * it will return the block group that contains the bytenr
 169  */
 170 static struct btrfs_block_group_cache *
 171 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 172                               int contains)
 173 {
 174         struct btrfs_block_group_cache *cache, *ret = NULL;
 175         struct rb_node *n;
 176         u64 end, start;
 177
 178         spin_lock(&info->block_group_cache_lock);
 179         n = info->block_group_cache_tree.rb_node;
 180
 181         while (n) {
 182                 cache = rb_entry(n, struct btrfs_block_group_cache,
 183                                  cache_node);
 184                 end = cache->key.objectid + cache->key.offset - 1;
 185                 start = cache->key.objectid;
 186
 187                 if (bytenr < start) {
 188                         if (!contains && (!ret || start < ret->key.objectid))
 189                                 ret = cache;
 190                         n = n->rb_left;
 191                 } else if (bytenr > start) {
 192                         if (contains && bytenr <= end) {
 193                                 ret = cache;
 194                                 break;
 195                         }
 196                         n = n->rb_right;
 197                 } else {
 198                         ret = cache;
 199                         break;
 200                 }
 201         }
 202         if (ret)
 203                 btrfs_get_block_group(ret);
 204         spin_unlock(&info->block_group_cache_lock);
 205
 206         return ret;
 207 }
 208
 209 static int add_excluded_extent(struct btrfs_root *root,
 210                                u64 start, u64 num_bytes)
 211 {
 212         u64 end = start + num_bytes - 1;
 213         set_extent_bits(&root->fs_info->freed_extents[0],
 214                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 215         set_extent_bits(&root->fs_info->freed_extents[1],
 216                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 217         return 0;
 218 }
 219
 220 static void free_excluded_extents(struct btrfs_root *root,
 221                                   struct btrfs_block_group_cache *cache)
 222 {
 223         u64 start, end;
 224
 225         start = cache->key.objectid;
 226         end = start + cache->key.offset - 1;
 227
 228         clear_extent_bits(&root->fs_info->freed_extents[0],
 229                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 230         clear_extent_bits(&root->fs_info->freed_extents[1],
 231                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 232 }
 233
 234 static int exclude_super_stripes(struct btrfs_root *root,
 235                                  struct btrfs_block_group_cache *cache)
 236 {
 237         u64 bytenr;
 238         u64 *logical;
 239         int stripe_len;
 240         int i, nr, ret;
 241
 242         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 243                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 244                 cache->bytes_super += stripe_len;
 245                 ret = add_excluded_extent(root, cache->key.objectid,
 246                                           stripe_len);
 247                 BUG_ON(ret);
 248         }
 249
 250         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 251                 bytenr = btrfs_sb_offset(i);
 252                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 253                                        cache->key.objectid, bytenr,
 254                                        0, &logical, &nr, &stripe_len);
 255                 BUG_ON(ret);
 256
 257                 while (nr--) {
 258                         cache->bytes_super += stripe_len;
 259                         ret = add_excluded_extent(root, logical[nr],
 260                                                   stripe_len);
 261                         BUG_ON(ret);
 262                 }
 263
 264                 kfree(logical);
 265         }
 266         return 0;
 267 }
 268
 269 static struct btrfs_caching_control *
 270 get_caching_control(struct btrfs_block_group_cache *cache)
 271 {
 272         struct btrfs_caching_control *ctl;
 273
 274         spin_lock(&cache->lock);
 275         if (cache->cached != BTRFS_CACHE_STARTED) {
 276                 spin_unlock(&cache->lock);
 277                 return NULL;
 278         }
 279
 280         /* We're loading it the fast way, so we don't have a caching_ctl. */
 281         if (!cache->caching_ctl) {
 282                 spin_unlock(&cache->lock);
 283                 return NULL;
 284         }
 285
 286         ctl = cache->caching_ctl;
 287         atomic_inc(&ctl->count);
 288         spin_unlock(&cache->lock);
 289         return ctl;
 290 }
 291
 292 static void put_caching_control(struct btrfs_caching_control *ctl)
 293 {
 294         if (atomic_dec_and_test(&ctl->count))
 295                 kfree(ctl);
 296 }
 297
 298 /*
 299  * this is only called by cache_block_group, since we could have freed extents
 300  * we need to check the pinned_extents for any extents that can't be used yet
 301  * since their free space will be released as soon as the transaction commits.
 302  */
 303 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 304                               struct btrfs_fs_info *info, u64 start, u64 end)
 305 {
 306         u64 extent_start, extent_end, size, total_added = 0;
 307         int ret;
 308
 309         while (start < end) {
 310                 ret = find_first_extent_bit(info->pinned_extents, start,
 311                                             &extent_start, &extent_end,
 312                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 313                 if (ret)
 314                         break;
 315
 316                 if (extent_start <= start) {
 317                         start = extent_end + 1;
 318                 } else if (extent_start > start && extent_start < end) {
 319                         size = extent_start - start;
 320                         total_added += size;
 321                         ret = btrfs_add_free_space(block_group, start,
 322                                                    size);
 323                         BUG_ON(ret);
 324                         start = extent_end + 1;
 325                 } else {
 326                         break;
 327                 }
 328         }
 329
 330         if (start < end) {
 331                 size = end - start;
 332                 total_added += size;
 333                 ret = btrfs_add_free_space(block_group, start, size);
 334                 BUG_ON(ret);
 335         }
 336
 337         return total_added;
 338 }
 339
 340 static noinline void caching_thread(struct btrfs_work *work)
 341 {
 342         struct btrfs_block_group_cache *block_group;
 343         struct btrfs_fs_info *fs_info;
 344         struct btrfs_caching_control *caching_ctl;
 345         struct btrfs_root *extent_root;
 346         struct btrfs_path *path;
 347         struct extent_buffer *leaf;
 348         struct btrfs_key key;
 349         u64 total_found = 0;
 350         u64 last = 0;
 351         u32 nritems;
 352         int ret = 0;
 353
 354         caching_ctl = container_of(work, struct btrfs_caching_control, work);
 355         block_group = caching_ctl->block_group;
 356         fs_info = block_group->fs_info;
 357         extent_root = fs_info->extent_root;
 358
 359         path = btrfs_alloc_path();
 360         if (!path)
 361                 goto out;
 362
 363         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 364
 365         /*
 366          * We don't want to deadlock with somebody trying to allocate a new
 367          * extent for the extent root while also trying to search the extent
 368          * root to add free space.  So we skip locking and search the commit
 369          * root, since its read-only
 370          */
 371         path->skip_locking = 1;
 372         path->search_commit_root = 1;
 373         path->reada = 1;
 374
 375         key.objectid = last;
 376         key.offset = 0;
 377         key.type = BTRFS_EXTENT_ITEM_KEY;
 378 again:
 379         mutex_lock(&caching_ctl->mutex);
 380         /* need to make sure the commit_root doesn't disappear */
 381         down_read(&fs_info->extent_commit_sem);
 382
 383         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 384         if (ret < 0)
 385                 goto err;
 386
 387         leaf = path->nodes[0];
 388         nritems = btrfs_header_nritems(leaf);
 389
 390         while (1) {
 391                 if (btrfs_fs_closing(fs_info) > 1) {
 392                         last = (u64)-1;
 393                         break;
 394                 }
 395
 396                 if (path->slots[0] < nritems) {
 397                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 398                 } else {
 399                         ret = find_next_key(path, 0, &key);
 400                         if (ret)
 401                                 break;
 402
 403                         if (need_resched() ||
 404                             btrfs_next_leaf(extent_root, path)) {
 405                                 caching_ctl->progress = last;
 406                                 btrfs_release_path(path);
 407                                 up_read(&fs_info->extent_commit_sem);
 408                                 mutex_unlock(&caching_ctl->mutex);
 409                                 cond_resched();
 410                                 goto again;
 411                         }
 412                         leaf = path->nodes[0];
 413                         nritems = btrfs_header_nritems(leaf);
 414                         continue;
 415                 }
 416
 417                 if (key.objectid < block_group->key.objectid) {
 418                         path->slots[0]++;
 419                         continue;
 420                 }
 421
 422                 if (key.objectid >= block_group->key.objectid +
 423                     block_group->key.offset)
 424                         break;
 425
 426                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 427                         total_found += add_new_free_space(block_group,
 428                                                           fs_info, last,
 429                                                           key.objectid);
 430                         last = key.objectid + key.offset;
 431
 432                         if (total_found > (1024 * 1024 * 2)) {
 433                                 total_found = 0;
 434                                 wake_up(&caching_ctl->wait);
 435                         }
 436                 }
 437                 path->slots[0]++;
 438         }
 439         ret = 0;
 440
 441         total_found += add_new_free_space(block_group, fs_info, last,
 442                                           block_group->key.objectid +
 443                                           block_group->key.offset);
 444         caching_ctl->progress = (u64)-1;
 445
 446         spin_lock(&block_group->lock);
 447         block_group->caching_ctl = NULL;
 448         block_group->cached = BTRFS_CACHE_FINISHED;
 449         spin_unlock(&block_group->lock);
 450
 451 err:
 452         btrfs_free_path(path);
 453         up_read(&fs_info->extent_commit_sem);
 454
 455         free_excluded_extents(extent_root, block_group);
 456
 457         mutex_unlock(&caching_ctl->mutex);
 458 out:
 459         wake_up(&caching_ctl->wait);
 460
 461         put_caching_control(caching_ctl);
 462         btrfs_put_block_group(block_group);
 463 }
 464
 465 static int cache_block_group(struct btrfs_block_group_cache *cache,
 466                              struct btrfs_trans_handle *trans,
 467                              struct btrfs_root *root,
 468                              int load_cache_only)
 469 {
 470         DEFINE_WAIT(wait);
 471         struct btrfs_fs_info *fs_info = cache->fs_info;
 472         struct btrfs_caching_control *caching_ctl;
 473         int ret = 0;
 474
 475         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 476         BUG_ON(!caching_ctl);
 477
 478         INIT_LIST_HEAD(&caching_ctl->list);
 479         mutex_init(&caching_ctl->mutex);
 480         init_waitqueue_head(&caching_ctl->wait);
 481         caching_ctl->block_group = cache;
 482         caching_ctl->progress = cache->key.objectid;
 483         atomic_set(&caching_ctl->count, 1);
 484         caching_ctl->work.func = caching_thread;
 485
 486         spin_lock(&cache->lock);
 487         /*
 488          * This should be a rare occasion, but this could happen I think in the
 489          * case where one thread starts to load the space cache info, and then
 490          * some other thread starts a transaction commit which tries to do an
 491          * allocation while the other thread is still loading the space cache
 492          * info.  The previous loop should have kept us from choosing this block
 493          * group, but if we've moved to the state where we will wait on caching
 494          * block groups we need to first check if we're doing a fast load here,
 495          * so we can wait for it to finish, otherwise we could end up allocating
 496          * from a block group who's cache gets evicted for one reason or
 497          * another.
 498          */
 499         while (cache->cached == BTRFS_CACHE_FAST) {
 500                 struct btrfs_caching_control *ctl;
 501
 502                 ctl = cache->caching_ctl;
 503                 atomic_inc(&ctl->count);
 504                 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
 505                 spin_unlock(&cache->lock);
 506
 507                 schedule();
 508
 509                 finish_wait(&ctl->wait, &wait);
 510                 put_caching_control(ctl);
 511                 spin_lock(&cache->lock);
 512         }
 513
 514         if (cache->cached != BTRFS_CACHE_NO) {
 515                 spin_unlock(&cache->lock);
 516                 kfree(caching_ctl);
 517                 return 0;
 518         }
 519         WARN_ON(cache->caching_ctl);
 520         cache->caching_ctl = caching_ctl;
 521         cache->cached = BTRFS_CACHE_FAST;
 522         spin_unlock(&cache->lock);
 523
 524         /*
 525          * We can't do the read from on-disk cache during a commit since we need
 526          * to have the normal tree locking.  Also if we are currently trying to
 527          * allocate blocks for the tree root we can't do the fast caching since
 528          * we likely hold important locks.
 529          */
 530         if (trans && (!trans->transaction->in_commit) &&
 531             (root && root != root->fs_info->tree_root) &&
 532             btrfs_test_opt(root, SPACE_CACHE)) {
 533                 ret = load_free_space_cache(fs_info, cache);
 534
 535                 spin_lock(&cache->lock);
 536                 if (ret == 1) {
 537                         cache->caching_ctl = NULL;
 538                         cache->cached = BTRFS_CACHE_FINISHED;
 539                         cache->last_byte_to_unpin = (u64)-1;
 540                 } else {
 541                         if (load_cache_only) {
 542                                 cache->caching_ctl = NULL;
 543                                 cache->cached = BTRFS_CACHE_NO;
 544                         } else {
 545                                 cache->cached = BTRFS_CACHE_STARTED;
 546                         }
 547                 }
 548                 spin_unlock(&cache->lock);
 549                 wake_up(&caching_ctl->wait);
 550                 if (ret == 1) {
 551                         put_caching_control(caching_ctl);
 552                         free_excluded_extents(fs_info->extent_root, cache);
 553                         return 0;
 554                 }
 555         } else {
 556                 /*
 557                  * We are not going to do the fast caching, set cached to the
 558                  * appropriate value and wakeup any waiters.
 559                  */
 560                 spin_lock(&cache->lock);
 561                 if (load_cache_only) {
 562                         cache->caching_ctl = NULL;
 563                         cache->cached = BTRFS_CACHE_NO;
 564                 } else {
 565                         cache->cached = BTRFS_CACHE_STARTED;
 566                 }
 567                 spin_unlock(&cache->lock);
 568                 wake_up(&caching_ctl->wait);
 569         }
 570
 571         if (load_cache_only) {
 572                 put_caching_control(caching_ctl);
 573                 return 0;
 574         }
 575
 576         down_write(&fs_info->extent_commit_sem);
 577         atomic_inc(&caching_ctl->count);
 578         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 579         up_write(&fs_info->extent_commit_sem);
 580
 581         btrfs_get_block_group(cache);
 582
 583         btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
 584
 585         return ret;
 586 }
 587
 588 /*
 589  * return the block group that starts at or after bytenr
 590  */
 591 static struct btrfs_block_group_cache *
 592 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 593 {
 594         struct btrfs_block_group_cache *cache;
 595
 596         cache = block_group_cache_tree_search(info, bytenr, 0);
 597
 598         return cache;
 599 }
 600
 601 /*
 602  * return the block group that contains the given bytenr
 603  */
 604 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 605                                                  struct btrfs_fs_info *info,
 606                                                  u64 bytenr)
 607 {
 608         struct btrfs_block_group_cache *cache;
 609
 610         cache = block_group_cache_tree_search(info, bytenr, 1);
 611
 612         return cache;
 613 }
 614
 615 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 616                                                   u64 flags)
 617 {
 618         struct list_head *head = &info->space_info;
 619         struct btrfs_space_info *found;
 620
 621         flags &= BTRFS_BLOCK_GROUP_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);
 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);
1011
1012         ret = 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         BUG_ON(ret);
1481
1482         leaf = path->nodes[0];
1483         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1484 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1485         if (item_size < sizeof(*ei)) {
1486                 if (!insert) {
1487                         err = -ENOENT;
1488                         goto out;
1489                 }
1490                 ret = convert_extent_item_v0(trans, root, path, owner,
1491                                              extra_size);
1492                 if (ret < 0) {
1493                         err = ret;
1494                         goto out;
1495                 }
1496                 leaf = path->nodes[0];
1497                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1498         }
1499 #endif
1500         BUG_ON(item_size < sizeof(*ei));
1501
1502         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1503         flags = btrfs_extent_flags(leaf, ei);
1504
1505         ptr = (unsigned long)(ei + 1);
1506         end = (unsigned long)ei + item_size;
1507
1508         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1509                 ptr += sizeof(struct btrfs_tree_block_info);
1510                 BUG_ON(ptr > end);
1511         } else {
1512                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1513         }
1514
1515         err = -ENOENT;
1516         while (1) {
1517                 if (ptr >= end) {
1518                         WARN_ON(ptr > end);
1519                         break;
1520                 }
1521                 iref = (struct btrfs_extent_inline_ref *)ptr;
1522                 type = btrfs_extent_inline_ref_type(leaf, iref);
1523                 if (want < type)
1524                         break;
1525                 if (want > type) {
1526                         ptr += btrfs_extent_inline_ref_size(type);
1527                         continue;
1528                 }
1529
1530                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1531                         struct btrfs_extent_data_ref *dref;
1532                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1533                         if (match_extent_data_ref(leaf, dref, root_objectid,
1534                                                   owner, offset)) {
1535                                 err = 0;
1536                                 break;
1537                         }
1538                         if (hash_extent_data_ref_item(leaf, dref) <
1539                             hash_extent_data_ref(root_objectid, owner, offset))
1540                                 break;
1541                 } else {
1542                         u64 ref_offset;
1543                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1544                         if (parent > 0) {
1545                                 if (parent == ref_offset) {
1546                                         err = 0;
1547                                         break;
1548                                 }
1549                                 if (ref_offset < parent)
1550                                         break;
1551                         } else {
1552                                 if (root_objectid == ref_offset) {
1553                                         err = 0;
1554                                         break;
1555                                 }
1556                                 if (ref_offset < root_objectid)
1557                                         break;
1558                         }
1559                 }
1560                 ptr += btrfs_extent_inline_ref_size(type);
1561         }
1562         if (err == -ENOENT && insert) {
1563                 if (item_size + extra_size >=
1564                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1565                         err = -EAGAIN;
1566                         goto out;
1567                 }
1568                 /*
1569                  * To add new inline back ref, we have to make sure
1570                  * there is no corresponding back ref item.
1571                  * For simplicity, we just do not add new inline back
1572                  * ref if there is any kind of item for this block
1573                  */
1574                 if (find_next_key(path, 0, &key) == 0 &&
1575                     key.objectid == bytenr &&
1576                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1577                         err = -EAGAIN;
1578                         goto out;
1579                 }
1580         }
1581         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1582 out:
1583         if (insert) {
1584                 path->keep_locks = 0;
1585                 btrfs_unlock_up_safe(path, 1);
1586         }
1587         return err;
1588 }
1589
1590 /*
1591  * helper to add new inline back ref
1592  */
1593 static noinline_for_stack
1594 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1595                                 struct btrfs_root *root,
1596                                 struct btrfs_path *path,
1597                                 struct btrfs_extent_inline_ref *iref,
1598                                 u64 parent, u64 root_objectid,
1599                                 u64 owner, u64 offset, int refs_to_add,
1600                                 struct btrfs_delayed_extent_op *extent_op)
1601 {
1602         struct extent_buffer *leaf;
1603         struct btrfs_extent_item *ei;
1604         unsigned long ptr;
1605         unsigned long end;
1606         unsigned long item_offset;
1607         u64 refs;
1608         int size;
1609         int type;
1610         int ret;
1611
1612         leaf = path->nodes[0];
1613         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1614         item_offset = (unsigned long)iref - (unsigned long)ei;
1615
1616         type = extent_ref_type(parent, owner);
1617         size = btrfs_extent_inline_ref_size(type);
1618
1619         ret = btrfs_extend_item(trans, root, path, size);
1620
1621         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1622         refs = btrfs_extent_refs(leaf, ei);
1623         refs += refs_to_add;
1624         btrfs_set_extent_refs(leaf, ei, refs);
1625         if (extent_op)
1626                 __run_delayed_extent_op(extent_op, leaf, ei);
1627
1628         ptr = (unsigned long)ei + item_offset;
1629         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1630         if (ptr < end - size)
1631                 memmove_extent_buffer(leaf, ptr + size, ptr,
1632                                       end - size - ptr);
1633
1634         iref = (struct btrfs_extent_inline_ref *)ptr;
1635         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1636         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1637                 struct btrfs_extent_data_ref *dref;
1638                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1639                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1640                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1641                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1642                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1643         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1644                 struct btrfs_shared_data_ref *sref;
1645                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1646                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1647                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1648         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1649                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1650         } else {
1651                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1652         }
1653         btrfs_mark_buffer_dirty(leaf);
1654         return 0;
1655 }
1656
1657 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1658                                  struct btrfs_root *root,
1659                                  struct btrfs_path *path,
1660                                  struct btrfs_extent_inline_ref **ref_ret,
1661                                  u64 bytenr, u64 num_bytes, u64 parent,
1662                                  u64 root_objectid, u64 owner, u64 offset)
1663 {
1664         int ret;
1665
1666         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1667                                            bytenr, num_bytes, parent,
1668                                            root_objectid, owner, offset, 0);
1669         if (ret != -ENOENT)
1670                 return ret;
1671
1672         btrfs_release_path(path);
1673         *ref_ret = NULL;
1674
1675         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1676                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1677                                             root_objectid);
1678         } else {
1679                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1680                                              root_objectid, owner, offset);
1681         }
1682         return ret;
1683 }
1684
1685 /*
1686  * helper to update/remove inline back ref
1687  */
1688 static noinline_for_stack
1689 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1690                                  struct btrfs_root *root,
1691                                  struct btrfs_path *path,
1692                                  struct btrfs_extent_inline_ref *iref,
1693                                  int refs_to_mod,
1694                                  struct btrfs_delayed_extent_op *extent_op)
1695 {
1696         struct extent_buffer *leaf;
1697         struct btrfs_extent_item *ei;
1698         struct btrfs_extent_data_ref *dref = NULL;
1699         struct btrfs_shared_data_ref *sref = NULL;
1700         unsigned long ptr;
1701         unsigned long end;
1702         u32 item_size;
1703         int size;
1704         int type;
1705         int ret;
1706         u64 refs;
1707
1708         leaf = path->nodes[0];
1709         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1710         refs = btrfs_extent_refs(leaf, ei);
1711         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1712         refs += refs_to_mod;
1713         btrfs_set_extent_refs(leaf, ei, refs);
1714         if (extent_op)
1715                 __run_delayed_extent_op(extent_op, leaf, ei);
1716
1717         type = btrfs_extent_inline_ref_type(leaf, iref);
1718
1719         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1720                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1721                 refs = btrfs_extent_data_ref_count(leaf, dref);
1722         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1723                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1724                 refs = btrfs_shared_data_ref_count(leaf, sref);
1725         } else {
1726                 refs = 1;
1727                 BUG_ON(refs_to_mod != -1);
1728         }
1729
1730         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1731         refs += refs_to_mod;
1732
1733         if (refs > 0) {
1734                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1735                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1736                 else
1737                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1738         } else {
1739                 size =  btrfs_extent_inline_ref_size(type);
1740                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1741                 ptr = (unsigned long)iref;
1742                 end = (unsigned long)ei + item_size;
1743                 if (ptr + size < end)
1744                         memmove_extent_buffer(leaf, ptr, ptr + size,
1745                                               end - ptr - size);
1746                 item_size -= size;
1747                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1748         }
1749         btrfs_mark_buffer_dirty(leaf);
1750         return 0;
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                 ret = update_inline_extent_backref(trans, root, path, iref,
1771                                                    refs_to_add, extent_op);
1772         } else if (ret == -ENOENT) {
1773                 ret = setup_inline_extent_backref(trans, root, path, iref,
1774                                                   parent, root_objectid,
1775                                                   owner, offset, refs_to_add,
1776                                                   extent_op);
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;
1807
1808         BUG_ON(!is_data && refs_to_drop != 1);
1809         if (iref) {
1810                 ret = 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         if (!ret) {
1838                 struct btrfs_bio_stripe *stripe = bbio->stripes;
1839                 int i;
1840
1841
1842                 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1843                         if (!stripe->dev->can_discard)
1844                                 continue;
1845
1846                         ret = btrfs_issue_discard(stripe->dev->bdev,
1847                                                   stripe->physical,
1848                                                   stripe->length);
1849                         if (!ret)
1850                                 discarded_bytes += stripe->length;
1851                         else if (ret != -EOPNOTSUPP)
1852                                 break;
1853
1854                         /*
1855                          * Just in case we get back EOPNOTSUPP for some reason,
1856                          * just ignore the return value so we don't screw up
1857                          * people calling discard_extent.
1858                          */
1859                         ret = 0;
1860                 }
1861                 kfree(bbio);
1862         }
1863
1864         if (actual_bytes)
1865                 *actual_bytes = discarded_bytes;
1866
1867
1868         return ret;
1869 }
1870
1871 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1872                          struct btrfs_root *root,
1873                          u64 bytenr, u64 num_bytes, u64 parent,
1874                          u64 root_objectid, u64 owner, u64 offset, int for_cow)
1875 {
1876         int ret;
1877         struct btrfs_fs_info *fs_info = root->fs_info;
1878
1879         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1880                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1881
1882         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1883                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1884                                         num_bytes,
1885                                         parent, root_objectid, (int)owner,
1886                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1887         } else {
1888                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1889                                         num_bytes,
1890                                         parent, root_objectid, owner, offset,
1891                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1892         }
1893         return ret;
1894 }
1895
1896 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1897                                   struct btrfs_root *root,
1898                                   u64 bytenr, u64 num_bytes,
1899                                   u64 parent, u64 root_objectid,
1900                                   u64 owner, u64 offset, int refs_to_add,
1901                                   struct btrfs_delayed_extent_op *extent_op)
1902 {
1903         struct btrfs_path *path;
1904         struct extent_buffer *leaf;
1905         struct btrfs_extent_item *item;
1906         u64 refs;
1907         int ret;
1908         int err = 0;
1909
1910         path = btrfs_alloc_path();
1911         if (!path)
1912                 return -ENOMEM;
1913
1914         path->reada = 1;
1915         path->leave_spinning = 1;
1916         /* this will setup the path even if it fails to insert the back ref */
1917         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1918                                            path, bytenr, num_bytes, parent,
1919                                            root_objectid, owner, offset,
1920                                            refs_to_add, extent_op);
1921         if (ret == 0)
1922                 goto out;
1923
1924         if (ret != -EAGAIN) {
1925                 err = ret;
1926                 goto out;
1927         }
1928
1929         leaf = path->nodes[0];
1930         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1931         refs = btrfs_extent_refs(leaf, item);
1932         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1933         if (extent_op)
1934                 __run_delayed_extent_op(extent_op, leaf, item);
1935
1936         btrfs_mark_buffer_dirty(leaf);
1937         btrfs_release_path(path);
1938
1939         path->reada = 1;
1940         path->leave_spinning = 1;
1941
1942         /* now insert the actual backref */
1943         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1944                                     path, bytenr, parent, root_objectid,
1945                                     owner, offset, refs_to_add);
1946         BUG_ON(ret);
1947 out:
1948         btrfs_free_path(path);
1949         return err;
1950 }
1951
1952 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1953                                 struct btrfs_root *root,
1954                                 struct btrfs_delayed_ref_node *node,
1955                                 struct btrfs_delayed_extent_op *extent_op,
1956                                 int insert_reserved)
1957 {
1958         int ret = 0;
1959         struct btrfs_delayed_data_ref *ref;
1960         struct btrfs_key ins;
1961         u64 parent = 0;
1962         u64 ref_root = 0;
1963         u64 flags = 0;
1964
1965         ins.objectid = node->bytenr;
1966         ins.offset = node->num_bytes;
1967         ins.type = BTRFS_EXTENT_ITEM_KEY;
1968
1969         ref = btrfs_delayed_node_to_data_ref(node);
1970         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1971                 parent = ref->parent;
1972         else
1973                 ref_root = ref->root;
1974
1975         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1976                 if (extent_op) {
1977                         BUG_ON(extent_op->update_key);
1978                         flags |= extent_op->flags_to_set;
1979                 }
1980                 ret = alloc_reserved_file_extent(trans, root,
1981                                                  parent, ref_root, flags,
1982                                                  ref->objectid, ref->offset,
1983                                                  &ins, node->ref_mod);
1984         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1985                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1986                                              node->num_bytes, parent,
1987                                              ref_root, ref->objectid,
1988                                              ref->offset, node->ref_mod,
1989                                              extent_op);
1990         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1991                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1992                                           node->num_bytes, parent,
1993                                           ref_root, ref->objectid,
1994                                           ref->offset, node->ref_mod,
1995                                           extent_op);
1996         } else {
1997                 BUG();
1998         }
1999         return ret;
2000 }
2001
2002 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2003                                     struct extent_buffer *leaf,
2004                                     struct btrfs_extent_item *ei)
2005 {
2006         u64 flags = btrfs_extent_flags(leaf, ei);
2007         if (extent_op->update_flags) {
2008                 flags |= extent_op->flags_to_set;
2009                 btrfs_set_extent_flags(leaf, ei, flags);
2010         }
2011
2012         if (extent_op->update_key) {
2013                 struct btrfs_tree_block_info *bi;
2014                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2015                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2016                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2017         }
2018 }
2019
2020 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2021                                  struct btrfs_root *root,
2022                                  struct btrfs_delayed_ref_node *node,
2023                                  struct btrfs_delayed_extent_op *extent_op)
2024 {
2025         struct btrfs_key key;
2026         struct btrfs_path *path;
2027         struct btrfs_extent_item *ei;
2028         struct extent_buffer *leaf;
2029         u32 item_size;
2030         int ret;
2031         int err = 0;
2032
2033         path = btrfs_alloc_path();
2034         if (!path)
2035                 return -ENOMEM;
2036
2037         key.objectid = node->bytenr;
2038         key.type = BTRFS_EXTENT_ITEM_KEY;
2039         key.offset = node->num_bytes;
2040
2041         path->reada = 1;
2042         path->leave_spinning = 1;
2043         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2044                                 path, 0, 1);
2045         if (ret < 0) {
2046                 err = ret;
2047                 goto out;
2048         }
2049         if (ret > 0) {
2050                 err = -EIO;
2051                 goto out;
2052         }
2053
2054         leaf = path->nodes[0];
2055         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2056 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2057         if (item_size < sizeof(*ei)) {
2058                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2059                                              path, (u64)-1, 0);
2060                 if (ret < 0) {
2061                         err = ret;
2062                         goto out;
2063                 }
2064                 leaf = path->nodes[0];
2065                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2066         }
2067 #endif
2068         BUG_ON(item_size < sizeof(*ei));
2069         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2070         __run_delayed_extent_op(extent_op, leaf, ei);
2071
2072         btrfs_mark_buffer_dirty(leaf);
2073 out:
2074         btrfs_free_path(path);
2075         return err;
2076 }
2077
2078 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2079                                 struct btrfs_root *root,
2080                                 struct btrfs_delayed_ref_node *node,
2081                                 struct btrfs_delayed_extent_op *extent_op,
2082                                 int insert_reserved)
2083 {
2084         int ret = 0;
2085         struct btrfs_delayed_tree_ref *ref;
2086         struct btrfs_key ins;
2087         u64 parent = 0;
2088         u64 ref_root = 0;
2089
2090         ins.objectid = node->bytenr;
2091         ins.offset = node->num_bytes;
2092         ins.type = BTRFS_EXTENT_ITEM_KEY;
2093
2094         ref = btrfs_delayed_node_to_tree_ref(node);
2095         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2096                 parent = ref->parent;
2097         else
2098                 ref_root = ref->root;
2099
2100         BUG_ON(node->ref_mod != 1);
2101         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2102                 BUG_ON(!extent_op || !extent_op->update_flags ||
2103                        !extent_op->update_key);
2104                 ret = alloc_reserved_tree_block(trans, root,
2105                                                 parent, ref_root,
2106                                                 extent_op->flags_to_set,
2107                                                 &extent_op->key,
2108                                                 ref->level, &ins);
2109         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2110                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2111                                              node->num_bytes, parent, ref_root,
2112                                              ref->level, 0, 1, extent_op);
2113         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2114                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2115                                           node->num_bytes, parent, ref_root,
2116                                           ref->level, 0, 1, extent_op);
2117         } else {
2118                 BUG();
2119         }
2120         return ret;
2121 }
2122
2123 /* helper function to actually process a single delayed ref entry */
2124 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2125                                struct btrfs_root *root,
2126                                struct btrfs_delayed_ref_node *node,
2127                                struct btrfs_delayed_extent_op *extent_op,
2128                                int insert_reserved)
2129 {
2130         int ret;
2131         if (btrfs_delayed_ref_is_head(node)) {
2132                 struct btrfs_delayed_ref_head *head;
2133                 /*
2134                  * we've hit the end of the chain and we were supposed
2135                  * to insert this extent into the tree.  But, it got
2136                  * deleted before we ever needed to insert it, so all
2137                  * we have to do is clean up the accounting
2138                  */
2139                 BUG_ON(extent_op);
2140                 head = btrfs_delayed_node_to_head(node);
2141                 if (insert_reserved) {
2142                         btrfs_pin_extent(root, node->bytenr,
2143                                          node->num_bytes, 1);
2144                         if (head->is_data) {
2145                                 ret = btrfs_del_csums(trans, root,
2146                                                       node->bytenr,
2147                                                       node->num_bytes);
2148                                 BUG_ON(ret);
2149                         }
2150                 }
2151                 mutex_unlock(&head->mutex);
2152                 return 0;
2153         }
2154
2155         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2156             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2157                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2158                                            insert_reserved);
2159         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2160                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2161                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2162                                            insert_reserved);
2163         else
2164                 BUG();
2165         return ret;
2166 }
2167
2168 static noinline struct btrfs_delayed_ref_node *
2169 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2170 {
2171         struct rb_node *node;
2172         struct btrfs_delayed_ref_node *ref;
2173         int action = BTRFS_ADD_DELAYED_REF;
2174 again:
2175         /*
2176          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2177          * this prevents ref count from going down to zero when
2178          * there still are pending delayed ref.
2179          */
2180         node = rb_prev(&head->node.rb_node);
2181         while (1) {
2182                 if (!node)
2183                         break;
2184                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2185                                 rb_node);
2186                 if (ref->bytenr != head->node.bytenr)
2187                         break;
2188                 if (ref->action == action)
2189                         return ref;
2190                 node = rb_prev(node);
2191         }
2192         if (action == BTRFS_ADD_DELAYED_REF) {
2193                 action = BTRFS_DROP_DELAYED_REF;
2194                 goto again;
2195         }
2196         return NULL;
2197 }
2198
2199 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2200                                        struct btrfs_root *root,
2201                                        struct list_head *cluster)
2202 {
2203         struct btrfs_delayed_ref_root *delayed_refs;
2204         struct btrfs_delayed_ref_node *ref;
2205         struct btrfs_delayed_ref_head *locked_ref = NULL;
2206         struct btrfs_delayed_extent_op *extent_op;
2207         int ret;
2208         int count = 0;
2209         int must_insert_reserved = 0;
2210
2211         delayed_refs = &trans->transaction->delayed_refs;
2212         while (1) {
2213                 if (!locked_ref) {
2214                         /* pick a new head ref from the cluster list */
2215                         if (list_empty(cluster))
2216                                 break;
2217
2218                         locked_ref = list_entry(cluster->next,
2219                                      struct btrfs_delayed_ref_head, cluster);
2220
2221                         /* grab the lock that says we are going to process
2222                          * all the refs for this head */
2223                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2224
2225                         /*
2226                          * we may have dropped the spin lock to get the head
2227                          * mutex lock, and that might have given someone else
2228                          * time to free the head.  If that's true, it has been
2229                          * removed from our list and we can move on.
2230                          */
2231                         if (ret == -EAGAIN) {
2232                                 locked_ref = NULL;
2233                                 count++;
2234                                 continue;
2235                         }
2236                 }
2237
2238                 /*
2239                  * locked_ref is the head node, so we have to go one
2240                  * node back for any delayed ref updates
2241                  */
2242                 ref = select_delayed_ref(locked_ref);
2243
2244                 if (ref && ref->seq &&
2245                     btrfs_check_delayed_seq(delayed_refs, ref->seq)) {
2246                         /*
2247                          * there are still refs with lower seq numbers in the
2248                          * process of being added. Don't run this ref yet.
2249                          */
2250                         list_del_init(&locked_ref->cluster);
2251                         mutex_unlock(&locked_ref->mutex);
2252                         locked_ref = NULL;
2253                         delayed_refs->num_heads_ready++;
2254                         spin_unlock(&delayed_refs->lock);
2255                         cond_resched();
2256                         spin_lock(&delayed_refs->lock);
2257                         continue;
2258                 }
2259
2260                 /*
2261                  * record the must insert reserved flag before we
2262                  * drop the spin lock.
2263                  */
2264                 must_insert_reserved = locked_ref->must_insert_reserved;
2265                 locked_ref->must_insert_reserved = 0;
2266
2267                 extent_op = locked_ref->extent_op;
2268                 locked_ref->extent_op = NULL;
2269
2270                 if (!ref) {
2271                         /* All delayed refs have been processed, Go ahead
2272                          * and send the head node to run_one_delayed_ref,
2273                          * so that any accounting fixes can happen
2274                          */
2275                         ref = &locked_ref->node;
2276
2277                         if (extent_op && must_insert_reserved) {
2278                                 kfree(extent_op);
2279                                 extent_op = NULL;
2280                         }
2281
2282                         if (extent_op) {
2283                                 spin_unlock(&delayed_refs->lock);
2284
2285                                 ret = run_delayed_extent_op(trans, root,
2286                                                             ref, extent_op);
2287                                 BUG_ON(ret);
2288                                 kfree(extent_op);
2289
2290                                 goto next;
2291                         }
2292
2293                         list_del_init(&locked_ref->cluster);
2294                         locked_ref = NULL;
2295                 }
2296
2297                 ref->in_tree = 0;
2298                 rb_erase(&ref->rb_node, &delayed_refs->root);
2299                 delayed_refs->num_entries--;
2300                 /*
2301                  * we modified num_entries, but as we're currently running
2302                  * delayed refs, skip
2303                  *     wake_up(&delayed_refs->seq_wait);
2304                  * here.
2305                  */
2306                 spin_unlock(&delayed_refs->lock);
2307
2308                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2309                                           must_insert_reserved);
2310                 BUG_ON(ret);
2311
2312                 btrfs_put_delayed_ref(ref);
2313                 kfree(extent_op);
2314                 count++;
2315 next:
2316                 do_chunk_alloc(trans, root->fs_info->extent_root,
2317                                2 * 1024 * 1024,
2318                                btrfs_get_alloc_profile(root, 0),
2319                                CHUNK_ALLOC_NO_FORCE);
2320                 cond_resched();
2321                 spin_lock(&delayed_refs->lock);
2322         }
2323         return count;
2324 }
2325
2326
2327 static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs,
2328                         unsigned long num_refs)
2329 {
2330         struct list_head *first_seq = delayed_refs->seq_head.next;
2331
2332         spin_unlock(&delayed_refs->lock);
2333         pr_debug("waiting for more refs (num %ld, first %p)\n",
2334                  num_refs, first_seq);
2335         wait_event(delayed_refs->seq_wait,
2336                    num_refs != delayed_refs->num_entries ||
2337                    delayed_refs->seq_head.next != first_seq);
2338         pr_debug("done waiting for more refs (num %ld, first %p)\n",
2339                  delayed_refs->num_entries, delayed_refs->seq_head.next);
2340         spin_lock(&delayed_refs->lock);
2341 }
2342
2343 /*
2344  * this starts processing the delayed reference count updates and
2345  * extent insertions we have queued up so far.  count can be
2346  * 0, which means to process everything in the tree at the start
2347  * of the run (but not newly added entries), or it can be some target
2348  * number you'd like to process.
2349  */
2350 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2351                            struct btrfs_root *root, unsigned long count)
2352 {
2353         struct rb_node *node;
2354         struct btrfs_delayed_ref_root *delayed_refs;
2355         struct btrfs_delayed_ref_node *ref;
2356         struct list_head cluster;
2357         int ret;
2358         u64 delayed_start;
2359         int run_all = count == (unsigned long)-1;
2360         int run_most = 0;
2361         unsigned long num_refs = 0;
2362         int consider_waiting;
2363
2364         if (root == root->fs_info->extent_root)
2365                 root = root->fs_info->tree_root;
2366
2367         do_chunk_alloc(trans, root->fs_info->extent_root,
2368                        2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2369                        CHUNK_ALLOC_NO_FORCE);
2370
2371         delayed_refs = &trans->transaction->delayed_refs;
2372         INIT_LIST_HEAD(&cluster);
2373 again:
2374         consider_waiting = 0;
2375         spin_lock(&delayed_refs->lock);
2376         if (count == 0) {
2377                 count = delayed_refs->num_entries * 2;
2378                 run_most = 1;
2379         }
2380         while (1) {
2381                 if (!(run_all || run_most) &&
2382                     delayed_refs->num_heads_ready < 64)
2383                         break;
2384
2385                 /*
2386                  * go find something we can process in the rbtree.  We start at
2387                  * the beginning of the tree, and then build a cluster
2388                  * of refs to process starting at the first one we are able to
2389                  * lock
2390                  */
2391                 delayed_start = delayed_refs->run_delayed_start;
2392                 ret = btrfs_find_ref_cluster(trans, &cluster,
2393                                              delayed_refs->run_delayed_start);
2394                 if (ret)
2395                         break;
2396
2397                 if (delayed_start >= delayed_refs->run_delayed_start) {
2398                         if (consider_waiting == 0) {
2399                                 /*
2400                                  * btrfs_find_ref_cluster looped. let's do one
2401                                  * more cycle. if we don't run any delayed ref
2402                                  * during that cycle (because we can't because
2403                                  * all of them are blocked) and if the number of
2404                                  * refs doesn't change, we avoid busy waiting.
2405                                  */
2406                                 consider_waiting = 1;
2407                                 num_refs = delayed_refs->num_entries;
2408                         } else {
2409                                 wait_for_more_refs(delayed_refs, num_refs);
2410                                 /*
2411                                  * after waiting, things have changed. we
2412                                  * dropped the lock and someone else might have
2413                                  * run some refs, built new clusters and so on.
2414                                  * therefore, we restart staleness detection.
2415                                  */
2416                                 consider_waiting = 0;
2417                         }
2418                 }
2419
2420                 ret = run_clustered_refs(trans, root, &cluster);
2421                 BUG_ON(ret < 0);
2422
2423                 count -= min_t(unsigned long, ret, count);
2424
2425                 if (count == 0)
2426                         break;
2427
2428                 if (ret || delayed_refs->run_delayed_start == 0) {
2429                         /* refs were run, let's reset staleness detection */
2430                         consider_waiting = 0;
2431                 }
2432         }
2433
2434         if (run_all) {
2435                 node = rb_first(&delayed_refs->root);
2436                 if (!node)
2437                         goto out;
2438                 count = (unsigned long)-1;
2439
2440                 while (node) {
2441                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2442                                        rb_node);
2443                         if (btrfs_delayed_ref_is_head(ref)) {
2444                                 struct btrfs_delayed_ref_head *head;
2445
2446                                 head = btrfs_delayed_node_to_head(ref);
2447                                 atomic_inc(&ref->refs);
2448
2449                                 spin_unlock(&delayed_refs->lock);
2450                                 /*
2451                                  * Mutex was contended, block until it's
2452                                  * released and try again
2453                                  */
2454                                 mutex_lock(&head->mutex);
2455                                 mutex_unlock(&head->mutex);
2456
2457                                 btrfs_put_delayed_ref(ref);
2458                                 cond_resched();
2459                                 goto again;
2460                         }
2461                         node = rb_next(node);
2462                 }
2463                 spin_unlock(&delayed_refs->lock);
2464                 schedule_timeout(1);
2465                 goto again;
2466         }
2467 out:
2468         spin_unlock(&delayed_refs->lock);
2469         return 0;
2470 }
2471
2472 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2473                                 struct btrfs_root *root,
2474                                 u64 bytenr, u64 num_bytes, u64 flags,
2475                                 int is_data)
2476 {
2477         struct btrfs_delayed_extent_op *extent_op;
2478         int ret;
2479
2480         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2481         if (!extent_op)
2482                 return -ENOMEM;
2483
2484         extent_op->flags_to_set = flags;
2485         extent_op->update_flags = 1;
2486         extent_op->update_key = 0;
2487         extent_op->is_data = is_data ? 1 : 0;
2488
2489         ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2490                                           num_bytes, extent_op);
2491         if (ret)
2492                 kfree(extent_op);
2493         return ret;
2494 }
2495
2496 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2497                                       struct btrfs_root *root,
2498                                       struct btrfs_path *path,
2499                                       u64 objectid, u64 offset, u64 bytenr)
2500 {
2501         struct btrfs_delayed_ref_head *head;
2502         struct btrfs_delayed_ref_node *ref;
2503         struct btrfs_delayed_data_ref *data_ref;
2504         struct btrfs_delayed_ref_root *delayed_refs;
2505         struct rb_node *node;
2506         int ret = 0;
2507
2508         ret = -ENOENT;
2509         delayed_refs = &trans->transaction->delayed_refs;
2510         spin_lock(&delayed_refs->lock);
2511         head = btrfs_find_delayed_ref_head(trans, bytenr);
2512         if (!head)
2513                 goto out;
2514
2515         if (!mutex_trylock(&head->mutex)) {
2516                 atomic_inc(&head->node.refs);
2517                 spin_unlock(&delayed_refs->lock);
2518
2519                 btrfs_release_path(path);
2520
2521                 /*
2522                  * Mutex was contended, block until it's released and let
2523                  * caller try again
2524                  */
2525                 mutex_lock(&head->mutex);
2526                 mutex_unlock(&head->mutex);
2527                 btrfs_put_delayed_ref(&head->node);
2528                 return -EAGAIN;
2529         }
2530
2531         node = rb_prev(&head->node.rb_node);
2532         if (!node)
2533                 goto out_unlock;
2534
2535         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2536
2537         if (ref->bytenr != bytenr)
2538                 goto out_unlock;
2539
2540         ret = 1;
2541         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2542                 goto out_unlock;
2543
2544         data_ref = btrfs_delayed_node_to_data_ref(ref);
2545
2546         node = rb_prev(node);
2547         if (node) {
2548                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2549                 if (ref->bytenr == bytenr)
2550                         goto out_unlock;
2551         }
2552
2553         if (data_ref->root != root->root_key.objectid ||
2554             data_ref->objectid != objectid || data_ref->offset != offset)
2555                 goto out_unlock;
2556
2557         ret = 0;
2558 out_unlock:
2559         mutex_unlock(&head->mutex);
2560 out:
2561         spin_unlock(&delayed_refs->lock);
2562         return ret;
2563 }
2564
2565 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2566                                         struct btrfs_root *root,
2567                                         struct btrfs_path *path,
2568                                         u64 objectid, u64 offset, u64 bytenr)
2569 {
2570         struct btrfs_root *extent_root = root->fs_info->extent_root;
2571         struct extent_buffer *leaf;
2572         struct btrfs_extent_data_ref *ref;
2573         struct btrfs_extent_inline_ref *iref;
2574         struct btrfs_extent_item *ei;
2575         struct btrfs_key key;
2576         u32 item_size;
2577         int ret;
2578
2579         key.objectid = bytenr;
2580         key.offset = (u64)-1;
2581         key.type = BTRFS_EXTENT_ITEM_KEY;
2582
2583         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2584         if (ret < 0)
2585                 goto out;
2586         BUG_ON(ret == 0);
2587
2588         ret = -ENOENT;
2589         if (path->slots[0] == 0)
2590                 goto out;
2591
2592         path->slots[0]--;
2593         leaf = path->nodes[0];
2594         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2595
2596         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2597                 goto out;
2598
2599         ret = 1;
2600         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2601 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2602         if (item_size < sizeof(*ei)) {
2603                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2604                 goto out;
2605         }
2606 #endif
2607         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2608
2609         if (item_size != sizeof(*ei) +
2610             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2611                 goto out;
2612
2613         if (btrfs_extent_generation(leaf, ei) <=
2614             btrfs_root_last_snapshot(&root->root_item))
2615                 goto out;
2616
2617         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2618         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2619             BTRFS_EXTENT_DATA_REF_KEY)
2620                 goto out;
2621
2622         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2623         if (btrfs_extent_refs(leaf, ei) !=
2624             btrfs_extent_data_ref_count(leaf, ref) ||
2625             btrfs_extent_data_ref_root(leaf, ref) !=
2626             root->root_key.objectid ||
2627             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2628             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2629                 goto out;
2630
2631         ret = 0;
2632 out:
2633         return ret;
2634 }
2635
2636 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2637                           struct btrfs_root *root,
2638                           u64 objectid, u64 offset, u64 bytenr)
2639 {
2640         struct btrfs_path *path;
2641         int ret;
2642         int ret2;
2643
2644         path = btrfs_alloc_path();
2645         if (!path)
2646                 return -ENOENT;
2647
2648         do {
2649                 ret = check_committed_ref(trans, root, path, objectid,
2650                                           offset, bytenr);
2651                 if (ret && ret != -ENOENT)
2652                         goto out;
2653
2654                 ret2 = check_delayed_ref(trans, root, path, objectid,
2655                                          offset, bytenr);
2656         } while (ret2 == -EAGAIN);
2657
2658         if (ret2 && ret2 != -ENOENT) {
2659                 ret = ret2;
2660                 goto out;
2661         }
2662
2663         if (ret != -ENOENT || ret2 != -ENOENT)
2664                 ret = 0;
2665 out:
2666         btrfs_free_path(path);
2667         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2668                 WARN_ON(ret > 0);
2669         return ret;
2670 }
2671
2672 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2673                            struct btrfs_root *root,
2674                            struct extent_buffer *buf,
2675                            int full_backref, int inc, int for_cow)
2676 {
2677         u64 bytenr;
2678         u64 num_bytes;
2679         u64 parent;
2680         u64 ref_root;
2681         u32 nritems;
2682         struct btrfs_key key;
2683         struct btrfs_file_extent_item *fi;
2684         int i;
2685         int level;
2686         int ret = 0;
2687         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2688                             u64, u64, u64, u64, u64, u64, int);
2689
2690         ref_root = btrfs_header_owner(buf);
2691         nritems = btrfs_header_nritems(buf);
2692         level = btrfs_header_level(buf);
2693
2694         if (!root->ref_cows && level == 0)
2695                 return 0;
2696
2697         if (inc)
2698                 process_func = btrfs_inc_extent_ref;
2699         else
2700                 process_func = btrfs_free_extent;
2701
2702         if (full_backref)
2703                 parent = buf->start;
2704         else
2705                 parent = 0;
2706
2707         for (i = 0; i < nritems; i++) {
2708                 if (level == 0) {
2709                         btrfs_item_key_to_cpu(buf, &key, i);
2710                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2711                                 continue;
2712                         fi = btrfs_item_ptr(buf, i,
2713                                             struct btrfs_file_extent_item);
2714                         if (btrfs_file_extent_type(buf, fi) ==
2715                             BTRFS_FILE_EXTENT_INLINE)
2716                                 continue;
2717                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2718                         if (bytenr == 0)
2719                                 continue;
2720
2721                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2722                         key.offset -= btrfs_file_extent_offset(buf, fi);
2723                         ret = process_func(trans, root, bytenr, num_bytes,
2724                                            parent, ref_root, key.objectid,
2725                                            key.offset, for_cow);
2726                         if (ret)
2727                                 goto fail;
2728                 } else {
2729                         bytenr = btrfs_node_blockptr(buf, i);
2730                         num_bytes = btrfs_level_size(root, level - 1);
2731                         ret = process_func(trans, root, bytenr, num_bytes,
2732                                            parent, ref_root, level - 1, 0,
2733                                            for_cow);
2734                         if (ret)
2735                                 goto fail;
2736                 }
2737         }
2738         return 0;
2739 fail:
2740         BUG();
2741         return ret;
2742 }
2743
2744 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2745                   struct extent_buffer *buf, int full_backref, int for_cow)
2746 {
2747         return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2748 }
2749
2750 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2751                   struct extent_buffer *buf, int full_backref, int for_cow)
2752 {
2753         return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2754 }
2755
2756 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2757                                  struct btrfs_root *root,
2758                                  struct btrfs_path *path,
2759                                  struct btrfs_block_group_cache *cache)
2760 {
2761         int ret;
2762         struct btrfs_root *extent_root = root->fs_info->extent_root;
2763         unsigned long bi;
2764         struct extent_buffer *leaf;
2765
2766         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2767         if (ret < 0)
2768                 goto fail;
2769         BUG_ON(ret);
2770
2771         leaf = path->nodes[0];
2772         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2773         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2774         btrfs_mark_buffer_dirty(leaf);
2775         btrfs_release_path(path);
2776 fail:
2777         if (ret)
2778                 return ret;
2779         return 0;
2780
2781 }
2782
2783 static struct btrfs_block_group_cache *
2784 next_block_group(struct btrfs_root *root,
2785                  struct btrfs_block_group_cache *cache)
2786 {
2787         struct rb_node *node;
2788         spin_lock(&root->fs_info->block_group_cache_lock);
2789         node = rb_next(&cache->cache_node);
2790         btrfs_put_block_group(cache);
2791         if (node) {
2792                 cache = rb_entry(node, struct btrfs_block_group_cache,
2793                                  cache_node);
2794                 btrfs_get_block_group(cache);
2795         } else
2796                 cache = NULL;
2797         spin_unlock(&root->fs_info->block_group_cache_lock);
2798         return cache;
2799 }
2800
2801 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2802                             struct btrfs_trans_handle *trans,
2803                             struct btrfs_path *path)
2804 {
2805         struct btrfs_root *root = block_group->fs_info->tree_root;
2806         struct inode *inode = NULL;
2807         u64 alloc_hint = 0;
2808         int dcs = BTRFS_DC_ERROR;
2809         int num_pages = 0;
2810         int retries = 0;
2811         int ret = 0;
2812
2813         /*
2814          * If this block group is smaller than 100 megs don't bother caching the
2815          * block group.
2816          */
2817         if (block_group->key.offset < (100 * 1024 * 1024)) {
2818                 spin_lock(&block_group->lock);
2819                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2820                 spin_unlock(&block_group->lock);
2821                 return 0;
2822         }
2823
2824 again:
2825         inode = lookup_free_space_inode(root, block_group, path);
2826         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2827                 ret = PTR_ERR(inode);
2828                 btrfs_release_path(path);
2829                 goto out;
2830         }
2831
2832         if (IS_ERR(inode)) {
2833                 BUG_ON(retries);
2834                 retries++;
2835
2836                 if (block_group->ro)
2837                         goto out_free;
2838
2839                 ret = create_free_space_inode(root, trans, block_group, path);
2840                 if (ret)
2841                         goto out_free;
2842                 goto again;
2843         }
2844
2845         /* We've already setup this transaction, go ahead and exit */
2846         if (block_group->cache_generation == trans->transid &&
2847             i_size_read(inode)) {
2848                 dcs = BTRFS_DC_SETUP;
2849                 goto out_put;
2850         }
2851
2852         /*
2853          * We want to set the generation to 0, that way if anything goes wrong
2854          * from here on out we know not to trust this cache when we load up next
2855          * time.
2856          */
2857         BTRFS_I(inode)->generation = 0;
2858         ret = btrfs_update_inode(trans, root, inode);
2859         WARN_ON(ret);
2860
2861         if (i_size_read(inode) > 0) {
2862                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2863                                                       inode);
2864                 if (ret)
2865                         goto out_put;
2866         }
2867
2868         spin_lock(&block_group->lock);
2869         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2870                 /* We're not cached, don't bother trying to write stuff out */
2871                 dcs = BTRFS_DC_WRITTEN;
2872                 spin_unlock(&block_group->lock);
2873                 goto out_put;
2874         }
2875         spin_unlock(&block_group->lock);
2876
2877         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2878         if (!num_pages)
2879                 num_pages = 1;
2880
2881         /*
2882          * Just to make absolutely sure we have enough space, we're going to
2883          * preallocate 12 pages worth of space for each block group.  In
2884          * practice we ought to use at most 8, but we need extra space so we can
2885          * add our header and have a terminator between the extents and the
2886          * bitmaps.
2887          */
2888         num_pages *= 16;
2889         num_pages *= PAGE_CACHE_SIZE;
2890
2891         ret = btrfs_check_data_free_space(inode, num_pages);
2892         if (ret)
2893                 goto out_put;
2894
2895         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2896                                               num_pages, num_pages,
2897                                               &alloc_hint);
2898         if (!ret)
2899                 dcs = BTRFS_DC_SETUP;
2900         btrfs_free_reserved_data_space(inode, num_pages);
2901
2902 out_put:
2903         iput(inode);
2904 out_free:
2905         btrfs_release_path(path);
2906 out:
2907         spin_lock(&block_group->lock);
2908         if (!ret && dcs == BTRFS_DC_SETUP)
2909                 block_group->cache_generation = trans->transid;
2910         block_group->disk_cache_state = dcs;
2911         spin_unlock(&block_group->lock);
2912
2913         return ret;
2914 }
2915
2916 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2917                                    struct btrfs_root *root)
2918 {
2919         struct btrfs_block_group_cache *cache;
2920         int err = 0;
2921         struct btrfs_path *path;
2922         u64 last = 0;
2923
2924         path = btrfs_alloc_path();
2925         if (!path)
2926                 return -ENOMEM;
2927
2928 again:
2929         while (1) {
2930                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2931                 while (cache) {
2932                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2933                                 break;
2934                         cache = next_block_group(root, cache);
2935                 }
2936                 if (!cache) {
2937                         if (last == 0)
2938                                 break;
2939                         last = 0;
2940                         continue;
2941                 }
2942                 err = cache_save_setup(cache, trans, path);
2943                 last = cache->key.objectid + cache->key.offset;
2944                 btrfs_put_block_group(cache);
2945         }
2946
2947         while (1) {
2948                 if (last == 0) {
2949                         err = btrfs_run_delayed_refs(trans, root,
2950                                                      (unsigned long)-1);
2951                         BUG_ON(err);
2952                 }
2953
2954                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2955                 while (cache) {
2956                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2957                                 btrfs_put_block_group(cache);
2958                                 goto again;
2959                         }
2960
2961                         if (cache->dirty)
2962                                 break;
2963                         cache = next_block_group(root, cache);
2964                 }
2965                 if (!cache) {
2966                         if (last == 0)
2967                                 break;
2968                         last = 0;
2969                         continue;
2970                 }
2971
2972                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2973                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2974                 cache->dirty = 0;
2975                 last = cache->key.objectid + cache->key.offset;
2976
2977                 err = write_one_cache_group(trans, root, path, cache);
2978                 BUG_ON(err);
2979                 btrfs_put_block_group(cache);
2980         }
2981
2982         while (1) {
2983                 /*
2984                  * I don't think this is needed since we're just marking our
2985                  * preallocated extent as written, but just in case it can't
2986                  * hurt.
2987                  */
2988                 if (last == 0) {
2989                         err = btrfs_run_delayed_refs(trans, root,
2990                                                      (unsigned long)-1);
2991                         BUG_ON(err);
2992                 }
2993
2994                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2995                 while (cache) {
2996                         /*
2997                          * Really this shouldn't happen, but it could if we
2998                          * couldn't write the entire preallocated extent and
2999                          * splitting the extent resulted in a new block.
3000                          */
3001                         if (cache->dirty) {
3002                                 btrfs_put_block_group(cache);
3003                                 goto again;
3004                         }
3005                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3006                                 break;
3007                         cache = next_block_group(root, cache);
3008                 }
3009                 if (!cache) {
3010                         if (last == 0)
3011                                 break;
3012                         last = 0;
3013                         continue;
3014                 }
3015
3016                 btrfs_write_out_cache(root, trans, cache, path);
3017
3018                 /*
3019                  * If we didn't have an error then the cache state is still
3020                  * NEED_WRITE, so we can set it to WRITTEN.
3021                  */
3022                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3023                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
3024                 last = cache->key.objectid + cache->key.offset;
3025                 btrfs_put_block_group(cache);
3026         }
3027
3028         btrfs_free_path(path);
3029         return 0;
3030 }
3031
3032 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3033 {
3034         struct btrfs_block_group_cache *block_group;
3035         int readonly = 0;
3036
3037         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3038         if (!block_group || block_group->ro)
3039                 readonly = 1;
3040         if (block_group)
3041                 btrfs_put_block_group(block_group);
3042         return readonly;
3043 }
3044
3045 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3046                              u64 total_bytes, u64 bytes_used,
3047                              struct btrfs_space_info **space_info)
3048 {
3049         struct btrfs_space_info *found;
3050         int i;
3051         int factor;
3052
3053         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3054                      BTRFS_BLOCK_GROUP_RAID10))
3055                 factor = 2;
3056         else
3057                 factor = 1;
3058
3059         found = __find_space_info(info, flags);
3060         if (found) {
3061                 spin_lock(&found->lock);
3062                 found->total_bytes += total_bytes;
3063                 found->disk_total += total_bytes * factor;
3064                 found->bytes_used += bytes_used;
3065                 found->disk_used += bytes_used * factor;
3066                 found->full = 0;
3067                 spin_unlock(&found->lock);
3068                 *space_info = found;
3069                 return 0;
3070         }
3071         found = kzalloc(sizeof(*found), GFP_NOFS);
3072         if (!found)
3073                 return -ENOMEM;
3074
3075         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3076                 INIT_LIST_HEAD(&found->block_groups[i]);
3077         init_rwsem(&found->groups_sem);
3078         spin_lock_init(&found->lock);
3079         found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3080         found->total_bytes = total_bytes;
3081         found->disk_total = total_bytes * factor;
3082         found->bytes_used = bytes_used;
3083         found->disk_used = bytes_used * factor;
3084         found->bytes_pinned = 0;
3085         found->bytes_reserved = 0;
3086         found->bytes_readonly = 0;
3087         found->bytes_may_use = 0;
3088         found->full = 0;
3089         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3090         found->chunk_alloc = 0;
3091         found->flush = 0;
3092         init_waitqueue_head(&found->wait);
3093         *space_info = found;
3094         list_add_rcu(&found->list, &info->space_info);
3095         return 0;
3096 }
3097
3098 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3099 {
3100         u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
3101
3102         /* chunk -> extended profile */
3103         if (extra_flags == 0)
3104                 extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3105
3106         if (flags & BTRFS_BLOCK_GROUP_DATA)
3107                 fs_info->avail_data_alloc_bits |= extra_flags;
3108         if (flags & BTRFS_BLOCK_GROUP_METADATA)
3109                 fs_info->avail_metadata_alloc_bits |= extra_flags;
3110         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3111                 fs_info->avail_system_alloc_bits |= extra_flags;
3112 }
3113
3114 /*
3115  * @flags: available profiles in extended format (see ctree.h)
3116  *
3117  * Returns reduced profile in chunk format.  If profile changing is in
3118  * progress (either running or paused) picks the target profile (if it's
3119  * already available), otherwise falls back to plain reducing.
3120  */
3121 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3122 {
3123         /*
3124          * we add in the count of missing devices because we want
3125          * to make sure that any RAID levels on a degraded FS
3126          * continue to be honored.
3127          */
3128         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3129                 root->fs_info->fs_devices->missing_devices;
3130
3131         /* pick restriper's target profile if it's available */
3132         spin_lock(&root->fs_info->balance_lock);
3133         if (root->fs_info->balance_ctl) {
3134                 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3135                 u64 tgt = 0;
3136
3137                 if ((flags & BTRFS_BLOCK_GROUP_DATA) &&
3138                     (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3139                     (flags & bctl->data.target)) {
3140                         tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3141                 } else if ((flags & BTRFS_BLOCK_GROUP_SYSTEM) &&
3142                            (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3143                            (flags & bctl->sys.target)) {
3144                         tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3145                 } else if ((flags & BTRFS_BLOCK_GROUP_METADATA) &&
3146                            (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3147                            (flags & bctl->meta.target)) {
3148                         tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3149                 }
3150
3151                 if (tgt) {
3152                         spin_unlock(&root->fs_info->balance_lock);
3153                         flags = tgt;
3154                         goto out;
3155                 }
3156         }
3157         spin_unlock(&root->fs_info->balance_lock);
3158
3159         if (num_devices == 1)
3160                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3161         if (num_devices < 4)
3162                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3163
3164         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3165             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3166                       BTRFS_BLOCK_GROUP_RAID10))) {
3167                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3168         }
3169
3170         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3171             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3172                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3173         }
3174
3175         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3176             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3177              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3178              (flags & BTRFS_BLOCK_GROUP_DUP))) {
3179                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3180         }
3181
3182 out:
3183         /* extended -> chunk profile */
3184         flags &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3185         return flags;
3186 }
3187
3188 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3189 {
3190         if (flags & BTRFS_BLOCK_GROUP_DATA)
3191                 flags |= root->fs_info->avail_data_alloc_bits;
3192         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3193                 flags |= root->fs_info->avail_system_alloc_bits;
3194         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3195                 flags |= root->fs_info->avail_metadata_alloc_bits;
3196
3197         return btrfs_reduce_alloc_profile(root, flags);
3198 }
3199
3200 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3201 {
3202         u64 flags;
3203
3204         if (data)
3205                 flags = BTRFS_BLOCK_GROUP_DATA;
3206         else if (root == root->fs_info->chunk_root)
3207                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3208         else
3209                 flags = BTRFS_BLOCK_GROUP_METADATA;
3210
3211         return get_alloc_profile(root, flags);
3212 }
3213
3214 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3215 {
3216         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3217                                                        BTRFS_BLOCK_GROUP_DATA);
3218 }
3219
3220 /*
3221  * This will check the space that the inode allocates from to make sure we have
3222  * enough space for bytes.
3223  */
3224 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3225 {
3226         struct btrfs_space_info *data_sinfo;
3227         struct btrfs_root *root = BTRFS_I(inode)->root;
3228         u64 used;
3229         int ret = 0, committed = 0, alloc_chunk = 1;
3230
3231         /* make sure bytes are sectorsize aligned */
3232         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3233
3234         if (root == root->fs_info->tree_root ||
3235             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3236                 alloc_chunk = 0;
3237                 committed = 1;
3238         }
3239
3240         data_sinfo = BTRFS_I(inode)->space_info;
3241         if (!data_sinfo)
3242                 goto alloc;
3243
3244 again:
3245         /* make sure we have enough space to handle the data first */
3246         spin_lock(&data_sinfo->lock);
3247         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3248                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3249                 data_sinfo->bytes_may_use;
3250
3251         if (used + bytes > data_sinfo->total_bytes) {
3252                 struct btrfs_trans_handle *trans;
3253
3254                 /*
3255                  * if we don't have enough free bytes in this space then we need
3256                  * to alloc a new chunk.
3257                  */
3258                 if (!data_sinfo->full && alloc_chunk) {
3259                         u64 alloc_target;
3260
3261                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3262                         spin_unlock(&data_sinfo->lock);
3263 alloc:
3264                         alloc_target = btrfs_get_alloc_profile(root, 1);
3265                         trans = btrfs_join_transaction(root);
3266                         if (IS_ERR(trans))
3267                                 return PTR_ERR(trans);
3268
3269                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3270                                              bytes + 2 * 1024 * 1024,
3271                                              alloc_target,
3272                                              CHUNK_ALLOC_NO_FORCE);
3273                         btrfs_end_transaction(trans, root);
3274                         if (ret < 0) {
3275                                 if (ret != -ENOSPC)
3276                                         return ret;
3277                                 else
3278                                         goto commit_trans;
3279                         }
3280
3281                         if (!data_sinfo) {
3282                                 btrfs_set_inode_space_info(root, inode);
3283                                 data_sinfo = BTRFS_I(inode)->space_info;
3284                         }
3285                         goto again;
3286                 }
3287
3288                 /*
3289                  * If we have less pinned bytes than we want to allocate then
3290                  * don't bother committing the transaction, it won't help us.
3291                  */
3292                 if (data_sinfo->bytes_pinned < bytes)
3293                         committed = 1;
3294                 spin_unlock(&data_sinfo->lock);
3295
3296                 /* commit the current transaction and try again */
3297 commit_trans:
3298                 if (!committed &&
3299                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3300                         committed = 1;
3301                         trans = btrfs_join_transaction(root);
3302                         if (IS_ERR(trans))
3303                                 return PTR_ERR(trans);
3304                         ret = btrfs_commit_transaction(trans, root);
3305                         if (ret)
3306                                 return ret;
3307                         goto again;
3308                 }
3309
3310                 return -ENOSPC;
3311         }
3312         data_sinfo->bytes_may_use += bytes;
3313         spin_unlock(&data_sinfo->lock);
3314
3315         return 0;
3316 }
3317
3318 /*
3319  * Called if we need to clear a data reservation for this inode.
3320  */
3321 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3322 {
3323         struct btrfs_root *root = BTRFS_I(inode)->root;
3324         struct btrfs_space_info *data_sinfo;
3325
3326         /* make sure bytes are sectorsize aligned */
3327         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3328
3329         data_sinfo = BTRFS_I(inode)->space_info;
3330         spin_lock(&data_sinfo->lock);
3331         data_sinfo->bytes_may_use -= bytes;
3332         spin_unlock(&data_sinfo->lock);
3333 }
3334
3335 static void force_metadata_allocation(struct btrfs_fs_info *info)
3336 {
3337         struct list_head *head = &info->space_info;
3338         struct btrfs_space_info *found;
3339
3340         rcu_read_lock();
3341         list_for_each_entry_rcu(found, head, list) {
3342                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3343                         found->force_alloc = CHUNK_ALLOC_FORCE;
3344         }
3345         rcu_read_unlock();
3346 }
3347
3348 static int should_alloc_chunk(struct btrfs_root *root,
3349                               struct btrfs_space_info *sinfo, u64 alloc_bytes,
3350                               int force)
3351 {
3352         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3353         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3354         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3355         u64 thresh;
3356
3357         if (force == CHUNK_ALLOC_FORCE)
3358                 return 1;
3359
3360         /*
3361          * We need to take into account the global rsv because for all intents
3362          * and purposes it's used space.  Don't worry about locking the
3363          * global_rsv, it doesn't change except when the transaction commits.
3364          */
3365         num_allocated += global_rsv->size;
3366
3367         /*
3368          * in limited mode, we want to have some free space up to
3369          * about 1% of the FS size.
3370          */
3371         if (force == CHUNK_ALLOC_LIMITED) {
3372                 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3373                 thresh = max_t(u64, 64 * 1024 * 1024,
3374                                div_factor_fine(thresh, 1));
3375
3376                 if (num_bytes - num_allocated < thresh)
3377                         return 1;
3378         }
3379         thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3380
3381         /* 256MB or 2% of the FS */
3382         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3383
3384         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3385                 return 0;
3386         return 1;
3387 }
3388
3389 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3390                           struct btrfs_root *extent_root, u64 alloc_bytes,
3391                           u64 flags, int force)
3392 {
3393         struct btrfs_space_info *space_info;
3394         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3395         int wait_for_alloc = 0;
3396         int ret = 0;
3397
3398         BUG_ON(!profile_is_valid(flags, 0));
3399
3400         space_info = __find_space_info(extent_root->fs_info, flags);
3401         if (!space_info) {
3402                 ret = update_space_info(extent_root->fs_info, flags,
3403                                         0, 0, &space_info);
3404                 BUG_ON(ret);
3405         }
3406         BUG_ON(!space_info);
3407
3408 again:
3409         spin_lock(&space_info->lock);
3410         if (space_info->force_alloc)
3411                 force = space_info->force_alloc;
3412         if (space_info->full) {
3413                 spin_unlock(&space_info->lock);
3414                 return 0;
3415         }
3416
3417         if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3418                 spin_unlock(&space_info->lock);
3419                 return 0;
3420         } else if (space_info->chunk_alloc) {
3421                 wait_for_alloc = 1;
3422         } else {
3423                 space_info->chunk_alloc = 1;
3424         }
3425
3426         spin_unlock(&space_info->lock);
3427
3428         mutex_lock(&fs_info->chunk_mutex);
3429
3430         /*
3431          * The chunk_mutex is held throughout the entirety of a chunk
3432          * allocation, so once we've acquired the chunk_mutex we know that the
3433          * other guy is done and we need to recheck and see if we should
3434          * allocate.
3435          */
3436         if (wait_for_alloc) {
3437                 mutex_unlock(&fs_info->chunk_mutex);
3438                 wait_for_alloc = 0;
3439                 goto again;
3440         }
3441
3442         /*
3443          * If we have mixed data/metadata chunks we want to make sure we keep
3444          * allocating mixed chunks instead of individual chunks.
3445          */
3446         if (btrfs_mixed_space_info(space_info))
3447                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3448
3449         /*
3450          * if we're doing a data chunk, go ahead and make sure that
3451          * we keep a reasonable number of metadata chunks allocated in the
3452          * FS as well.
3453          */
3454         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3455                 fs_info->data_chunk_allocations++;
3456                 if (!(fs_info->data_chunk_allocations %
3457                       fs_info->metadata_ratio))
3458                         force_metadata_allocation(fs_info);
3459         }
3460
3461         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3462         if (ret < 0 && ret != -ENOSPC)
3463                 goto out;
3464
3465         spin_lock(&space_info->lock);
3466         if (ret)
3467                 space_info->full = 1;
3468         else
3469                 ret = 1;
3470
3471         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3472         space_info->chunk_alloc = 0;
3473         spin_unlock(&space_info->lock);
3474 out:
3475         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3476         return ret;
3477 }
3478
3479 /*
3480  * shrink metadata reservation for delalloc
3481  */
3482 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3483                            bool wait_ordered)
3484 {
3485         struct btrfs_block_rsv *block_rsv;
3486         struct btrfs_space_info *space_info;
3487         struct btrfs_trans_handle *trans;
3488         u64 reserved;
3489         u64 max_reclaim;
3490         u64 reclaimed = 0;
3491         long time_left;
3492         unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3493         int loops = 0;
3494         unsigned long progress;
3495
3496         trans = (struct btrfs_trans_handle *)current->journal_info;
3497         block_rsv = &root->fs_info->delalloc_block_rsv;
3498         space_info = block_rsv->space_info;
3499
3500         smp_mb();
3501         reserved = space_info->bytes_may_use;
3502         progress = space_info->reservation_progress;
3503
3504         if (reserved == 0)
3505                 return 0;
3506
3507         smp_mb();
3508         if (root->fs_info->delalloc_bytes == 0) {
3509                 if (trans)
3510                         return 0;
3511                 btrfs_wait_ordered_extents(root, 0, 0);
3512                 return 0;
3513         }
3514
3515         max_reclaim = min(reserved, to_reclaim);
3516         nr_pages = max_t(unsigned long, nr_pages,
3517                          max_reclaim >> PAGE_CACHE_SHIFT);
3518         while (loops < 1024) {
3519                 /* have the flusher threads jump in and do some IO */
3520                 smp_mb();
3521                 nr_pages = min_t(unsigned long, nr_pages,
3522                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3523                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3524                                                 WB_REASON_FS_FREE_SPACE);
3525
3526                 spin_lock(&space_info->lock);
3527                 if (reserved > space_info->bytes_may_use)
3528                         reclaimed += reserved - space_info->bytes_may_use;
3529                 reserved = space_info->bytes_may_use;
3530                 spin_unlock(&space_info->lock);
3531
3532                 loops++;
3533
3534                 if (reserved == 0 || reclaimed >= max_reclaim)
3535                         break;
3536
3537                 if (trans && trans->transaction->blocked)
3538                         return -EAGAIN;
3539
3540                 if (wait_ordered && !trans) {
3541                         btrfs_wait_ordered_extents(root, 0, 0);
3542                 } else {
3543                         time_left = schedule_timeout_interruptible(1);
3544
3545                         /* We were interrupted, exit */
3546                         if (time_left)
3547                                 break;
3548                 }
3549
3550                 /* we've kicked the IO a few times, if anything has been freed,
3551                  * exit.  There is no sense in looping here for a long time
3552                  * when we really need to commit the transaction, or there are
3553                  * just too many writers without enough free space
3554                  */
3555
3556                 if (loops > 3) {
3557                         smp_mb();
3558                         if (progress != space_info->reservation_progress)
3559                                 break;
3560                 }
3561
3562         }
3563
3564         return reclaimed >= to_reclaim;
3565 }
3566
3567 /**
3568  * maybe_commit_transaction - possibly commit the transaction if its ok to
3569  * @root - the root we're allocating for
3570  * @bytes - the number of bytes we want to reserve
3571  * @force - force the commit
3572  *
3573  * This will check to make sure that committing the transaction will actually
3574  * get us somewhere and then commit the transaction if it does.  Otherwise it
3575  * will return -ENOSPC.
3576  */
3577 static int may_commit_transaction(struct btrfs_root *root,
3578                                   struct btrfs_space_info *space_info,
3579                                   u64 bytes, int force)
3580 {
3581         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3582         struct btrfs_trans_handle *trans;
3583
3584         trans = (struct btrfs_trans_handle *)current->journal_info;
3585         if (trans)
3586                 return -EAGAIN;
3587
3588         if (force)
3589                 goto commit;
3590
3591         /* See if there is enough pinned space to make this reservation */
3592         spin_lock(&space_info->lock);
3593         if (space_info->bytes_pinned >= bytes) {
3594                 spin_unlock(&space_info->lock);
3595                 goto commit;
3596         }
3597         spin_unlock(&space_info->lock);
3598
3599         /*
3600          * See if there is some space in the delayed insertion reservation for
3601          * this reservation.
3602          */
3603         if (space_info != delayed_rsv->space_info)
3604                 return -ENOSPC;
3605
3606         spin_lock(&delayed_rsv->lock);
3607         if (delayed_rsv->size < bytes) {
3608                 spin_unlock(&delayed_rsv->lock);
3609                 return -ENOSPC;
3610         }
3611         spin_unlock(&delayed_rsv->lock);
3612
3613 commit:
3614         trans = btrfs_join_transaction(root);
3615         if (IS_ERR(trans))
3616                 return -ENOSPC;
3617
3618         return btrfs_commit_transaction(trans, root);
3619 }
3620
3621 /**
3622  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3623  * @root - the root we're allocating for
3624  * @block_rsv - the block_rsv we're allocating for
3625  * @orig_bytes - the number of bytes we want
3626  * @flush - wether or not we can flush to make our reservation
3627  *
3628  * This will reserve orgi_bytes number of bytes from the space info associated
3629  * with the block_rsv.  If there is not enough space it will make an attempt to
3630  * flush out space to make room.  It will do this by flushing delalloc if
3631  * possible or committing the transaction.  If flush is 0 then no attempts to
3632  * regain reservations will be made and this will fail if there is not enough
3633  * space already.
3634  */
3635 static int reserve_metadata_bytes(struct btrfs_root *root,
3636                                   struct btrfs_block_rsv *block_rsv,
3637                                   u64 orig_bytes, int flush)
3638 {
3639         struct btrfs_space_info *space_info = block_rsv->space_info;
3640         u64 used;
3641         u64 num_bytes = orig_bytes;
3642         int retries = 0;
3643         int ret = 0;
3644         bool committed = false;
3645         bool flushing = false;
3646         bool wait_ordered = false;
3647
3648 again:
3649         ret = 0;
3650         spin_lock(&space_info->lock);
3651         /*
3652          * We only want to wait if somebody other than us is flushing and we are
3653          * actually alloed to flush.
3654          */
3655         while (flush && !flushing && space_info->flush) {
3656                 spin_unlock(&space_info->lock);
3657                 /*
3658                  * If we have a trans handle we can't wait because the flusher
3659                  * may have to commit the transaction, which would mean we would
3660                  * deadlock since we are waiting for the flusher to finish, but
3661                  * hold the current transaction open.
3662                  */
3663                 if (current->journal_info)
3664                         return -EAGAIN;
3665                 ret = wait_event_interruptible(space_info->wait,
3666                                                !space_info->flush);
3667                 /* Must have been interrupted, return */
3668                 if (ret)
3669                         return -EINTR;
3670
3671                 spin_lock(&space_info->lock);
3672         }
3673
3674         ret = -ENOSPC;
3675         used = space_info->bytes_used + space_info->bytes_reserved +
3676                 space_info->bytes_pinned + space_info->bytes_readonly +
3677                 space_info->bytes_may_use;
3678
3679         /*
3680          * The idea here is that we've not already over-reserved the block group
3681          * then we can go ahead and save our reservation first and then start
3682          * flushing if we need to.  Otherwise if we've already overcommitted
3683          * lets start flushing stuff first and then come back and try to make
3684          * our reservation.
3685          */
3686         if (used <= space_info->total_bytes) {
3687                 if (used + orig_bytes <= space_info->total_bytes) {
3688                         space_info->bytes_may_use += orig_bytes;
3689                         ret = 0;
3690                 } else {
3691                         /*
3692                          * Ok set num_bytes to orig_bytes since we aren't
3693                          * overocmmitted, this way we only try and reclaim what
3694                          * we need.
3695                          */
3696                         num_bytes = orig_bytes;
3697                 }
3698         } else {
3699                 /*
3700                  * Ok we're over committed, set num_bytes to the overcommitted
3701                  * amount plus the amount of bytes that we need for this
3702                  * reservation.
3703                  */
3704                 wait_ordered = true;
3705                 num_bytes = used - space_info->total_bytes +
3706                         (orig_bytes * (retries + 1));
3707         }
3708
3709         if (ret) {
3710                 u64 profile = btrfs_get_alloc_profile(root, 0);
3711                 u64 avail;
3712
3713                 /*
3714                  * If we have a lot of space that's pinned, don't bother doing
3715                  * the overcommit dance yet and just commit the transaction.
3716                  */
3717                 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3718                 do_div(avail, 10);
3719                 if (space_info->bytes_pinned >= avail && flush && !committed) {
3720                         space_info->flush = 1;
3721                         flushing = true;
3722                         spin_unlock(&space_info->lock);
3723                         ret = may_commit_transaction(root, space_info,
3724                                                      orig_bytes, 1);
3725                         if (ret)
3726                                 goto out;
3727                         committed = true;
3728                         goto again;
3729                 }
3730
3731                 spin_lock(&root->fs_info->free_chunk_lock);
3732                 avail = root->fs_info->free_chunk_space;
3733
3734                 /*
3735                  * If we have dup, raid1 or raid10 then only half of the free
3736                  * space is actually useable.
3737                  */
3738                 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3739                                BTRFS_BLOCK_GROUP_RAID1 |
3740                                BTRFS_BLOCK_GROUP_RAID10))
3741                         avail >>= 1;
3742
3743                 /*
3744                  * If we aren't flushing don't let us overcommit too much, say
3745                  * 1/8th of the space.  If we can flush, let it overcommit up to
3746                  * 1/2 of the space.
3747                  */
3748                 if (flush)
3749                         avail >>= 3;
3750                 else
3751                         avail >>= 1;
3752                  spin_unlock(&root->fs_info->free_chunk_lock);
3753
3754                 if (used + num_bytes < space_info->total_bytes + avail) {
3755                         space_info->bytes_may_use += orig_bytes;
3756                         ret = 0;
3757                 } else {
3758                         wait_ordered = true;
3759                 }
3760         }
3761
3762         /*
3763          * Couldn't make our reservation, save our place so while we're trying
3764          * to reclaim space we can actually use it instead of somebody else
3765          * stealing it from us.
3766          */
3767         if (ret && flush) {
3768                 flushing = true;
3769                 space_info->flush = 1;
3770         }
3771
3772         spin_unlock(&space_info->lock);
3773
3774         if (!ret || !flush)
3775                 goto out;
3776
3777         /*
3778          * We do synchronous shrinking since we don't actually unreserve
3779          * metadata until after the IO is completed.
3780          */
3781         ret = shrink_delalloc(root, num_bytes, wait_ordered);
3782         if (ret < 0)
3783                 goto out;
3784
3785         ret = 0;
3786
3787         /*
3788          * So if we were overcommitted it's possible that somebody else flushed
3789          * out enough space and we simply didn't have enough space to reclaim,
3790          * so go back around and try again.
3791          */
3792         if (retries < 2) {
3793                 wait_ordered = true;
3794                 retries++;
3795                 goto again;
3796         }
3797
3798         ret = -ENOSPC;
3799         if (committed)
3800                 goto out;
3801
3802         ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3803         if (!ret) {
3804                 committed = true;
3805                 goto again;
3806         }
3807
3808 out:
3809         if (flushing) {
3810                 spin_lock(&space_info->lock);
3811                 space_info->flush = 0;
3812                 wake_up_all(&space_info->wait);
3813                 spin_unlock(&space_info->lock);
3814         }
3815         return ret;
3816 }
3817
3818 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3819                                              struct btrfs_root *root)
3820 {
3821         struct btrfs_block_rsv *block_rsv = NULL;
3822
3823         if (root->ref_cows || root == root->fs_info->csum_root)
3824                 block_rsv = trans->block_rsv;
3825
3826         if (!block_rsv)
3827                 block_rsv = root->block_rsv;
3828
3829         if (!block_rsv)
3830                 block_rsv = &root->fs_info->empty_block_rsv;
3831
3832         return block_rsv;
3833 }
3834
3835 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3836                                u64 num_bytes)
3837 {
3838         int ret = -ENOSPC;
3839         spin_lock(&block_rsv->lock);
3840         if (block_rsv->reserved >= num_bytes) {
3841                 block_rsv->reserved -= num_bytes;
3842                 if (block_rsv->reserved < block_rsv->size)
3843                         block_rsv->full = 0;
3844                 ret = 0;
3845         }
3846         spin_unlock(&block_rsv->lock);
3847         return ret;
3848 }
3849
3850 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3851                                 u64 num_bytes, int update_size)
3852 {
3853         spin_lock(&block_rsv->lock);
3854         block_rsv->reserved += num_bytes;
3855         if (update_size)
3856                 block_rsv->size += num_bytes;
3857         else if (block_rsv->reserved >= block_rsv->size)
3858                 block_rsv->full = 1;
3859         spin_unlock(&block_rsv->lock);
3860 }
3861
3862 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3863                                     struct btrfs_block_rsv *dest, u64 num_bytes)
3864 {
3865         struct btrfs_space_info *space_info = block_rsv->space_info;
3866
3867         spin_lock(&block_rsv->lock);
3868         if (num_bytes == (u64)-1)
3869                 num_bytes = block_rsv->size;
3870         block_rsv->size -= num_bytes;
3871         if (block_rsv->reserved >= block_rsv->size) {
3872                 num_bytes = block_rsv->reserved - block_rsv->size;
3873                 block_rsv->reserved = block_rsv->size;
3874                 block_rsv->full = 1;
3875         } else {
3876                 num_bytes = 0;
3877         }
3878         spin_unlock(&block_rsv->lock);
3879
3880         if (num_bytes > 0) {
3881                 if (dest) {
3882                         spin_lock(&dest->lock);
3883                         if (!dest->full) {
3884                                 u64 bytes_to_add;
3885
3886                                 bytes_to_add = dest->size - dest->reserved;
3887                                 bytes_to_add = min(num_bytes, bytes_to_add);
3888                                 dest->reserved += bytes_to_add;
3889                                 if (dest->reserved >= dest->size)
3890                                         dest->full = 1;
3891                                 num_bytes -= bytes_to_add;
3892                         }
3893                         spin_unlock(&dest->lock);
3894                 }
3895                 if (num_bytes) {
3896                         spin_lock(&space_info->lock);
3897                         space_info->bytes_may_use -= num_bytes;
3898                         space_info->reservation_progress++;
3899                         spin_unlock(&space_info->lock);
3900                 }
3901         }
3902 }
3903
3904 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3905                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3906 {
3907         int ret;
3908
3909         ret = block_rsv_use_bytes(src, num_bytes);
3910         if (ret)
3911                 return ret;
3912
3913         block_rsv_add_bytes(dst, num_bytes, 1);
3914         return 0;
3915 }
3916
3917 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3918 {
3919         memset(rsv, 0, sizeof(*rsv));
3920         spin_lock_init(&rsv->lock);
3921 }
3922
3923 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3924 {
3925         struct btrfs_block_rsv *block_rsv;
3926         struct btrfs_fs_info *fs_info = root->fs_info;
3927
3928         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3929         if (!block_rsv)
3930                 return NULL;
3931
3932         btrfs_init_block_rsv(block_rsv);
3933         block_rsv->space_info = __find_space_info(fs_info,
3934                                                   BTRFS_BLOCK_GROUP_METADATA);
3935         return block_rsv;
3936 }
3937
3938 void btrfs_free_block_rsv(struct btrfs_root *root,
3939                           struct btrfs_block_rsv *rsv)
3940 {
3941         btrfs_block_rsv_release(root, rsv, (u64)-1);
3942         kfree(rsv);
3943 }
3944
3945 static inline int __block_rsv_add(struct btrfs_root *root,
3946                                   struct btrfs_block_rsv *block_rsv,
3947                                   u64 num_bytes, int flush)
3948 {
3949         int ret;
3950
3951         if (num_bytes == 0)
3952                 return 0;
3953
3954         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3955         if (!ret) {
3956                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3957                 return 0;
3958         }
3959
3960         return ret;
3961 }
3962
3963 int btrfs_block_rsv_add(struct btrfs_root *root,
3964                         struct btrfs_block_rsv *block_rsv,
3965                         u64 num_bytes)
3966 {
3967         return __block_rsv_add(root, block_rsv, num_bytes, 1);
3968 }
3969
3970 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3971                                 struct btrfs_block_rsv *block_rsv,
3972                                 u64 num_bytes)
3973 {
3974         return __block_rsv_add(root, block_rsv, num_bytes, 0);
3975 }
3976
3977 int btrfs_block_rsv_check(struct btrfs_root *root,
3978                           struct btrfs_block_rsv *block_rsv, int min_factor)
3979 {
3980         u64 num_bytes = 0;
3981         int ret = -ENOSPC;
3982
3983         if (!block_rsv)
3984                 return 0;
3985
3986         spin_lock(&block_rsv->lock);
3987         num_bytes = div_factor(block_rsv->size, min_factor);
3988         if (block_rsv->reserved >= num_bytes)
3989                 ret = 0;
3990         spin_unlock(&block_rsv->lock);
3991
3992         return ret;
3993 }
3994
3995 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
3996                                            struct btrfs_block_rsv *block_rsv,
3997                                            u64 min_reserved, int flush)
3998 {
3999         u64 num_bytes = 0;
4000         int ret = -ENOSPC;
4001
4002         if (!block_rsv)
4003                 return 0;
4004
4005         spin_lock(&block_rsv->lock);
4006         num_bytes = min_reserved;
4007         if (block_rsv->reserved >= num_bytes)
4008                 ret = 0;
4009         else
4010                 num_bytes -= block_rsv->reserved;
4011         spin_unlock(&block_rsv->lock);
4012
4013         if (!ret)
4014                 return 0;
4015
4016         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4017         if (!ret) {
4018                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4019                 return 0;
4020         }
4021
4022         return ret;
4023 }
4024
4025 int btrfs_block_rsv_refill(struct btrfs_root *root,
4026                            struct btrfs_block_rsv *block_rsv,
4027                            u64 min_reserved)
4028 {
4029         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4030 }
4031
4032 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4033                                    struct btrfs_block_rsv *block_rsv,
4034                                    u64 min_reserved)
4035 {
4036         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4037 }
4038
4039 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4040                             struct btrfs_block_rsv *dst_rsv,
4041                             u64 num_bytes)
4042 {
4043         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4044 }
4045
4046 void btrfs_block_rsv_release(struct btrfs_root *root,
4047                              struct btrfs_block_rsv *block_rsv,
4048                              u64 num_bytes)
4049 {
4050         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4051         if (global_rsv->full || global_rsv == block_rsv ||
4052             block_rsv->space_info != global_rsv->space_info)
4053                 global_rsv = NULL;
4054         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
4055 }
4056
4057 /*
4058  * helper to calculate size of global block reservation.
4059  * the desired value is sum of space used by extent tree,
4060  * checksum tree and root tree
4061  */
4062 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4063 {
4064         struct btrfs_space_info *sinfo;
4065         u64 num_bytes;
4066         u64 meta_used;
4067         u64 data_used;
4068         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4069
4070         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4071         spin_lock(&sinfo->lock);
4072         data_used = sinfo->bytes_used;
4073         spin_unlock(&sinfo->lock);
4074
4075         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4076         spin_lock(&sinfo->lock);
4077         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4078                 data_used = 0;
4079         meta_used = sinfo->bytes_used;
4080         spin_unlock(&sinfo->lock);
4081
4082         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4083                     csum_size * 2;
4084         num_bytes += div64_u64(data_used + meta_used, 50);
4085
4086         if (num_bytes * 3 > meta_used)
4087                 num_bytes = div64_u64(meta_used, 3);
4088
4089         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4090 }
4091
4092 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4093 {
4094         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4095         struct btrfs_space_info *sinfo = block_rsv->space_info;
4096         u64 num_bytes;
4097
4098         num_bytes = calc_global_metadata_size(fs_info);
4099
4100         spin_lock(&block_rsv->lock);
4101         spin_lock(&sinfo->lock);
4102
4103         block_rsv->size = num_bytes;
4104
4105         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4106                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4107                     sinfo->bytes_may_use;
4108
4109         if (sinfo->total_bytes > num_bytes) {
4110                 num_bytes = sinfo->total_bytes - num_bytes;
4111                 block_rsv->reserved += num_bytes;
4112                 sinfo->bytes_may_use += num_bytes;
4113         }
4114
4115         if (block_rsv->reserved >= block_rsv->size) {
4116                 num_bytes = block_rsv->reserved - block_rsv->size;
4117                 sinfo->bytes_may_use -= num_bytes;
4118                 sinfo->reservation_progress++;
4119                 block_rsv->reserved = block_rsv->size;
4120                 block_rsv->full = 1;
4121         }
4122
4123         spin_unlock(&sinfo->lock);
4124         spin_unlock(&block_rsv->lock);
4125 }
4126
4127 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4128 {
4129         struct btrfs_space_info *space_info;
4130
4131         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4132         fs_info->chunk_block_rsv.space_info = space_info;
4133
4134         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4135         fs_info->global_block_rsv.space_info = space_info;
4136         fs_info->delalloc_block_rsv.space_info = space_info;
4137         fs_info->trans_block_rsv.space_info = space_info;
4138         fs_info->empty_block_rsv.space_info = space_info;
4139         fs_info->delayed_block_rsv.space_info = space_info;
4140
4141         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4142         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4143         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4144         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4145         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4146
4147         update_global_block_rsv(fs_info);
4148 }
4149
4150 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4151 {
4152         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
4153         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4154         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4155         WARN_ON(fs_info->trans_block_rsv.size > 0);
4156         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4157         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4158         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4159         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4160         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4161 }
4162
4163 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4164                                   struct btrfs_root *root)
4165 {
4166         if (!trans->bytes_reserved)
4167                 return;
4168
4169         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4170         trans->bytes_reserved = 0;
4171 }
4172
4173 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4174                                   struct inode *inode)
4175 {
4176         struct btrfs_root *root = BTRFS_I(inode)->root;
4177         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4178         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4179
4180         /*
4181          * We need to hold space in order to delete our orphan item once we've
4182          * added it, so this takes the reservation so we can release it later
4183          * when we are truly done with the orphan item.
4184          */
4185         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4186         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4187 }
4188
4189 void btrfs_orphan_release_metadata(struct inode *inode)
4190 {
4191         struct btrfs_root *root = BTRFS_I(inode)->root;
4192         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4193         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4194 }
4195
4196 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4197                                 struct btrfs_pending_snapshot *pending)
4198 {
4199         struct btrfs_root *root = pending->root;
4200         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4201         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4202         /*
4203          * two for root back/forward refs, two for directory entries
4204          * and one for root of the snapshot.
4205          */
4206         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4207         dst_rsv->space_info = src_rsv->space_info;
4208         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4209 }
4210
4211 /**
4212  * drop_outstanding_extent - drop an outstanding extent
4213  * @inode: the inode we're dropping the extent for
4214  *
4215  * This is called when we are freeing up an outstanding extent, either called
4216  * after an error or after an extent is written.  This will return the number of
4217  * reserved extents that need to be freed.  This must be called with
4218  * BTRFS_I(inode)->lock held.
4219  */
4220 static unsigned drop_outstanding_extent(struct inode *inode)
4221 {
4222         unsigned drop_inode_space = 0;
4223         unsigned dropped_extents = 0;
4224
4225         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4226         BTRFS_I(inode)->outstanding_extents--;
4227
4228         if (BTRFS_I(inode)->outstanding_extents == 0 &&
4229             BTRFS_I(inode)->delalloc_meta_reserved) {
4230                 drop_inode_space = 1;
4231                 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4232         }
4233
4234         /*
4235          * If we have more or the same amount of outsanding extents than we have
4236          * reserved then we need to leave the reserved extents count alone.
4237          */
4238         if (BTRFS_I(inode)->outstanding_extents >=
4239             BTRFS_I(inode)->reserved_extents)
4240                 return drop_inode_space;
4241
4242         dropped_extents = BTRFS_I(inode)->reserved_extents -
4243                 BTRFS_I(inode)->outstanding_extents;
4244         BTRFS_I(inode)->reserved_extents -= dropped_extents;
4245         return dropped_extents + drop_inode_space;
4246 }
4247
4248 /**
4249  * calc_csum_metadata_size - return the amount of metada space that must be
4250  *      reserved/free'd for the given bytes.
4251  * @inode: the inode we're manipulating
4252  * @num_bytes: the number of bytes in question
4253  * @reserve: 1 if we are reserving space, 0 if we are freeing space
4254  *
4255  * This adjusts the number of csum_bytes in the inode and then returns the
4256  * correct amount of metadata that must either be reserved or freed.  We
4257  * calculate how many checksums we can fit into one leaf and then divide the
4258  * number of bytes that will need to be checksumed by this value to figure out
4259  * how many checksums will be required.  If we are adding bytes then the number
4260  * may go up and we will return the number of additional bytes that must be
4261  * reserved.  If it is going down we will return the number of bytes that must
4262  * be freed.
4263  *
4264  * This must be called with BTRFS_I(inode)->lock held.
4265  */
4266 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4267                                    int reserve)
4268 {
4269         struct btrfs_root *root = BTRFS_I(inode)->root;
4270         u64 csum_size;
4271         int num_csums_per_leaf;
4272         int num_csums;
4273         int old_csums;
4274
4275         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4276             BTRFS_I(inode)->csum_bytes == 0)
4277                 return 0;
4278
4279         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4280         if (reserve)
4281                 BTRFS_I(inode)->csum_bytes += num_bytes;
4282         else
4283                 BTRFS_I(inode)->csum_bytes -= num_bytes;
4284         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4285         num_csums_per_leaf = (int)div64_u64(csum_size,
4286                                             sizeof(struct btrfs_csum_item) +
4287                                             sizeof(struct btrfs_disk_key));
4288         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4289         num_csums = num_csums + num_csums_per_leaf - 1;
4290         num_csums = num_csums / num_csums_per_leaf;
4291
4292         old_csums = old_csums + num_csums_per_leaf - 1;
4293         old_csums = old_csums / num_csums_per_leaf;
4294
4295         /* No change, no need to reserve more */
4296         if (old_csums == num_csums)
4297                 return 0;
4298
4299         if (reserve)
4300                 return btrfs_calc_trans_metadata_size(root,
4301                                                       num_csums - old_csums);
4302
4303         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4304 }
4305
4306 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4307 {
4308         struct btrfs_root *root = BTRFS_I(inode)->root;
4309         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4310         u64 to_reserve = 0;
4311         u64 csum_bytes;
4312         unsigned nr_extents = 0;
4313         int extra_reserve = 0;
4314         int flush = 1;
4315         int ret;
4316
4317         /* Need to be holding the i_mutex here if we aren't free space cache */
4318         if (btrfs_is_free_space_inode(root, inode))
4319                 flush = 0;
4320         else
4321                 WARN_ON(!mutex_is_locked(&inode->i_mutex));
4322
4323         if (flush && btrfs_transaction_in_commit(root->fs_info))
4324                 schedule_timeout(1);
4325
4326         num_bytes = ALIGN(num_bytes, root->sectorsize);
4327
4328         spin_lock(&BTRFS_I(inode)->lock);
4329         BTRFS_I(inode)->outstanding_extents++;
4330
4331         if (BTRFS_I(inode)->outstanding_extents >
4332             BTRFS_I(inode)->reserved_extents)
4333                 nr_extents = BTRFS_I(inode)->outstanding_extents -
4334                         BTRFS_I(inode)->reserved_extents;
4335
4336         /*
4337          * Add an item to reserve for updating the inode when we complete the
4338          * delalloc io.
4339          */
4340         if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4341                 nr_extents++;
4342                 extra_reserve = 1;
4343         }
4344
4345         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4346         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4347         csum_bytes = BTRFS_I(inode)->csum_bytes;
4348         spin_unlock(&BTRFS_I(inode)->lock);
4349
4350         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4351         if (ret) {
4352                 u64 to_free = 0;
4353                 unsigned dropped;
4354
4355                 spin_lock(&BTRFS_I(inode)->lock);
4356                 dropped = drop_outstanding_extent(inode);
4357                 /*
4358                  * If the inodes csum_bytes is the same as the original
4359                  * csum_bytes then we know we haven't raced with any free()ers
4360                  * so we can just reduce our inodes csum bytes and carry on.
4361                  * Otherwise we have to do the normal free thing to account for
4362                  * the case that the free side didn't free up its reserve
4363                  * because of this outstanding reservation.
4364                  */
4365                 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4366                         calc_csum_metadata_size(inode, num_bytes, 0);
4367                 else
4368                         to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4369                 spin_unlock(&BTRFS_I(inode)->lock);
4370                 if (dropped)
4371                         to_free += btrfs_calc_trans_metadata_size(root, dropped);
4372
4373                 if (to_free)
4374                         btrfs_block_rsv_release(root, block_rsv, to_free);
4375                 return ret;
4376         }
4377
4378         spin_lock(&BTRFS_I(inode)->lock);
4379         if (extra_reserve) {
4380                 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4381                 nr_extents--;
4382         }
4383         BTRFS_I(inode)->reserved_extents += nr_extents;
4384         spin_unlock(&BTRFS_I(inode)->lock);
4385
4386         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4387
4388         return 0;
4389 }
4390
4391 /**
4392  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4393  * @inode: the inode to release the reservation for
4394  * @num_bytes: the number of bytes we're releasing
4395  *
4396  * This will release the metadata reservation for an inode.  This can be called
4397  * once we complete IO for a given set of bytes to release their metadata
4398  * reservations.
4399  */
4400 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4401 {
4402         struct btrfs_root *root = BTRFS_I(inode)->root;
4403         u64 to_free = 0;
4404         unsigned dropped;
4405
4406         num_bytes = ALIGN(num_bytes, root->sectorsize);
4407         spin_lock(&BTRFS_I(inode)->lock);
4408         dropped = drop_outstanding_extent(inode);
4409
4410         to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4411         spin_unlock(&BTRFS_I(inode)->lock);
4412         if (dropped > 0)
4413                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4414
4415         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4416                                 to_free);
4417 }
4418
4419 /**
4420  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4421  * @inode: inode we're writing to
4422  * @num_bytes: the number of bytes we want to allocate
4423  *
4424  * This will do the following things
4425  *
4426  * o reserve space in the data space info for num_bytes
4427  * o reserve space in the metadata space info based on number of outstanding
4428  *   extents and how much csums will be needed
4429  * o add to the inodes ->delalloc_bytes
4430  * o add it to the fs_info's delalloc inodes list.
4431  *
4432  * This will return 0 for success and -ENOSPC if there is no space left.
4433  */
4434 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4435 {
4436         int ret;
4437
4438         ret = btrfs_check_data_free_space(inode, num_bytes);
4439         if (ret)
4440                 return ret;
4441
4442         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4443         if (ret) {
4444                 btrfs_free_reserved_data_space(inode, num_bytes);
4445                 return ret;
4446         }
4447
4448         return 0;
4449 }
4450
4451 /**
4452  * btrfs_delalloc_release_space - release data and metadata space for delalloc
4453  * @inode: inode we're releasing space for
4454  * @num_bytes: the number of bytes we want to free up
4455  *
4456  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
4457  * called in the case that we don't need the metadata AND data reservations
4458  * anymore.  So if there is an error or we insert an inline extent.
4459  *
4460  * This function will release the metadata space that was not used and will
4461  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4462  * list if there are no delalloc bytes left.
4463  */
4464 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4465 {
4466         btrfs_delalloc_release_metadata(inode, num_bytes);
4467         btrfs_free_reserved_data_space(inode, num_bytes);
4468 }
4469
4470 static int update_block_group(struct btrfs_trans_handle *trans,
4471                               struct btrfs_root *root,
4472                               u64 bytenr, u64 num_bytes, int alloc)
4473 {
4474         struct btrfs_block_group_cache *cache = NULL;
4475         struct btrfs_fs_info *info = root->fs_info;
4476         u64 total = num_bytes;
4477         u64 old_val;
4478         u64 byte_in_group;
4479         int factor;
4480
4481         /* block accounting for super block */
4482         spin_lock(&info->delalloc_lock);
4483         old_val = btrfs_super_bytes_used(info->super_copy);
4484         if (alloc)
4485                 old_val += num_bytes;
4486         else
4487                 old_val -= num_bytes;
4488         btrfs_set_super_bytes_used(info->super_copy, old_val);
4489         spin_unlock(&info->delalloc_lock);
4490
4491         while (total) {
4492                 cache = btrfs_lookup_block_group(info, bytenr);
4493                 if (!cache)
4494                         return -1;
4495                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4496                                     BTRFS_BLOCK_GROUP_RAID1 |
4497                                     BTRFS_BLOCK_GROUP_RAID10))
4498                         factor = 2;
4499                 else
4500                         factor = 1;
4501                 /*
4502                  * If this block group has free space cache written out, we
4503                  * need to make sure to load it if we are removing space.  This
4504                  * is because we need the unpinning stage to actually add the
4505                  * space back to the block group, otherwise we will leak space.
4506                  */
4507                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4508                         cache_block_group(cache, trans, NULL, 1);
4509
4510                 byte_in_group = bytenr - cache->key.objectid;
4511                 WARN_ON(byte_in_group > cache->key.offset);
4512
4513                 spin_lock(&cache->space_info->lock);
4514                 spin_lock(&cache->lock);
4515
4516                 if (btrfs_test_opt(root, SPACE_CACHE) &&
4517                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4518                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4519
4520                 cache->dirty = 1;
4521                 old_val = btrfs_block_group_used(&cache->item);
4522                 num_bytes = min(total, cache->key.offset - byte_in_group);
4523                 if (alloc) {
4524                         old_val += num_bytes;
4525                         btrfs_set_block_group_used(&cache->item, old_val);
4526                         cache->reserved -= num_bytes;
4527                         cache->space_info->bytes_reserved -= num_bytes;
4528                         cache->space_info->bytes_used += num_bytes;
4529                         cache->space_info->disk_used += num_bytes * factor;
4530                         spin_unlock(&cache->lock);
4531                         spin_unlock(&cache->space_info->lock);
4532                 } else {
4533                         old_val -= num_bytes;
4534                         btrfs_set_block_group_used(&cache->item, old_val);
4535                         cache->pinned += num_bytes;
4536                         cache->space_info->bytes_pinned += num_bytes;
4537                         cache->space_info->bytes_used -= num_bytes;
4538                         cache->space_info->disk_used -= num_bytes * factor;
4539                         spin_unlock(&cache->lock);
4540                         spin_unlock(&cache->space_info->lock);
4541
4542                         set_extent_dirty(info->pinned_extents,
4543                                          bytenr, bytenr + num_bytes - 1,
4544                                          GFP_NOFS | __GFP_NOFAIL);
4545                 }
4546                 btrfs_put_block_group(cache);
4547                 total -= num_bytes;
4548                 bytenr += num_bytes;
4549         }
4550         return 0;
4551 }
4552
4553 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4554 {
4555         struct btrfs_block_group_cache *cache;
4556         u64 bytenr;
4557
4558         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4559         if (!cache)
4560                 return 0;
4561
4562         bytenr = cache->key.objectid;
4563         btrfs_put_block_group(cache);
4564
4565         return bytenr;
4566 }
4567
4568 static int pin_down_extent(struct btrfs_root *root,
4569                            struct btrfs_block_group_cache *cache,
4570                            u64 bytenr, u64 num_bytes, int reserved)
4571 {
4572         spin_lock(&cache->space_info->lock);
4573         spin_lock(&cache->lock);
4574         cache->pinned += num_bytes;
4575         cache->space_info->bytes_pinned += num_bytes;
4576         if (reserved) {
4577                 cache->reserved -= num_bytes;
4578                 cache->space_info->bytes_reserved -= num_bytes;
4579         }
4580         spin_unlock(&cache->lock);
4581         spin_unlock(&cache->space_info->lock);
4582
4583         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4584                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4585         return 0;
4586 }
4587
4588 /*
4589  * this function must be called within transaction
4590  */
4591 int btrfs_pin_extent(struct btrfs_root *root,
4592                      u64 bytenr, u64 num_bytes, int reserved)
4593 {
4594         struct btrfs_block_group_cache *cache;
4595
4596         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4597         BUG_ON(!cache);
4598
4599         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4600
4601         btrfs_put_block_group(cache);
4602         return 0;
4603 }
4604
4605 /*
4606  * this function must be called within transaction
4607  */
4608 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4609                                     struct btrfs_root *root,
4610                                     u64 bytenr, u64 num_bytes)
4611 {
4612         struct btrfs_block_group_cache *cache;
4613
4614         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4615         BUG_ON(!cache);
4616
4617         /*
4618          * pull in the free space cache (if any) so that our pin
4619          * removes the free space from the cache.  We have load_only set
4620          * to one because the slow code to read in the free extents does check
4621          * the pinned extents.
4622          */
4623         cache_block_group(cache, trans, root, 1);
4624
4625         pin_down_extent(root, cache, bytenr, num_bytes, 0);
4626
4627         /* remove us from the free space cache (if we're there at all) */
4628         btrfs_remove_free_space(cache, bytenr, num_bytes);
4629         btrfs_put_block_group(cache);
4630         return 0;
4631 }
4632
4633 /**
4634  * btrfs_update_reserved_bytes - update the block_group and space info counters
4635  * @cache:      The cache we are manipulating
4636  * @num_bytes:  The number of bytes in question
4637  * @reserve:    One of the reservation enums
4638  *
4639  * This is called by the allocator when it reserves space, or by somebody who is
4640  * freeing space that was never actually used on disk.  For example if you
4641  * reserve some space for a new leaf in transaction A and before transaction A
4642  * commits you free that leaf, you call this with reserve set to 0 in order to
4643  * clear the reservation.
4644  *
4645  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4646  * ENOSPC accounting.  For data we handle the reservation through clearing the
4647  * delalloc bits in the io_tree.  We have to do this since we could end up
4648  * allocating less disk space for the amount of data we have reserved in the
4649  * case of compression.
4650  *
4651  * If this is a reservation and the block group has become read only we cannot
4652  * make the reservation and return -EAGAIN, otherwise this function always
4653  * succeeds.
4654  */
4655 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4656                                        u64 num_bytes, int reserve)
4657 {
4658         struct btrfs_space_info *space_info = cache->space_info;
4659         int ret = 0;
4660         spin_lock(&space_info->lock);
4661         spin_lock(&cache->lock);
4662         if (reserve != RESERVE_FREE) {
4663                 if (cache->ro) {
4664                         ret = -EAGAIN;
4665                 } else {
4666                         cache->reserved += num_bytes;
4667                         space_info->bytes_reserved += num_bytes;
4668                         if (reserve == RESERVE_ALLOC) {
4669                                 BUG_ON(space_info->bytes_may_use < num_bytes);
4670                                 space_info->bytes_may_use -= num_bytes;
4671                         }
4672                 }
4673         } else {
4674                 if (cache->ro)
4675                         space_info->bytes_readonly += num_bytes;
4676                 cache->reserved -= num_bytes;
4677                 space_info->bytes_reserved -= num_bytes;
4678                 space_info->reservation_progress++;
4679         }
4680         spin_unlock(&cache->lock);
4681         spin_unlock(&space_info->lock);
4682         return ret;
4683 }
4684
4685 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4686                                 struct btrfs_root *root)
4687 {
4688         struct btrfs_fs_info *fs_info = root->fs_info;
4689         struct btrfs_caching_control *next;
4690         struct btrfs_caching_control *caching_ctl;
4691         struct btrfs_block_group_cache *cache;
4692
4693         down_write(&fs_info->extent_commit_sem);
4694
4695         list_for_each_entry_safe(caching_ctl, next,
4696                                  &fs_info->caching_block_groups, list) {
4697                 cache = caching_ctl->block_group;
4698                 if (block_group_cache_done(cache)) {
4699                         cache->last_byte_to_unpin = (u64)-1;
4700                         list_del_init(&caching_ctl->list);
4701                         put_caching_control(caching_ctl);
4702                 } else {
4703                         cache->last_byte_to_unpin = caching_ctl->progress;
4704                 }
4705         }
4706
4707         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4708                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4709         else
4710                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4711
4712         up_write(&fs_info->extent_commit_sem);
4713
4714         update_global_block_rsv(fs_info);
4715         return 0;
4716 }
4717
4718 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4719 {
4720         struct btrfs_fs_info *fs_info = root->fs_info;
4721         struct btrfs_block_group_cache *cache = NULL;
4722         u64 len;
4723
4724         while (start <= end) {
4725                 if (!cache ||
4726                     start >= cache->key.objectid + cache->key.offset) {
4727                         if (cache)
4728                                 btrfs_put_block_group(cache);
4729                         cache = btrfs_lookup_block_group(fs_info, start);
4730                         BUG_ON(!cache);
4731                 }
4732
4733                 len = cache->key.objectid + cache->key.offset - start;
4734                 len = min(len, end + 1 - start);
4735
4736                 if (start < cache->last_byte_to_unpin) {
4737                         len = min(len, cache->last_byte_to_unpin - start);
4738                         btrfs_add_free_space(cache, start, len);
4739                 }
4740
4741                 start += len;
4742
4743                 spin_lock(&cache->space_info->lock);
4744                 spin_lock(&cache->lock);
4745                 cache->pinned -= len;
4746                 cache->space_info->bytes_pinned -= len;
4747                 if (cache->ro)
4748                         cache->space_info->bytes_readonly += len;
4749                 spin_unlock(&cache->lock);
4750                 spin_unlock(&cache->space_info->lock);
4751         }
4752
4753         if (cache)
4754                 btrfs_put_block_group(cache);
4755         return 0;
4756 }
4757
4758 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4759                                struct btrfs_root *root)
4760 {
4761         struct btrfs_fs_info *fs_info = root->fs_info;
4762         struct extent_io_tree *unpin;
4763         u64 start;
4764         u64 end;
4765         int ret;
4766
4767         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4768                 unpin = &fs_info->freed_extents[1];
4769         else
4770                 unpin = &fs_info->freed_extents[0];
4771
4772         while (1) {
4773                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4774                                             EXTENT_DIRTY);
4775                 if (ret)
4776                         break;
4777
4778                 if (btrfs_test_opt(root, DISCARD))
4779                         ret = btrfs_discard_extent(root, start,
4780                                                    end + 1 - start, NULL);
4781
4782                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4783                 unpin_extent_range(root, start, end);
4784                 cond_resched();
4785         }
4786
4787         return 0;
4788 }
4789
4790 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4791                                 struct btrfs_root *root,
4792                                 u64 bytenr, u64 num_bytes, u64 parent,
4793                                 u64 root_objectid, u64 owner_objectid,
4794                                 u64 owner_offset, int refs_to_drop,
4795                                 struct btrfs_delayed_extent_op *extent_op)
4796 {
4797         struct btrfs_key key;
4798         struct btrfs_path *path;
4799         struct btrfs_fs_info *info = root->fs_info;
4800         struct btrfs_root *extent_root = info->extent_root;
4801         struct extent_buffer *leaf;
4802         struct btrfs_extent_item *ei;
4803         struct btrfs_extent_inline_ref *iref;
4804         int ret;
4805         int is_data;
4806         int extent_slot = 0;
4807         int found_extent = 0;
4808         int num_to_del = 1;
4809         u32 item_size;
4810         u64 refs;
4811
4812         path = btrfs_alloc_path();
4813         if (!path)
4814                 return -ENOMEM;
4815
4816         path->reada = 1;
4817         path->leave_spinning = 1;
4818
4819         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4820         BUG_ON(!is_data && refs_to_drop != 1);
4821
4822         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4823                                     bytenr, num_bytes, parent,
4824                                     root_objectid, owner_objectid,
4825                                     owner_offset);
4826         if (ret == 0) {
4827                 extent_slot = path->slots[0];
4828                 while (extent_slot >= 0) {
4829                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4830                                               extent_slot);
4831                         if (key.objectid != bytenr)
4832                                 break;
4833                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4834                             key.offset == num_bytes) {
4835                                 found_extent = 1;
4836                                 break;
4837                         }
4838                         if (path->slots[0] - extent_slot > 5)
4839                                 break;
4840                         extent_slot--;
4841                 }
4842 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4843                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4844                 if (found_extent && item_size < sizeof(*ei))
4845                         found_extent = 0;
4846 #endif
4847                 if (!found_extent) {
4848                         BUG_ON(iref);
4849                         ret = remove_extent_backref(trans, extent_root, path,
4850                                                     NULL, refs_to_drop,
4851                                                     is_data);
4852                         BUG_ON(ret);
4853                         btrfs_release_path(path);
4854                         path->leave_spinning = 1;
4855
4856                         key.objectid = bytenr;
4857                         key.type = BTRFS_EXTENT_ITEM_KEY;
4858                         key.offset = num_bytes;
4859
4860                         ret = btrfs_search_slot(trans, extent_root,
4861                                                 &key, path, -1, 1);
4862                         if (ret) {
4863                                 printk(KERN_ERR "umm, got %d back from search"
4864                                        ", was looking for %llu\n", ret,
4865                                        (unsigned long long)bytenr);
4866                                 if (ret > 0)
4867                                         btrfs_print_leaf(extent_root,
4868                                                          path->nodes[0]);
4869                         }
4870                         BUG_ON(ret);
4871                         extent_slot = path->slots[0];
4872                 }
4873         } else {
4874                 btrfs_print_leaf(extent_root, path->nodes[0]);
4875                 WARN_ON(1);
4876                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4877                        "parent %llu root %llu  owner %llu offset %llu\n",
4878                        (unsigned long long)bytenr,
4879                        (unsigned long long)parent,
4880                        (unsigned long long)root_objectid,
4881                        (unsigned long long)owner_objectid,
4882                        (unsigned long long)owner_offset);
4883         }
4884
4885         leaf = path->nodes[0];
4886         item_size = btrfs_item_size_nr(leaf, extent_slot);
4887 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4888         if (item_size < sizeof(*ei)) {
4889                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4890                 ret = convert_extent_item_v0(trans, extent_root, path,
4891                                              owner_objectid, 0);
4892                 BUG_ON(ret < 0);
4893
4894                 btrfs_release_path(path);
4895                 path->leave_spinning = 1;
4896
4897                 key.objectid = bytenr;
4898                 key.type = BTRFS_EXTENT_ITEM_KEY;
4899                 key.offset = num_bytes;
4900
4901                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4902                                         -1, 1);
4903                 if (ret) {
4904                         printk(KERN_ERR "umm, got %d back from search"
4905                                ", was looking for %llu\n", ret,
4906                                (unsigned long long)bytenr);
4907                         btrfs_print_leaf(extent_root, path->nodes[0]);
4908                 }
4909                 BUG_ON(ret);
4910                 extent_slot = path->slots[0];
4911                 leaf = path->nodes[0];
4912                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4913         }
4914 #endif
4915         BUG_ON(item_size < sizeof(*ei));
4916         ei = btrfs_item_ptr(leaf, extent_slot,
4917                             struct btrfs_extent_item);
4918         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4919                 struct btrfs_tree_block_info *bi;
4920                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4921                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4922                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4923         }
4924
4925         refs = btrfs_extent_refs(leaf, ei);
4926         BUG_ON(refs < refs_to_drop);
4927         refs -= refs_to_drop;
4928
4929         if (refs > 0) {
4930                 if (extent_op)
4931                         __run_delayed_extent_op(extent_op, leaf, ei);
4932                 /*
4933                  * In the case of inline back ref, reference count will
4934                  * be updated by remove_extent_backref
4935                  */
4936                 if (iref) {
4937                         BUG_ON(!found_extent);
4938                 } else {
4939                         btrfs_set_extent_refs(leaf, ei, refs);
4940                         btrfs_mark_buffer_dirty(leaf);
4941                 }
4942                 if (found_extent) {
4943                         ret = remove_extent_backref(trans, extent_root, path,
4944                                                     iref, refs_to_drop,
4945                                                     is_data);
4946                         BUG_ON(ret);
4947                 }
4948         } else {
4949                 if (found_extent) {
4950                         BUG_ON(is_data && refs_to_drop !=
4951                                extent_data_ref_count(root, path, iref));
4952                         if (iref) {
4953                                 BUG_ON(path->slots[0] != extent_slot);
4954                         } else {
4955                                 BUG_ON(path->slots[0] != extent_slot + 1);
4956                                 path->slots[0] = extent_slot;
4957                                 num_to_del = 2;
4958                         }
4959                 }
4960
4961                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4962                                       num_to_del);
4963                 BUG_ON(ret);
4964                 btrfs_release_path(path);
4965
4966                 if (is_data) {
4967                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4968                         BUG_ON(ret);
4969                 } else {
4970                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4971                              bytenr >> PAGE_CACHE_SHIFT,
4972                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4973                 }
4974
4975                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4976                 BUG_ON(ret);
4977         }
4978         btrfs_free_path(path);
4979         return ret;
4980 }
4981
4982 /*
4983  * when we free an block, it is possible (and likely) that we free the last
4984  * delayed ref for that extent as well.  This searches the delayed ref tree for
4985  * a given extent, and if there are no other delayed refs to be processed, it
4986  * removes it from the tree.
4987  */
4988 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4989                                       struct btrfs_root *root, u64 bytenr)
4990 {
4991         struct btrfs_delayed_ref_head *head;
4992         struct btrfs_delayed_ref_root *delayed_refs;
4993         struct btrfs_delayed_ref_node *ref;
4994         struct rb_node *node;
4995         int ret = 0;
4996
4997         delayed_refs = &trans->transaction->delayed_refs;
4998         spin_lock(&delayed_refs->lock);
4999         head = btrfs_find_delayed_ref_head(trans, bytenr);
5000         if (!head)
5001                 goto out;
5002
5003         node = rb_prev(&head->node.rb_node);
5004         if (!node)
5005                 goto out;
5006
5007         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5008
5009         /* there are still entries for this ref, we can't drop it */
5010         if (ref->bytenr == bytenr)
5011                 goto out;
5012
5013         if (head->extent_op) {
5014                 if (!head->must_insert_reserved)
5015                         goto out;
5016                 kfree(head->extent_op);
5017                 head->extent_op = NULL;
5018         }
5019
5020         /*
5021          * waiting for the lock here would deadlock.  If someone else has it
5022          * locked they are already in the process of dropping it anyway
5023          */
5024         if (!mutex_trylock(&head->mutex))
5025                 goto out;
5026
5027         /*
5028          * at this point we have a head with no other entries.  Go
5029          * ahead and process it.
5030          */
5031         head->node.in_tree = 0;
5032         rb_erase(&head->node.rb_node, &delayed_refs->root);
5033
5034         delayed_refs->num_entries--;
5035         if (waitqueue_active(&delayed_refs->seq_wait))
5036                 wake_up(&delayed_refs->seq_wait);
5037
5038         /*
5039          * we don't take a ref on the node because we're removing it from the
5040          * tree, so we just steal the ref the tree was holding.
5041          */
5042         delayed_refs->num_heads--;
5043         if (list_empty(&head->cluster))
5044                 delayed_refs->num_heads_ready--;
5045
5046         list_del_init(&head->cluster);
5047         spin_unlock(&delayed_refs->lock);
5048
5049         BUG_ON(head->extent_op);
5050         if (head->must_insert_reserved)
5051                 ret = 1;
5052
5053         mutex_unlock(&head->mutex);
5054         btrfs_put_delayed_ref(&head->node);
5055         return ret;
5056 out:
5057         spin_unlock(&delayed_refs->lock);
5058         return 0;
5059 }
5060
5061 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5062                            struct btrfs_root *root,
5063                            struct extent_buffer *buf,
5064                            u64 parent, int last_ref, int for_cow)
5065 {
5066         struct btrfs_block_group_cache *cache = NULL;
5067         int ret;
5068
5069         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5070                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5071                                         buf->start, buf->len,
5072                                         parent, root->root_key.objectid,
5073                                         btrfs_header_level(buf),
5074                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5075                 BUG_ON(ret);
5076         }
5077
5078         if (!last_ref)
5079                 return;
5080
5081         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5082
5083         if (btrfs_header_generation(buf) == trans->transid) {
5084                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5085                         ret = check_ref_cleanup(trans, root, buf->start);
5086                         if (!ret)
5087                                 goto out;
5088                 }
5089
5090                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5091                         pin_down_extent(root, cache, buf->start, buf->len, 1);
5092                         goto out;
5093                 }
5094
5095                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5096
5097                 btrfs_add_free_space(cache, buf->start, buf->len);
5098                 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5099         }
5100 out:
5101         /*
5102          * Deleting the buffer, clear the corrupt flag since it doesn't matter
5103          * anymore.
5104          */
5105         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5106         btrfs_put_block_group(cache);
5107 }
5108
5109 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5110                       u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5111                       u64 owner, u64 offset, int for_cow)
5112 {
5113         int ret;
5114         struct btrfs_fs_info *fs_info = root->fs_info;
5115
5116         /*
5117          * tree log blocks never actually go into the extent allocation
5118          * tree, just update pinning info and exit early.
5119          */
5120         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5121                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5122                 /* unlocks the pinned mutex */
5123                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5124                 ret = 0;
5125         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5126                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5127                                         num_bytes,
5128                                         parent, root_objectid, (int)owner,
5129                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5130                 BUG_ON(ret);
5131         } else {
5132                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5133                                                 num_bytes,
5134                                                 parent, root_objectid, owner,
5135                                                 offset, BTRFS_DROP_DELAYED_REF,
5136                                                 NULL, for_cow);
5137                 BUG_ON(ret);
5138         }
5139         return ret;
5140 }
5141
5142 static u64 stripe_align(struct btrfs_root *root, u64 val)
5143 {
5144         u64 mask = ((u64)root->stripesize - 1);
5145         u64 ret = (val + mask) & ~mask;
5146         return ret;
5147 }
5148
5149 /*
5150  * when we wait for progress in the block group caching, its because
5151  * our allocation attempt failed at least once.  So, we must sleep
5152  * and let some progress happen before we try again.
5153  *
5154  * This function will sleep at least once waiting for new free space to
5155  * show up, and then it will check the block group free space numbers
5156  * for our min num_bytes.  Another option is to have it go ahead
5157  * and look in the rbtree for a free extent of a given size, but this
5158  * is a good start.
5159  */
5160 static noinline int
5161 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5162                                 u64 num_bytes)
5163 {
5164         struct btrfs_caching_control *caching_ctl;
5165         DEFINE_WAIT(wait);
5166
5167         caching_ctl = get_caching_control(cache);
5168         if (!caching_ctl)
5169                 return 0;
5170
5171         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5172                    (cache->free_space_ctl->free_space >= num_bytes));
5173
5174         put_caching_control(caching_ctl);
5175         return 0;
5176 }
5177
5178 static noinline int
5179 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5180 {
5181         struct btrfs_caching_control *caching_ctl;
5182         DEFINE_WAIT(wait);
5183
5184         caching_ctl = get_caching_control(cache);
5185         if (!caching_ctl)
5186                 return 0;
5187
5188         wait_event(caching_ctl->wait, block_group_cache_done(cache));
5189
5190         put_caching_control(caching_ctl);
5191         return 0;
5192 }
5193
5194 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5195 {
5196         int index;
5197         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5198                 index = 0;
5199         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5200                 index = 1;
5201         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5202                 index = 2;
5203         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5204                 index = 3;
5205         else
5206                 index = 4;
5207         return index;
5208 }
5209
5210 enum btrfs_loop_type {
5211         LOOP_FIND_IDEAL = 0,
5212         LOOP_CACHING_NOWAIT = 1,
5213         LOOP_CACHING_WAIT = 2,
5214         LOOP_ALLOC_CHUNK = 3,
5215         LOOP_NO_EMPTY_SIZE = 4,
5216 };
5217
5218 /*
5219  * walks the btree of allocated extents and find a hole of a given size.
5220  * The key ins is changed to record the hole:
5221  * ins->objectid == block start
5222  * ins->flags = BTRFS_EXTENT_ITEM_KEY
5223  * ins->offset == number of blocks
5224  * Any available blocks before search_start are skipped.
5225  */
5226 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5227                                      struct btrfs_root *orig_root,
5228                                      u64 num_bytes, u64 empty_size,
5229                                      u64 search_start, u64 search_end,
5230                                      u64 hint_byte, struct btrfs_key *ins,
5231                                      u64 data)
5232 {
5233         int ret = 0;
5234         struct btrfs_root *root = orig_root->fs_info->extent_root;
5235         struct btrfs_free_cluster *last_ptr = NULL;
5236         struct btrfs_block_group_cache *block_group = NULL;
5237         struct btrfs_block_group_cache *used_block_group;
5238         int empty_cluster = 2 * 1024 * 1024;
5239         int allowed_chunk_alloc = 0;
5240         int done_chunk_alloc = 0;
5241         struct btrfs_space_info *space_info;
5242         int loop = 0;
5243         int index = 0;
5244         int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5245                 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5246         bool found_uncached_bg = false;
5247         bool failed_cluster_refill = false;
5248         bool failed_alloc = false;
5249         bool use_cluster = true;
5250         bool have_caching_bg = false;
5251         u64 ideal_cache_percent = 0;
5252         u64 ideal_cache_offset = 0;
5253
5254         WARN_ON(num_bytes < root->sectorsize);
5255         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5256         ins->objectid = 0;
5257         ins->offset = 0;
5258
5259         space_info = __find_space_info(root->fs_info, data);
5260         if (!space_info) {
5261                 printk(KERN_ERR "No space info for %llu\n", data);
5262                 return -ENOSPC;
5263         }
5264
5265         /*
5266          * If the space info is for both data and metadata it means we have a
5267          * small filesystem and we can't use the clustering stuff.
5268          */
5269         if (btrfs_mixed_space_info(space_info))
5270                 use_cluster = false;
5271
5272         if (orig_root->ref_cows || empty_size)
5273                 allowed_chunk_alloc = 1;
5274
5275         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5276                 last_ptr = &root->fs_info->meta_alloc_cluster;
5277                 if (!btrfs_test_opt(root, SSD))
5278                         empty_cluster = 64 * 1024;
5279         }
5280
5281         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5282             btrfs_test_opt(root, SSD)) {
5283                 last_ptr = &root->fs_info->data_alloc_cluster;
5284         }
5285
5286         if (last_ptr) {
5287                 spin_lock(&last_ptr->lock);
5288                 if (last_ptr->block_group)
5289                         hint_byte = last_ptr->window_start;
5290                 spin_unlock(&last_ptr->lock);
5291         }
5292
5293         search_start = max(search_start, first_logical_byte(root, 0));
5294         search_start = max(search_start, hint_byte);
5295
5296         if (!last_ptr)
5297                 empty_cluster = 0;
5298
5299         if (search_start == hint_byte) {
5300 ideal_cache:
5301                 block_group = btrfs_lookup_block_group(root->fs_info,
5302                                                        search_start);
5303                 used_block_group = block_group;
5304                 /*
5305                  * we don't want to use the block group if it doesn't match our
5306                  * allocation bits, or if its not cached.
5307                  *
5308                  * However if we are re-searching with an ideal block group
5309                  * picked out then we don't care that the block group is cached.
5310                  */
5311                 if (block_group && block_group_bits(block_group, data) &&
5312                     (block_group->cached != BTRFS_CACHE_NO ||
5313                      search_start == ideal_cache_offset)) {
5314                         down_read(&space_info->groups_sem);
5315                         if (list_empty(&block_group->list) ||
5316                             block_group->ro) {
5317                                 /*
5318                                  * someone is removing this block group,
5319                                  * we can't jump into the have_block_group
5320                                  * target because our list pointers are not
5321                                  * valid
5322                                  */
5323                                 btrfs_put_block_group(block_group);
5324                                 up_read(&space_info->groups_sem);
5325                         } else {
5326                                 index = get_block_group_index(block_group);
5327                                 goto have_block_group;
5328                         }
5329                 } else if (block_group) {
5330                         btrfs_put_block_group(block_group);
5331                 }
5332         }
5333 search:
5334         have_caching_bg = false;
5335         down_read(&space_info->groups_sem);
5336         list_for_each_entry(block_group, &space_info->block_groups[index],
5337                             list) {
5338                 u64 offset;
5339                 int cached;
5340
5341                 used_block_group = block_group;
5342                 btrfs_get_block_group(block_group);
5343                 search_start = block_group->key.objectid;
5344
5345                 /*
5346                  * this can happen if we end up cycling through all the
5347                  * raid types, but we want to make sure we only allocate
5348                  * for the proper type.
5349                  */
5350                 if (!block_group_bits(block_group, data)) {
5351                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
5352                                 BTRFS_BLOCK_GROUP_RAID1 |
5353                                 BTRFS_BLOCK_GROUP_RAID10;
5354
5355                         /*
5356                          * if they asked for extra copies and this block group
5357                          * doesn't provide them, bail.  This does allow us to
5358                          * fill raid0 from raid1.
5359                          */
5360                         if ((data & extra) && !(block_group->flags & extra))
5361                                 goto loop;
5362                 }
5363
5364 have_block_group:
5365                 cached = block_group_cache_done(block_group);
5366                 if (unlikely(!cached)) {
5367                         u64 free_percent;
5368
5369                         found_uncached_bg = true;
5370                         ret = cache_block_group(block_group, trans,
5371                                                 orig_root, 1);
5372                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5373                                 goto alloc;
5374
5375                         free_percent = btrfs_block_group_used(&block_group->item);
5376                         free_percent *= 100;
5377                         free_percent = div64_u64(free_percent,
5378                                                  block_group->key.offset);
5379                         free_percent = 100 - free_percent;
5380                         if (free_percent > ideal_cache_percent &&
5381                             likely(!block_group->ro)) {
5382                                 ideal_cache_offset = block_group->key.objectid;
5383                                 ideal_cache_percent = free_percent;
5384                         }
5385
5386                         /*
5387                          * The caching workers are limited to 2 threads, so we
5388                          * can queue as much work as we care to.
5389                          */
5390                         if (loop > LOOP_FIND_IDEAL) {
5391                                 ret = cache_block_group(block_group, trans,
5392                                                         orig_root, 0);
5393                                 BUG_ON(ret);
5394                         }
5395
5396                         /*
5397                          * If loop is set for cached only, try the next block
5398                          * group.
5399                          */
5400                         if (loop == LOOP_FIND_IDEAL)
5401                                 goto loop;
5402                 }
5403
5404 alloc:
5405                 if (unlikely(block_group->ro))
5406                         goto loop;
5407
5408                 /*
5409                  * Ok we want to try and use the cluster allocator, so
5410                  * lets look there
5411                  */
5412                 if (last_ptr) {
5413                         /*
5414                          * the refill lock keeps out other
5415                          * people trying to start a new cluster
5416                          */
5417                         spin_lock(&last_ptr->refill_lock);
5418                         used_block_group = last_ptr->block_group;
5419                         if (used_block_group != block_group &&
5420                             (!used_block_group ||
5421                              used_block_group->ro ||
5422                              !block_group_bits(used_block_group, data))) {
5423                                 used_block_group = block_group;
5424                                 goto refill_cluster;
5425                         }
5426
5427                         if (used_block_group != block_group)
5428                                 btrfs_get_block_group(used_block_group);
5429
5430                         offset = btrfs_alloc_from_cluster(used_block_group,
5431                           last_ptr, num_bytes, used_block_group->key.objectid);
5432                         if (offset) {
5433                                 /* we have a block, we're done */
5434                                 spin_unlock(&last_ptr->refill_lock);
5435                                 goto checks;
5436                         }
5437
5438                         WARN_ON(last_ptr->block_group != used_block_group);
5439                         if (used_block_group != block_group) {
5440                                 btrfs_put_block_group(used_block_group);
5441                                 used_block_group = block_group;
5442                         }
5443 refill_cluster:
5444                         BUG_ON(used_block_group != block_group);
5445                         /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5446                          * set up a new clusters, so lets just skip it
5447                          * and let the allocator find whatever block
5448                          * it can find.  If we reach this point, we
5449                          * will have tried the cluster allocator
5450                          * plenty of times and not have found
5451                          * anything, so we are likely way too
5452                          * fragmented for the clustering stuff to find
5453                          * anything.
5454                          *
5455                          * However, if the cluster is taken from the
5456                          * current block group, release the cluster
5457                          * first, so that we stand a better chance of
5458                          * succeeding in the unclustered
5459                          * allocation.  */
5460                         if (loop >= LOOP_NO_EMPTY_SIZE &&
5461                             last_ptr->block_group != block_group) {
5462                                 spin_unlock(&last_ptr->refill_lock);
5463                                 goto unclustered_alloc;
5464                         }
5465
5466                         /*
5467                          * this cluster didn't work out, free it and
5468                          * start over
5469                          */
5470                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5471
5472                         if (loop >= LOOP_NO_EMPTY_SIZE) {
5473                                 spin_unlock(&last_ptr->refill_lock);
5474                                 goto unclustered_alloc;
5475                         }
5476
5477                         /* allocate a cluster in this block group */
5478                         ret = btrfs_find_space_cluster(trans, root,
5479                                                block_group, last_ptr,
5480                                                search_start, num_bytes,
5481                                                empty_cluster + empty_size);
5482                         if (ret == 0) {
5483                                 /*
5484                                  * now pull our allocation out of this
5485                                  * cluster
5486                                  */
5487                                 offset = btrfs_alloc_from_cluster(block_group,
5488                                                   last_ptr, num_bytes,
5489                                                   search_start);
5490                                 if (offset) {
5491                                         /* we found one, proceed */
5492                                         spin_unlock(&last_ptr->refill_lock);
5493                                         goto checks;
5494                                 }
5495                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5496                                    && !failed_cluster_refill) {
5497                                 spin_unlock(&last_ptr->refill_lock);
5498
5499                                 failed_cluster_refill = true;
5500                                 wait_block_group_cache_progress(block_group,
5501                                        num_bytes + empty_cluster + empty_size);
5502                                 goto have_block_group;
5503                         }
5504
5505                         /*
5506                          * at this point we either didn't find a cluster
5507                          * or we weren't able to allocate a block from our
5508                          * cluster.  Free the cluster we've been trying
5509                          * to use, and go to the next block group
5510                          */
5511                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5512                         spin_unlock(&last_ptr->refill_lock);
5513                         goto loop;
5514                 }
5515
5516 unclustered_alloc:
5517                 spin_lock(&block_group->free_space_ctl->tree_lock);
5518                 if (cached &&
5519                     block_group->free_space_ctl->free_space <
5520                     num_bytes + empty_cluster + empty_size) {
5521                         spin_unlock(&block_group->free_space_ctl->tree_lock);
5522                         goto loop;
5523                 }
5524                 spin_unlock(&block_group->free_space_ctl->tree_lock);
5525
5526                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5527                                                     num_bytes, empty_size);
5528                 /*
5529                  * If we didn't find a chunk, and we haven't failed on this
5530                  * block group before, and this block group is in the middle of
5531                  * caching and we are ok with waiting, then go ahead and wait
5532                  * for progress to be made, and set failed_alloc to true.
5533                  *
5534                  * If failed_alloc is true then we've already waited on this
5535                  * block group once and should move on to the next block group.
5536                  */
5537                 if (!offset && !failed_alloc && !cached &&
5538                     loop > LOOP_CACHING_NOWAIT) {
5539                         wait_block_group_cache_progress(block_group,
5540                                                 num_bytes + empty_size);
5541                         failed_alloc = true;
5542                         goto have_block_group;
5543                 } else if (!offset) {
5544                         if (!cached)
5545                                 have_caching_bg = true;
5546                         goto loop;
5547                 }
5548 checks:
5549                 search_start = stripe_align(root, offset);
5550                 /* move on to the next group */
5551                 if (search_start + num_bytes >= search_end) {
5552                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5553                         goto loop;
5554                 }
5555
5556                 /* move on to the next group */
5557                 if (search_start + num_bytes >
5558                     used_block_group->key.objectid + used_block_group->key.offset) {
5559                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5560                         goto loop;
5561                 }
5562
5563                 if (offset < search_start)
5564                         btrfs_add_free_space(used_block_group, offset,
5565                                              search_start - offset);
5566                 BUG_ON(offset > search_start);
5567
5568                 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5569                                                   alloc_type);
5570                 if (ret == -EAGAIN) {
5571                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5572                         goto loop;
5573                 }
5574
5575                 /* we are all good, lets return */
5576                 ins->objectid = search_start;
5577                 ins->offset = num_bytes;
5578
5579                 if (offset < search_start)
5580                         btrfs_add_free_space(used_block_group, offset,
5581                                              search_start - offset);
5582                 BUG_ON(offset > search_start);
5583                 if (used_block_group != block_group)
5584                         btrfs_put_block_group(used_block_group);
5585                 btrfs_put_block_group(block_group);
5586                 break;
5587 loop:
5588                 failed_cluster_refill = false;
5589                 failed_alloc = false;
5590                 BUG_ON(index != get_block_group_index(block_group));
5591                 if (used_block_group != block_group)
5592                         btrfs_put_block_group(used_block_group);
5593                 btrfs_put_block_group(block_group);
5594         }
5595         up_read(&space_info->groups_sem);
5596
5597         if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5598                 goto search;
5599
5600         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5601                 goto search;
5602
5603         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5604          *                      for them to make caching progress.  Also
5605          *                      determine the best possible bg to cache
5606          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5607          *                      caching kthreads as we move along
5608          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5609          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5610          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5611          *                      again
5612          */
5613         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5614                 index = 0;
5615                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5616                         found_uncached_bg = false;
5617                         loop++;
5618                         if (!ideal_cache_percent)
5619                                 goto search;
5620
5621                         /*
5622                          * 1 of the following 2 things have happened so far
5623                          *
5624                          * 1) We found an ideal block group for caching that
5625                          * is mostly full and will cache quickly, so we might
5626                          * as well wait for it.
5627                          *
5628                          * 2) We searched for cached only and we didn't find
5629                          * anything, and we didn't start any caching kthreads
5630                          * either, so chances are we will loop through and
5631                          * start a couple caching kthreads, and then come back
5632                          * around and just wait for them.  This will be slower
5633                          * because we will have 2 caching kthreads reading at
5634                          * the same time when we could have just started one
5635                          * and waited for it to get far enough to give us an
5636                          * allocation, so go ahead and go to the wait caching
5637                          * loop.
5638                          */
5639                         loop = LOOP_CACHING_WAIT;
5640                         search_start = ideal_cache_offset;
5641                         ideal_cache_percent = 0;
5642                         goto ideal_cache;
5643                 } else if (loop == LOOP_FIND_IDEAL) {
5644                         /*
5645                          * Didn't find a uncached bg, wait on anything we find
5646                          * next.
5647                          */
5648                         loop = LOOP_CACHING_WAIT;
5649                         goto search;
5650                 }
5651
5652                 loop++;
5653
5654                 if (loop == LOOP_ALLOC_CHUNK) {
5655                        if (allowed_chunk_alloc) {
5656                                 ret = do_chunk_alloc(trans, root, num_bytes +
5657                                                      2 * 1024 * 1024, data,
5658                                                      CHUNK_ALLOC_LIMITED);
5659                                 allowed_chunk_alloc = 0;
5660                                 if (ret == 1)
5661                                         done_chunk_alloc = 1;
5662                         } else if (!done_chunk_alloc &&
5663                                    space_info->force_alloc ==
5664                                    CHUNK_ALLOC_NO_FORCE) {
5665                                 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5666                         }
5667
5668                        /*
5669                         * We didn't allocate a chunk, go ahead and drop the
5670                         * empty size and loop again.
5671                         */
5672                        if (!done_chunk_alloc)
5673                                loop = LOOP_NO_EMPTY_SIZE;
5674                 }
5675
5676                 if (loop == LOOP_NO_EMPTY_SIZE) {
5677                         empty_size = 0;
5678                         empty_cluster = 0;
5679                 }
5680
5681                 goto search;
5682         } else if (!ins->objectid) {
5683                 ret = -ENOSPC;
5684         } else if (ins->objectid) {
5685                 ret = 0;
5686         }
5687
5688         return ret;
5689 }
5690
5691 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5692                             int dump_block_groups)
5693 {
5694         struct btrfs_block_group_cache *cache;
5695         int index = 0;
5696
5697         spin_lock(&info->lock);
5698         printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5699                (unsigned long long)info->flags,
5700                (unsigned long long)(info->total_bytes - info->bytes_used -
5701                                     info->bytes_pinned - info->bytes_reserved -
5702                                     info->bytes_readonly),
5703                (info->full) ? "" : "not ");
5704         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5705                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5706                (unsigned long long)info->total_bytes,
5707                (unsigned long long)info->bytes_used,
5708                (unsigned long long)info->bytes_pinned,
5709                (unsigned long long)info->bytes_reserved,
5710                (unsigned long long)info->bytes_may_use,
5711                (unsigned long long)info->bytes_readonly);
5712         spin_unlock(&info->lock);
5713
5714         if (!dump_block_groups)
5715                 return;
5716
5717         down_read(&info->groups_sem);
5718 again:
5719         list_for_each_entry(cache, &info->block_groups[index], list) {
5720                 spin_lock(&cache->lock);
5721                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5722                        "%llu pinned %llu reserved\n",
5723                        (unsigned long long)cache->key.objectid,
5724                        (unsigned long long)cache->key.offset,
5725                        (unsigned long long)btrfs_block_group_used(&cache->item),
5726                        (unsigned long long)cache->pinned,
5727                        (unsigned long long)cache->reserved);
5728                 btrfs_dump_free_space(cache, bytes);
5729                 spin_unlock(&cache->lock);
5730         }
5731         if (++index < BTRFS_NR_RAID_TYPES)
5732                 goto again;
5733         up_read(&info->groups_sem);
5734 }
5735
5736 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5737                          struct btrfs_root *root,
5738                          u64 num_bytes, u64 min_alloc_size,
5739                          u64 empty_size, u64 hint_byte,
5740                          u64 search_end, struct btrfs_key *ins,
5741                          u64 data)
5742 {
5743         int ret;
5744         u64 search_start = 0;
5745
5746         data = btrfs_get_alloc_profile(root, data);
5747 again:
5748         /*
5749          * the only place that sets empty_size is btrfs_realloc_node, which
5750          * is not called recursively on allocations
5751          */
5752         if (empty_size || root->ref_cows)
5753                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5754                                      num_bytes + 2 * 1024 * 1024, data,
5755                                      CHUNK_ALLOC_NO_FORCE);
5756
5757         WARN_ON(num_bytes < root->sectorsize);
5758         ret = find_free_extent(trans, root, num_bytes, empty_size,
5759                                search_start, search_end, hint_byte,
5760                                ins, data);
5761
5762         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5763                 num_bytes = num_bytes >> 1;
5764                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5765                 num_bytes = max(num_bytes, min_alloc_size);
5766                 do_chunk_alloc(trans, root->fs_info->extent_root,
5767                                num_bytes, data, CHUNK_ALLOC_FORCE);
5768                 goto again;
5769         }
5770         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5771                 struct btrfs_space_info *sinfo;
5772
5773                 sinfo = __find_space_info(root->fs_info, data);
5774                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5775                        "wanted %llu\n", (unsigned long long)data,
5776                        (unsigned long long)num_bytes);
5777                 dump_space_info(sinfo, num_bytes, 1);
5778         }
5779
5780         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5781
5782         return ret;
5783 }
5784
5785 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5786                                         u64 start, u64 len, int pin)
5787 {
5788         struct btrfs_block_group_cache *cache;
5789         int ret = 0;
5790
5791         cache = btrfs_lookup_block_group(root->fs_info, start);
5792         if (!cache) {
5793                 printk(KERN_ERR "Unable to find block group for %llu\n",
5794                        (unsigned long long)start);
5795                 return -ENOSPC;
5796         }
5797
5798         if (btrfs_test_opt(root, DISCARD))
5799                 ret = btrfs_discard_extent(root, start, len, NULL);
5800
5801         if (pin)
5802                 pin_down_extent(root, cache, start, len, 1);
5803         else {
5804                 btrfs_add_free_space(cache, start, len);
5805                 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5806         }
5807         btrfs_put_block_group(cache);
5808
5809         trace_btrfs_reserved_extent_free(root, start, len);
5810
5811         return ret;
5812 }
5813
5814 int btrfs_free_reserved_extent(struct btrfs_root *root,
5815                                         u64 start, u64 len)
5816 {
5817         return __btrfs_free_reserved_extent(root, start, len, 0);
5818 }
5819
5820 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5821                                        u64 start, u64 len)
5822 {
5823         return __btrfs_free_reserved_extent(root, start, len, 1);
5824 }
5825
5826 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5827                                       struct btrfs_root *root,
5828                                       u64 parent, u64 root_objectid,
5829                                       u64 flags, u64 owner, u64 offset,
5830                                       struct btrfs_key *ins, int ref_mod)
5831 {
5832         int ret;
5833         struct btrfs_fs_info *fs_info = root->fs_info;
5834         struct btrfs_extent_item *extent_item;
5835         struct btrfs_extent_inline_ref *iref;
5836         struct btrfs_path *path;
5837         struct extent_buffer *leaf;
5838         int type;
5839         u32 size;
5840
5841         if (parent > 0)
5842                 type = BTRFS_SHARED_DATA_REF_KEY;
5843         else
5844                 type = BTRFS_EXTENT_DATA_REF_KEY;
5845
5846         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5847
5848         path = btrfs_alloc_path();
5849         if (!path)
5850                 return -ENOMEM;
5851
5852         path->leave_spinning = 1;
5853         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5854                                       ins, size);
5855         BUG_ON(ret);
5856
5857         leaf = path->nodes[0];
5858         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5859                                      struct btrfs_extent_item);
5860         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5861         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5862         btrfs_set_extent_flags(leaf, extent_item,
5863                                flags | BTRFS_EXTENT_FLAG_DATA);
5864
5865         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5866         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5867         if (parent > 0) {
5868                 struct btrfs_shared_data_ref *ref;
5869                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5870                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5871                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5872         } else {
5873                 struct btrfs_extent_data_ref *ref;
5874                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5875                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5876                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5877                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5878                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5879         }
5880
5881         btrfs_mark_buffer_dirty(path->nodes[0]);
5882         btrfs_free_path(path);
5883
5884         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5885         if (ret) {
5886                 printk(KERN_ERR "btrfs update block group failed for %llu "
5887                        "%llu\n", (unsigned long long)ins->objectid,
5888                        (unsigned long long)ins->offset);
5889                 BUG();
5890         }
5891         return ret;
5892 }
5893
5894 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5895                                      struct btrfs_root *root,
5896                                      u64 parent, u64 root_objectid,
5897                                      u64 flags, struct btrfs_disk_key *key,
5898                                      int level, struct btrfs_key *ins)
5899 {
5900         int ret;
5901         struct btrfs_fs_info *fs_info = root->fs_info;
5902         struct btrfs_extent_item *extent_item;
5903         struct btrfs_tree_block_info *block_info;
5904         struct btrfs_extent_inline_ref *iref;
5905         struct btrfs_path *path;
5906         struct extent_buffer *leaf;
5907         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5908
5909         path = btrfs_alloc_path();
5910         if (!path)
5911                 return -ENOMEM;
5912
5913         path->leave_spinning = 1;
5914         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5915                                       ins, size);
5916         BUG_ON(ret);
5917
5918         leaf = path->nodes[0];
5919         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5920                                      struct btrfs_extent_item);
5921         btrfs_set_extent_refs(leaf, extent_item, 1);
5922         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5923         btrfs_set_extent_flags(leaf, extent_item,
5924                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5925         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5926
5927         btrfs_set_tree_block_key(leaf, block_info, key);
5928         btrfs_set_tree_block_level(leaf, block_info, level);
5929
5930         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5931         if (parent > 0) {
5932                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5933                 btrfs_set_extent_inline_ref_type(leaf, iref,
5934                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5935                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5936         } else {
5937                 btrfs_set_extent_inline_ref_type(leaf, iref,
5938                                                  BTRFS_TREE_BLOCK_REF_KEY);
5939                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5940         }
5941
5942         btrfs_mark_buffer_dirty(leaf);
5943         btrfs_free_path(path);
5944
5945         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5946         if (ret) {
5947                 printk(KERN_ERR "btrfs update block group failed for %llu "
5948                        "%llu\n", (unsigned long long)ins->objectid,
5949                        (unsigned long long)ins->offset);
5950                 BUG();
5951         }
5952         return ret;
5953 }
5954
5955 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5956                                      struct btrfs_root *root,
5957                                      u64 root_objectid, u64 owner,
5958                                      u64 offset, struct btrfs_key *ins)
5959 {
5960         int ret;
5961
5962         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5963
5964         ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
5965                                          ins->offset, 0,
5966                                          root_objectid, owner, offset,
5967                                          BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
5968         return ret;
5969 }
5970
5971 /*
5972  * this is used by the tree logging recovery code.  It records that
5973  * an extent has been allocated and makes sure to clear the free
5974  * space cache bits as well
5975  */
5976 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5977                                    struct btrfs_root *root,
5978                                    u64 root_objectid, u64 owner, u64 offset,
5979                                    struct btrfs_key *ins)
5980 {
5981         int ret;
5982         struct btrfs_block_group_cache *block_group;
5983         struct btrfs_caching_control *caching_ctl;
5984         u64 start = ins->objectid;
5985         u64 num_bytes = ins->offset;
5986
5987         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5988         cache_block_group(block_group, trans, NULL, 0);
5989         caching_ctl = get_caching_control(block_group);
5990
5991         if (!caching_ctl) {
5992                 BUG_ON(!block_group_cache_done(block_group));
5993                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5994                 BUG_ON(ret);
5995         } else {
5996                 mutex_lock(&caching_ctl->mutex);
5997
5998                 if (start >= caching_ctl->progress) {
5999                         ret = add_excluded_extent(root, start, num_bytes);
6000                         BUG_ON(ret);
6001                 } else if (start + num_bytes <= caching_ctl->progress) {
6002                         ret = btrfs_remove_free_space(block_group,
6003                                                       start, num_bytes);
6004                         BUG_ON(ret);
6005                 } else {
6006                         num_bytes = caching_ctl->progress - start;
6007                         ret = btrfs_remove_free_space(block_group,
6008                                                       start, num_bytes);
6009                         BUG_ON(ret);
6010
6011                         start = caching_ctl->progress;
6012                         num_bytes = ins->objectid + ins->offset -
6013                                     caching_ctl->progress;
6014                         ret = add_excluded_extent(root, start, num_bytes);
6015                         BUG_ON(ret);
6016                 }
6017
6018                 mutex_unlock(&caching_ctl->mutex);
6019                 put_caching_control(caching_ctl);
6020         }
6021
6022         ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6023                                           RESERVE_ALLOC_NO_ACCOUNT);
6024         BUG_ON(ret);
6025         btrfs_put_block_group(block_group);
6026         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6027                                          0, owner, offset, ins, 1);
6028         return ret;
6029 }
6030
6031 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6032                                             struct btrfs_root *root,
6033                                             u64 bytenr, u32 blocksize,
6034                                             int level)
6035 {
6036         struct extent_buffer *buf;
6037
6038         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6039         if (!buf)
6040                 return ERR_PTR(-ENOMEM);
6041         btrfs_set_header_generation(buf, trans->transid);
6042         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6043         btrfs_tree_lock(buf);
6044         clean_tree_block(trans, root, buf);
6045
6046         btrfs_set_lock_blocking(buf);
6047         btrfs_set_buffer_uptodate(buf);
6048
6049         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6050                 /*
6051                  * we allow two log transactions at a time, use different
6052                  * EXENT bit to differentiate dirty pages.
6053                  */
6054                 if (root->log_transid % 2 == 0)
6055                         set_extent_dirty(&root->dirty_log_pages, buf->start,
6056                                         buf->start + buf->len - 1, GFP_NOFS);
6057                 else
6058                         set_extent_new(&root->dirty_log_pages, buf->start,
6059                                         buf->start + buf->len - 1, GFP_NOFS);
6060         } else {
6061                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6062                          buf->start + buf->len - 1, GFP_NOFS);
6063         }
6064         trans->blocks_used++;
6065         /* this returns a buffer locked for blocking */
6066         return buf;
6067 }
6068
6069 static struct btrfs_block_rsv *
6070 use_block_rsv(struct btrfs_trans_handle *trans,
6071               struct btrfs_root *root, u32 blocksize)
6072 {
6073         struct btrfs_block_rsv *block_rsv;
6074         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6075         int ret;
6076
6077         block_rsv = get_block_rsv(trans, root);
6078
6079         if (block_rsv->size == 0) {
6080                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6081                 /*
6082                  * If we couldn't reserve metadata bytes try and use some from
6083                  * the global reserve.
6084                  */
6085                 if (ret && block_rsv != global_rsv) {
6086                         ret = block_rsv_use_bytes(global_rsv, blocksize);
6087                         if (!ret)
6088                                 return global_rsv;
6089                         return ERR_PTR(ret);
6090                 } else if (ret) {
6091                         return ERR_PTR(ret);
6092                 }
6093                 return block_rsv;
6094         }
6095
6096         ret = block_rsv_use_bytes(block_rsv, blocksize);
6097         if (!ret)
6098                 return block_rsv;
6099         if (ret) {
6100                 static DEFINE_RATELIMIT_STATE(_rs,
6101                                 DEFAULT_RATELIMIT_INTERVAL,
6102                                 /*DEFAULT_RATELIMIT_BURST*/ 2);
6103                 if (__ratelimit(&_rs)) {
6104                         printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6105                         WARN_ON(1);
6106                 }
6107                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6108                 if (!ret) {
6109                         return block_rsv;
6110                 } else if (ret && block_rsv != global_rsv) {
6111                         ret = block_rsv_use_bytes(global_rsv, blocksize);
6112                         if (!ret)
6113                                 return global_rsv;
6114                 }
6115         }
6116
6117         return ERR_PTR(-ENOSPC);
6118 }
6119
6120 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
6121 {
6122         block_rsv_add_bytes(block_rsv, blocksize, 0);
6123         block_rsv_release_bytes(block_rsv, NULL, 0);
6124 }
6125
6126 /*
6127  * finds a free extent and does all the dirty work required for allocation
6128  * returns the key for the extent through ins, and a tree buffer for
6129  * the first block of the extent through buf.
6130  *
6131  * returns the tree buffer or NULL.
6132  */
6133 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6134                                         struct btrfs_root *root, u32 blocksize,
6135                                         u64 parent, u64 root_objectid,
6136                                         struct btrfs_disk_key *key, int level,
6137                                         u64 hint, u64 empty_size, int for_cow)
6138 {
6139         struct btrfs_key ins;
6140         struct btrfs_block_rsv *block_rsv;
6141         struct extent_buffer *buf;
6142         u64 flags = 0;
6143         int ret;
6144
6145
6146         block_rsv = use_block_rsv(trans, root, blocksize);
6147         if (IS_ERR(block_rsv))
6148                 return ERR_CAST(block_rsv);
6149
6150         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6151                                    empty_size, hint, (u64)-1, &ins, 0);
6152         if (ret) {
6153                 unuse_block_rsv(block_rsv, blocksize);
6154                 return ERR_PTR(ret);
6155         }
6156
6157         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6158                                     blocksize, level);
6159         BUG_ON(IS_ERR(buf));
6160
6161         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6162                 if (parent == 0)
6163                         parent = ins.objectid;
6164                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6165         } else
6166                 BUG_ON(parent > 0);
6167
6168         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6169                 struct btrfs_delayed_extent_op *extent_op;
6170                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6171                 BUG_ON(!extent_op);
6172                 if (key)
6173                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
6174                 else
6175                         memset(&extent_op->key, 0, sizeof(extent_op->key));
6176                 extent_op->flags_to_set = flags;
6177                 extent_op->update_key = 1;
6178                 extent_op->update_flags = 1;
6179                 extent_op->is_data = 0;
6180
6181                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6182                                         ins.objectid,
6183                                         ins.offset, parent, root_objectid,
6184                                         level, BTRFS_ADD_DELAYED_EXTENT,
6185                                         extent_op, for_cow);
6186                 BUG_ON(ret);
6187         }
6188         return buf;
6189 }
6190
6191 struct walk_control {
6192         u64 refs[BTRFS_MAX_LEVEL];
6193         u64 flags[BTRFS_MAX_LEVEL];
6194         struct btrfs_key update_progress;
6195         int stage;
6196         int level;
6197         int shared_level;
6198         int update_ref;
6199         int keep_locks;
6200         int reada_slot;
6201         int reada_count;
6202         int for_reloc;
6203 };
6204
6205 #define DROP_REFERENCE  1
6206 #define UPDATE_BACKREF  2
6207
6208 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6209                                      struct btrfs_root *root,
6210                                      struct walk_control *wc,
6211                                      struct btrfs_path *path)
6212 {
6213         u64 bytenr;
6214         u64 generation;
6215         u64 refs;
6216         u64 flags;
6217         u32 nritems;
6218         u32 blocksize;
6219         struct btrfs_key key;
6220         struct extent_buffer *eb;
6221         int ret;
6222         int slot;
6223         int nread = 0;
6224
6225         if (path->slots[wc->level] < wc->reada_slot) {
6226                 wc->reada_count = wc->reada_count * 2 / 3;
6227                 wc->reada_count = max(wc->reada_count, 2);
6228         } else {
6229                 wc->reada_count = wc->reada_count * 3 / 2;
6230                 wc->reada_count = min_t(int, wc->reada_count,
6231                                         BTRFS_NODEPTRS_PER_BLOCK(root));
6232         }
6233
6234         eb = path->nodes[wc->level];
6235         nritems = btrfs_header_nritems(eb);
6236         blocksize = btrfs_level_size(root, wc->level - 1);
6237
6238         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6239                 if (nread >= wc->reada_count)
6240                         break;
6241
6242                 cond_resched();
6243                 bytenr = btrfs_node_blockptr(eb, slot);
6244                 generation = btrfs_node_ptr_generation(eb, slot);
6245
6246                 if (slot == path->slots[wc->level])
6247                         goto reada;
6248
6249                 if (wc->stage == UPDATE_BACKREF &&
6250                     generation <= root->root_key.offset)
6251                         continue;
6252
6253                 /* We don't lock the tree block, it's OK to be racy here */
6254                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6255                                                &refs, &flags);
6256                 BUG_ON(ret);
6257                 BUG_ON(refs == 0);
6258
6259                 if (wc->stage == DROP_REFERENCE) {
6260                         if (refs == 1)
6261                                 goto reada;
6262
6263                         if (wc->level == 1 &&
6264                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6265                                 continue;
6266                         if (!wc->update_ref ||
6267                             generation <= root->root_key.offset)
6268                                 continue;
6269                         btrfs_node_key_to_cpu(eb, &key, slot);
6270                         ret = btrfs_comp_cpu_keys(&key,
6271                                                   &wc->update_progress);
6272                         if (ret < 0)
6273                                 continue;
6274                 } else {
6275                         if (wc->level == 1 &&
6276                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6277                                 continue;
6278                 }
6279 reada:
6280                 ret = readahead_tree_block(root, bytenr, blocksize,
6281                                            generation);
6282                 if (ret)
6283                         break;
6284                 nread++;
6285         }
6286         wc->reada_slot = slot;
6287 }
6288
6289 /*
6290  * hepler to process tree block while walking down the tree.
6291  *
6292  * when wc->stage == UPDATE_BACKREF, this function updates
6293  * back refs for pointers in the block.
6294  *
6295  * NOTE: return value 1 means we should stop walking down.
6296  */
6297 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6298                                    struct btrfs_root *root,
6299                                    struct btrfs_path *path,
6300                                    struct walk_control *wc, int lookup_info)
6301 {
6302         int level = wc->level;
6303         struct extent_buffer *eb = path->nodes[level];
6304         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6305         int ret;
6306
6307         if (wc->stage == UPDATE_BACKREF &&
6308             btrfs_header_owner(eb) != root->root_key.objectid)
6309                 return 1;
6310
6311         /*
6312          * when reference count of tree block is 1, it won't increase
6313          * again. once full backref flag is set, we never clear it.
6314          */
6315         if (lookup_info &&
6316             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6317              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6318                 BUG_ON(!path->locks[level]);
6319                 ret = btrfs_lookup_extent_info(trans, root,
6320                                                eb->start, eb->len,
6321                                                &wc->refs[level],
6322                                                &wc->flags[level]);
6323                 BUG_ON(ret);
6324                 BUG_ON(wc->refs[level] == 0);
6325         }
6326
6327         if (wc->stage == DROP_REFERENCE) {
6328                 if (wc->refs[level] > 1)
6329                         return 1;
6330
6331                 if (path->locks[level] && !wc->keep_locks) {
6332                         btrfs_tree_unlock_rw(eb, path->locks[level]);
6333                         path->locks[level] = 0;
6334                 }
6335                 return 0;
6336         }
6337
6338         /* wc->stage == UPDATE_BACKREF */
6339         if (!(wc->flags[level] & flag)) {
6340                 BUG_ON(!path->locks[level]);
6341                 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6342                 BUG_ON(ret);
6343                 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6344                 BUG_ON(ret);
6345                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6346                                                   eb->len, flag, 0);
6347                 BUG_ON(ret);
6348                 wc->flags[level] |= flag;
6349         }
6350
6351         /*
6352          * the block is shared by multiple trees, so it's not good to
6353          * keep the tree lock
6354          */
6355         if (path->locks[level] && level > 0) {
6356                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6357                 path->locks[level] = 0;
6358         }
6359         return 0;
6360 }
6361
6362 /*
6363  * hepler to process tree block pointer.
6364  *
6365  * when wc->stage == DROP_REFERENCE, this function checks
6366  * reference count of the block pointed to. if the block
6367  * is shared and we need update back refs for the subtree
6368  * rooted at the block, this function changes wc->stage to
6369  * UPDATE_BACKREF. if the block is shared and there is no
6370  * need to update back, this function drops the reference
6371  * to the block.
6372  *
6373  * NOTE: return value 1 means we should stop walking down.
6374  */
6375 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6376                                  struct btrfs_root *root,
6377                                  struct btrfs_path *path,
6378                                  struct walk_control *wc, int *lookup_info)
6379 {
6380         u64 bytenr;
6381         u64 generation;
6382         u64 parent;
6383         u32 blocksize;
6384         struct btrfs_key key;
6385         struct extent_buffer *next;
6386         int level = wc->level;
6387         int reada = 0;
6388         int ret = 0;
6389
6390         generation = btrfs_node_ptr_generation(path->nodes[level],
6391                                                path->slots[level]);
6392         /*
6393          * if the lower level block was created before the snapshot
6394          * was created, we know there is no need to update back refs
6395          * for the subtree
6396          */
6397         if (wc->stage == UPDATE_BACKREF &&
6398             generation <= root->root_key.offset) {
6399                 *lookup_info = 1;
6400                 return 1;
6401         }
6402
6403         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6404         blocksize = btrfs_level_size(root, level - 1);
6405
6406         next = btrfs_find_tree_block(root, bytenr, blocksize);
6407         if (!next) {
6408                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6409                 if (!next)
6410                         return -ENOMEM;
6411                 reada = 1;
6412         }
6413         btrfs_tree_lock(next);
6414         btrfs_set_lock_blocking(next);
6415
6416         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6417                                        &wc->refs[level - 1],
6418                                        &wc->flags[level - 1]);
6419         BUG_ON(ret);
6420         BUG_ON(wc->refs[level - 1] == 0);
6421         *lookup_info = 0;
6422
6423         if (wc->stage == DROP_REFERENCE) {
6424                 if (wc->refs[level - 1] > 1) {
6425                         if (level == 1 &&
6426                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6427                                 goto skip;
6428
6429                         if (!wc->update_ref ||
6430                             generation <= root->root_key.offset)
6431                                 goto skip;
6432
6433                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6434                                               path->slots[level]);
6435                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6436                         if (ret < 0)
6437                                 goto skip;
6438
6439                         wc->stage = UPDATE_BACKREF;
6440                         wc->shared_level = level - 1;
6441                 }
6442         } else {
6443                 if (level == 1 &&
6444                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6445                         goto skip;
6446         }
6447
6448         if (!btrfs_buffer_uptodate(next, generation)) {
6449                 btrfs_tree_unlock(next);
6450                 free_extent_buffer(next);
6451                 next = NULL;
6452                 *lookup_info = 1;
6453         }
6454
6455         if (!next) {
6456                 if (reada && level == 1)
6457                         reada_walk_down(trans, root, wc, path);
6458                 next = read_tree_block(root, bytenr, blocksize, generation);
6459                 if (!next)
6460                         return -EIO;
6461                 btrfs_tree_lock(next);
6462                 btrfs_set_lock_blocking(next);
6463         }
6464
6465         level--;
6466         BUG_ON(level != btrfs_header_level(next));
6467         path->nodes[level] = next;
6468         path->slots[level] = 0;
6469         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6470         wc->level = level;
6471         if (wc->level == 1)
6472                 wc->reada_slot = 0;
6473         return 0;
6474 skip:
6475         wc->refs[level - 1] = 0;
6476         wc->flags[level - 1] = 0;
6477         if (wc->stage == DROP_REFERENCE) {
6478                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6479                         parent = path->nodes[level]->start;
6480                 } else {
6481                         BUG_ON(root->root_key.objectid !=
6482                                btrfs_header_owner(path->nodes[level]));
6483                         parent = 0;
6484                 }
6485
6486                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6487                                 root->root_key.objectid, level - 1, 0, 0);
6488                 BUG_ON(ret);
6489         }
6490         btrfs_tree_unlock(next);
6491         free_extent_buffer(next);
6492         *lookup_info = 1;
6493         return 1;
6494 }
6495
6496 /*
6497  * hepler to process tree block while walking up the tree.
6498  *
6499  * when wc->stage == DROP_REFERENCE, this function drops
6500  * reference count on the block.
6501  *
6502  * when wc->stage == UPDATE_BACKREF, this function changes
6503  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6504  * to UPDATE_BACKREF previously while processing the block.
6505  *
6506  * NOTE: return value 1 means we should stop walking up.
6507  */
6508 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6509                                  struct btrfs_root *root,
6510                                  struct btrfs_path *path,
6511                                  struct walk_control *wc)
6512 {
6513         int ret;
6514         int level = wc->level;
6515         struct extent_buffer *eb = path->nodes[level];
6516         u64 parent = 0;
6517
6518         if (wc->stage == UPDATE_BACKREF) {
6519                 BUG_ON(wc->shared_level < level);
6520                 if (level < wc->shared_level)
6521                         goto out;
6522
6523                 ret = find_next_key(path, level + 1, &wc->update_progress);
6524                 if (ret > 0)
6525                         wc->update_ref = 0;
6526
6527                 wc->stage = DROP_REFERENCE;
6528                 wc->shared_level = -1;
6529                 path->slots[level] = 0;
6530
6531                 /*
6532                  * check reference count again if the block isn't locked.
6533                  * we should start walking down the tree again if reference
6534                  * count is one.
6535                  */
6536                 if (!path->locks[level]) {
6537                         BUG_ON(level == 0);
6538                         btrfs_tree_lock(eb);
6539                         btrfs_set_lock_blocking(eb);
6540                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6541
6542                         ret = btrfs_lookup_extent_info(trans, root,
6543                                                        eb->start, eb->len,
6544                                                        &wc->refs[level],
6545                                                        &wc->flags[level]);
6546                         BUG_ON(ret);
6547                         BUG_ON(wc->refs[level] == 0);
6548                         if (wc->refs[level] == 1) {
6549                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6550                                 return 1;
6551                         }
6552                 }
6553         }
6554
6555         /* wc->stage == DROP_REFERENCE */
6556         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6557
6558         if (wc->refs[level] == 1) {
6559                 if (level == 0) {
6560                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6561                                 ret = btrfs_dec_ref(trans, root, eb, 1,
6562                                                     wc->for_reloc);
6563                         else
6564                                 ret = btrfs_dec_ref(trans, root, eb, 0,
6565                                                     wc->for_reloc);
6566                         BUG_ON(ret);
6567                 }
6568                 /* make block locked assertion in clean_tree_block happy */
6569                 if (!path->locks[level] &&
6570                     btrfs_header_generation(eb) == trans->transid) {
6571                         btrfs_tree_lock(eb);
6572                         btrfs_set_lock_blocking(eb);
6573                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6574                 }
6575                 clean_tree_block(trans, root, eb);
6576         }
6577
6578         if (eb == root->node) {
6579                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6580                         parent = eb->start;
6581                 else
6582                         BUG_ON(root->root_key.objectid !=
6583                                btrfs_header_owner(eb));
6584         } else {
6585                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6586                         parent = path->nodes[level + 1]->start;
6587                 else
6588                         BUG_ON(root->root_key.objectid !=
6589                                btrfs_header_owner(path->nodes[level + 1]));
6590         }
6591
6592         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0);
6593 out:
6594         wc->refs[level] = 0;
6595         wc->flags[level] = 0;
6596         return 0;
6597 }
6598
6599 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6600                                    struct btrfs_root *root,
6601                                    struct btrfs_path *path,
6602                                    struct walk_control *wc)
6603 {
6604         int level = wc->level;
6605         int lookup_info = 1;
6606         int ret;
6607
6608         while (level >= 0) {
6609                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6610                 if (ret > 0)
6611                         break;
6612
6613                 if (level == 0)
6614                         break;
6615
6616                 if (path->slots[level] >=
6617                     btrfs_header_nritems(path->nodes[level]))
6618                         break;
6619
6620                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6621                 if (ret > 0) {
6622                         path->slots[level]++;
6623                         continue;
6624                 } else if (ret < 0)
6625                         return ret;
6626                 level = wc->level;
6627         }
6628         return 0;
6629 }
6630
6631 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6632                                  struct btrfs_root *root,
6633                                  struct btrfs_path *path,
6634                                  struct walk_control *wc, int max_level)
6635 {
6636         int level = wc->level;
6637         int ret;
6638
6639         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6640         while (level < max_level && path->nodes[level]) {
6641                 wc->level = level;
6642                 if (path->slots[level] + 1 <
6643                     btrfs_header_nritems(path->nodes[level])) {
6644                         path->slots[level]++;
6645                         return 0;
6646                 } else {
6647                         ret = walk_up_proc(trans, root, path, wc);
6648                         if (ret > 0)
6649                                 return 0;
6650
6651                         if (path->locks[level]) {
6652                                 btrfs_tree_unlock_rw(path->nodes[level],
6653                                                      path->locks[level]);
6654                                 path->locks[level] = 0;
6655                         }
6656                         free_extent_buffer(path->nodes[level]);
6657                         path->nodes[level] = NULL;
6658                         level++;
6659                 }
6660         }
6661         return 1;
6662 }
6663
6664 /*
6665  * drop a subvolume tree.
6666  *
6667  * this function traverses the tree freeing any blocks that only
6668  * referenced by the tree.
6669  *
6670  * when a shared tree block is found. this function decreases its
6671  * reference count by one. if update_ref is true, this function
6672  * also make sure backrefs for the shared block and all lower level
6673  * blocks are properly updated.
6674  */
6675 void btrfs_drop_snapshot(struct btrfs_root *root,
6676                          struct btrfs_block_rsv *block_rsv, int update_ref,
6677                          int for_reloc)
6678 {
6679         struct btrfs_path *path;
6680         struct btrfs_trans_handle *trans;
6681         struct btrfs_root *tree_root = root->fs_info->tree_root;
6682         struct btrfs_root_item *root_item = &root->root_item;
6683         struct walk_control *wc;
6684         struct btrfs_key key;
6685         int err = 0;
6686         int ret;
6687         int level;
6688
6689         path = btrfs_alloc_path();
6690         if (!path) {
6691                 err = -ENOMEM;
6692                 goto out;
6693         }
6694
6695         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6696         if (!wc) {
6697                 btrfs_free_path(path);
6698                 err = -ENOMEM;
6699                 goto out;
6700         }
6701
6702         trans = btrfs_start_transaction(tree_root, 0);
6703         BUG_ON(IS_ERR(trans));
6704
6705         if (block_rsv)
6706                 trans->block_rsv = block_rsv;
6707
6708         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6709                 level = btrfs_header_level(root->node);
6710                 path->nodes[level] = btrfs_lock_root_node(root);
6711                 btrfs_set_lock_blocking(path->nodes[level]);
6712                 path->slots[level] = 0;
6713                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6714                 memset(&wc->update_progress, 0,
6715                        sizeof(wc->update_progress));
6716         } else {
6717                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6718                 memcpy(&wc->update_progress, &key,
6719                        sizeof(wc->update_progress));
6720
6721                 level = root_item->drop_level;
6722                 BUG_ON(level == 0);
6723                 path->lowest_level = level;
6724                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6725                 path->lowest_level = 0;
6726                 if (ret < 0) {
6727                         err = ret;
6728                         goto out_free;
6729                 }
6730                 WARN_ON(ret > 0);
6731
6732                 /*
6733                  * unlock our path, this is safe because only this
6734                  * function is allowed to delete this snapshot
6735                  */
6736                 btrfs_unlock_up_safe(path, 0);
6737
6738                 level = btrfs_header_level(root->node);
6739                 while (1) {
6740                         btrfs_tree_lock(path->nodes[level]);
6741                         btrfs_set_lock_blocking(path->nodes[level]);
6742
6743                         ret = btrfs_lookup_extent_info(trans, root,
6744                                                 path->nodes[level]->start,
6745                                                 path->nodes[level]->len,
6746                                                 &wc->refs[level],
6747                                                 &wc->flags[level]);
6748                         BUG_ON(ret);
6749                         BUG_ON(wc->refs[level] == 0);
6750
6751                         if (level == root_item->drop_level)
6752                                 break;
6753
6754                         btrfs_tree_unlock(path->nodes[level]);
6755                         WARN_ON(wc->refs[level] != 1);
6756                         level--;
6757                 }
6758         }
6759
6760         wc->level = level;
6761         wc->shared_level = -1;
6762         wc->stage = DROP_REFERENCE;
6763         wc->update_ref = update_ref;
6764         wc->keep_locks = 0;
6765         wc->for_reloc = for_reloc;
6766         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6767
6768         while (1) {
6769                 ret = walk_down_tree(trans, root, path, wc);
6770                 if (ret < 0) {
6771                         err = ret;
6772                         break;
6773                 }
6774
6775                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6776                 if (ret < 0) {
6777                         err = ret;
6778                         break;
6779                 }
6780
6781                 if (ret > 0) {
6782                         BUG_ON(wc->stage != DROP_REFERENCE);
6783                         break;
6784                 }
6785
6786                 if (wc->stage == DROP_REFERENCE) {
6787                         level = wc->level;
6788                         btrfs_node_key(path->nodes[level],
6789                                        &root_item->drop_progress,
6790                                        path->slots[level]);
6791                         root_item->drop_level = level;
6792                 }
6793
6794                 BUG_ON(wc->level == 0);
6795                 if (btrfs_should_end_transaction(trans, tree_root)) {
6796                         ret = btrfs_update_root(trans, tree_root,
6797                                                 &root->root_key,
6798                                                 root_item);
6799                         BUG_ON(ret);
6800
6801                         btrfs_end_transaction_throttle(trans, tree_root);
6802                         trans = btrfs_start_transaction(tree_root, 0);
6803                         BUG_ON(IS_ERR(trans));
6804                         if (block_rsv)
6805                                 trans->block_rsv = block_rsv;
6806                 }
6807         }
6808         btrfs_release_path(path);
6809         BUG_ON(err);
6810
6811         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6812         BUG_ON(ret);
6813
6814         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6815                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6816                                            NULL, NULL);
6817                 BUG_ON(ret < 0);
6818                 if (ret > 0) {
6819                         /* if we fail to delete the orphan item this time
6820                          * around, it'll get picked up the next time.
6821                          *
6822                          * The most common failure here is just -ENOENT.
6823                          */
6824                         btrfs_del_orphan_item(trans, tree_root,
6825                                               root->root_key.objectid);
6826                 }
6827         }
6828
6829         if (root->in_radix) {
6830                 btrfs_free_fs_root(tree_root->fs_info, root);
6831         } else {
6832                 free_extent_buffer(root->node);
6833                 free_extent_buffer(root->commit_root);
6834                 kfree(root);
6835         }
6836 out_free:
6837         btrfs_end_transaction_throttle(trans, tree_root);
6838         kfree(wc);
6839         btrfs_free_path(path);
6840 out:
6841         if (err)
6842                 btrfs_std_error(root->fs_info, err);
6843         return;
6844 }
6845
6846 /*
6847  * drop subtree rooted at tree block 'node'.
6848  *
6849  * NOTE: this function will unlock and release tree block 'node'
6850  * only used by relocation code
6851  */
6852 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6853                         struct btrfs_root *root,
6854                         struct extent_buffer *node,
6855                         struct extent_buffer *parent)
6856 {
6857         struct btrfs_path *path;
6858         struct walk_control *wc;
6859         int level;
6860         int parent_level;
6861         int ret = 0;
6862         int wret;
6863
6864         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6865
6866         path = btrfs_alloc_path();
6867         if (!path)
6868                 return -ENOMEM;
6869
6870         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6871         if (!wc) {
6872                 btrfs_free_path(path);
6873                 return -ENOMEM;
6874         }
6875
6876         btrfs_assert_tree_locked(parent);
6877         parent_level = btrfs_header_level(parent);
6878         extent_buffer_get(parent);
6879         path->nodes[parent_level] = parent;
6880         path->slots[parent_level] = btrfs_header_nritems(parent);
6881
6882         btrfs_assert_tree_locked(node);
6883         level = btrfs_header_level(node);
6884         path->nodes[level] = node;
6885         path->slots[level] = 0;
6886         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6887
6888         wc->refs[parent_level] = 1;
6889         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6890         wc->level = level;
6891         wc->shared_level = -1;
6892         wc->stage = DROP_REFERENCE;
6893         wc->update_ref = 0;
6894         wc->keep_locks = 1;
6895         wc->for_reloc = 1;
6896         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6897
6898         while (1) {
6899                 wret = walk_down_tree(trans, root, path, wc);
6900                 if (wret < 0) {
6901                         ret = wret;
6902                         break;
6903                 }
6904
6905                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6906                 if (wret < 0)
6907                         ret = wret;
6908                 if (wret != 0)
6909                         break;
6910         }
6911
6912         kfree(wc);
6913         btrfs_free_path(path);
6914         return ret;
6915 }
6916
6917 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6918 {
6919         u64 num_devices;
6920         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6921                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6922
6923         if (root->fs_info->balance_ctl) {
6924                 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
6925                 u64 tgt = 0;
6926
6927                 /* pick restriper's target profile and return */
6928                 if (flags & BTRFS_BLOCK_GROUP_DATA &&
6929                     bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6930                         tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
6931                 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
6932                            bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6933                         tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
6934                 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
6935                            bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6936                         tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
6937                 }
6938
6939                 if (tgt) {
6940                         /* extended -> chunk profile */
6941                         tgt &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
6942                         return tgt;
6943                 }
6944         }
6945
6946         /*
6947          * we add in the count of missing devices because we want
6948          * to make sure that any RAID levels on a degraded FS
6949          * continue to be honored.
6950          */
6951         num_devices = root->fs_info->fs_devices->rw_devices +
6952                 root->fs_info->fs_devices->missing_devices;
6953
6954         if (num_devices == 1) {
6955                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6956                 stripped = flags & ~stripped;
6957
6958                 /* turn raid0 into single device chunks */
6959                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6960                         return stripped;
6961
6962                 /* turn mirroring into duplication */
6963                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6964                              BTRFS_BLOCK_GROUP_RAID10))
6965                         return stripped | BTRFS_BLOCK_GROUP_DUP;
6966                 return flags;
6967         } else {
6968                 /* they already had raid on here, just return */
6969                 if (flags & stripped)
6970                         return flags;
6971
6972                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6973                 stripped = flags & ~stripped;
6974
6975                 /* switch duplicated blocks with raid1 */
6976                 if (flags & BTRFS_BLOCK_GROUP_DUP)
6977                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
6978
6979                 /* turn single device chunks into raid0 */
6980                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6981         }
6982         return flags;
6983 }
6984
6985 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6986 {
6987         struct btrfs_space_info *sinfo = cache->space_info;
6988         u64 num_bytes;
6989         u64 min_allocable_bytes;
6990         int ret = -ENOSPC;
6991
6992
6993         /*
6994          * We need some metadata space and system metadata space for
6995          * allocating chunks in some corner cases until we force to set
6996          * it to be readonly.
6997          */
6998         if ((sinfo->flags &
6999              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7000             !force)
7001                 min_allocable_bytes = 1 * 1024 * 1024;
7002         else
7003                 min_allocable_bytes = 0;
7004
7005         spin_lock(&sinfo->lock);
7006         spin_lock(&cache->lock);
7007
7008         if (cache->ro) {
7009                 ret = 0;
7010                 goto out;
7011         }
7012
7013         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7014                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7015
7016         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7017             sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7018             min_allocable_bytes <= sinfo->total_bytes) {
7019                 sinfo->bytes_readonly += num_bytes;
7020                 cache->ro = 1;
7021                 ret = 0;
7022         }
7023 out:
7024         spin_unlock(&cache->lock);
7025         spin_unlock(&sinfo->lock);
7026         return ret;
7027 }
7028
7029 int btrfs_set_block_group_ro(struct btrfs_root *root,
7030                              struct btrfs_block_group_cache *cache)
7031
7032 {
7033         struct btrfs_trans_handle *trans;
7034         u64 alloc_flags;
7035         int ret;
7036
7037         BUG_ON(cache->ro);
7038
7039         trans = btrfs_join_transaction(root);
7040         BUG_ON(IS_ERR(trans));
7041
7042         alloc_flags = update_block_group_flags(root, cache->flags);
7043         if (alloc_flags != cache->flags)
7044                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7045                                CHUNK_ALLOC_FORCE);
7046
7047         ret = set_block_group_ro(cache, 0);
7048         if (!ret)
7049                 goto out;
7050         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7051         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7052                              CHUNK_ALLOC_FORCE);
7053         if (ret < 0)
7054                 goto out;
7055         ret = set_block_group_ro(cache, 0);
7056 out:
7057         btrfs_end_transaction(trans, root);
7058         return ret;
7059 }
7060
7061 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7062                             struct btrfs_root *root, u64 type)
7063 {
7064         u64 alloc_flags = get_alloc_profile(root, type);
7065         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7066                               CHUNK_ALLOC_FORCE);
7067 }
7068
7069 /*
7070  * helper to account the unused space of all the readonly block group in the
7071  * list. takes mirrors into account.
7072  */
7073 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7074 {
7075         struct btrfs_block_group_cache *block_group;
7076         u64 free_bytes = 0;
7077         int factor;
7078
7079         list_for_each_entry(block_group, groups_list, list) {
7080                 spin_lock(&block_group->lock);
7081
7082                 if (!block_group->ro) {
7083                         spin_unlock(&block_group->lock);
7084                         continue;
7085                 }
7086
7087                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7088                                           BTRFS_BLOCK_GROUP_RAID10 |
7089                                           BTRFS_BLOCK_GROUP_DUP))
7090                         factor = 2;
7091                 else
7092                         factor = 1;
7093
7094                 free_bytes += (block_group->key.offset -
7095                                btrfs_block_group_used(&block_group->item)) *
7096                                factor;
7097
7098                 spin_unlock(&block_group->lock);
7099         }
7100
7101         return free_bytes;
7102 }
7103
7104 /*
7105  * helper to account the unused space of all the readonly block group in the
7106  * space_info. takes mirrors into account.
7107  */
7108 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7109 {
7110         int i;
7111         u64 free_bytes = 0;
7112
7113         spin_lock(&sinfo->lock);
7114
7115         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7116                 if (!list_empty(&sinfo->block_groups[i]))
7117                         free_bytes += __btrfs_get_ro_block_group_free_space(
7118                                                 &sinfo->block_groups[i]);
7119
7120         spin_unlock(&sinfo->lock);
7121
7122         return free_bytes;
7123 }
7124
7125 int btrfs_set_block_group_rw(struct btrfs_root *root,
7126                               struct btrfs_block_group_cache *cache)
7127 {
7128         struct btrfs_space_info *sinfo = cache->space_info;
7129         u64 num_bytes;
7130
7131         BUG_ON(!cache->ro);
7132
7133         spin_lock(&sinfo->lock);
7134         spin_lock(&cache->lock);
7135         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7136                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7137         sinfo->bytes_readonly -= num_bytes;
7138         cache->ro = 0;
7139         spin_unlock(&cache->lock);
7140         spin_unlock(&sinfo->lock);
7141         return 0;
7142 }
7143
7144 /*
7145  * checks to see if its even possible to relocate this block group.
7146  *
7147  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7148  * ok to go ahead and try.
7149  */
7150 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7151 {
7152         struct btrfs_block_group_cache *block_group;
7153         struct btrfs_space_info *space_info;
7154         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7155         struct btrfs_device *device;
7156         u64 min_free;
7157         u64 dev_min = 1;
7158         u64 dev_nr = 0;
7159         int index;
7160         int full = 0;
7161         int ret = 0;
7162
7163         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7164
7165         /* odd, couldn't find the block group, leave it alone */
7166         if (!block_group)
7167                 return -1;
7168
7169         min_free = btrfs_block_group_used(&block_group->item);
7170
7171         /* no bytes used, we're good */
7172         if (!min_free)
7173                 goto out;
7174
7175         space_info = block_group->space_info;
7176         spin_lock(&space_info->lock);
7177
7178         full = space_info->full;
7179
7180         /*
7181          * if this is the last block group we have in this space, we can't
7182          * relocate it unless we're able to allocate a new chunk below.
7183          *
7184          * Otherwise, we need to make sure we have room in the space to handle
7185          * all of the extents from this block group.  If we can, we're good
7186          */
7187         if ((space_info->total_bytes != block_group->key.offset) &&
7188             (space_info->bytes_used + space_info->bytes_reserved +
7189              space_info->bytes_pinned + space_info->bytes_readonly +
7190              min_free < space_info->total_bytes)) {
7191                 spin_unlock(&space_info->lock);
7192                 goto out;
7193         }
7194         spin_unlock(&space_info->lock);
7195
7196         /*
7197          * ok we don't have enough space, but maybe we have free space on our
7198          * devices to allocate new chunks for relocation, so loop through our
7199          * alloc devices and guess if we have enough space.  However, if we
7200          * were marked as full, then we know there aren't enough chunks, and we
7201          * can just return.
7202          */
7203         ret = -1;
7204         if (full)
7205                 goto out;
7206
7207         /*
7208          * index:
7209          *      0: raid10
7210          *      1: raid1
7211          *      2: dup
7212          *      3: raid0
7213          *      4: single
7214          */
7215         index = get_block_group_index(block_group);
7216         if (index == 0) {
7217                 dev_min = 4;
7218                 /* Divide by 2 */
7219                 min_free >>= 1;
7220         } else if (index == 1) {
7221                 dev_min = 2;
7222         } else if (index == 2) {
7223                 /* Multiply by 2 */
7224                 min_free <<= 1;
7225         } else if (index == 3) {
7226                 dev_min = fs_devices->rw_devices;
7227                 do_div(min_free, dev_min);
7228         }
7229
7230         mutex_lock(&root->fs_info->chunk_mutex);
7231         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7232                 u64 dev_offset;
7233
7234                 /*
7235                  * check to make sure we can actually find a chunk with enough
7236                  * space to fit our block group in.
7237                  */
7238                 if (device->total_bytes > device->bytes_used + min_free) {
7239                         ret = find_free_dev_extent(device, min_free,
7240                                                    &dev_offset, NULL);
7241                         if (!ret)
7242                                 dev_nr++;
7243
7244                         if (dev_nr >= dev_min)
7245                                 break;
7246
7247                         ret = -1;
7248                 }
7249         }
7250         mutex_unlock(&root->fs_info->chunk_mutex);
7251 out:
7252         btrfs_put_block_group(block_group);
7253         return ret;
7254 }
7255
7256 static int find_first_block_group(struct btrfs_root *root,
7257                 struct btrfs_path *path, struct btrfs_key *key)
7258 {
7259         int ret = 0;
7260         struct btrfs_key found_key;
7261         struct extent_buffer *leaf;
7262         int slot;
7263
7264         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7265         if (ret < 0)
7266                 goto out;
7267
7268         while (1) {
7269                 slot = path->slots[0];
7270                 leaf = path->nodes[0];
7271                 if (slot >= btrfs_header_nritems(leaf)) {
7272                         ret = btrfs_next_leaf(root, path);
7273                         if (ret == 0)
7274                                 continue;
7275                         if (ret < 0)
7276                                 goto out;
7277                         break;
7278                 }
7279                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7280
7281                 if (found_key.objectid >= key->objectid &&
7282                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7283                         ret = 0;
7284                         goto out;
7285                 }
7286                 path->slots[0]++;
7287         }
7288 out:
7289         return ret;
7290 }
7291
7292 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7293 {
7294         struct btrfs_block_group_cache *block_group;
7295         u64 last = 0;
7296
7297         while (1) {
7298                 struct inode *inode;
7299
7300                 block_group = btrfs_lookup_first_block_group(info, last);
7301                 while (block_group) {
7302                         spin_lock(&block_group->lock);
7303                         if (block_group->iref)
7304                                 break;
7305                         spin_unlock(&block_group->lock);
7306                         block_group = next_block_group(info->tree_root,
7307                                                        block_group);
7308                 }
7309                 if (!block_group) {
7310                         if (last == 0)
7311                                 break;
7312                         last = 0;
7313                         continue;
7314                 }
7315
7316                 inode = block_group->inode;
7317                 block_group->iref = 0;
7318                 block_group->inode = NULL;
7319                 spin_unlock(&block_group->lock);
7320                 iput(inode);
7321                 last = block_group->key.objectid + block_group->key.offset;
7322                 btrfs_put_block_group(block_group);
7323         }
7324 }
7325
7326 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7327 {
7328         struct btrfs_block_group_cache *block_group;
7329         struct btrfs_space_info *space_info;
7330         struct btrfs_caching_control *caching_ctl;
7331         struct rb_node *n;
7332
7333         down_write(&info->extent_commit_sem);
7334         while (!list_empty(&info->caching_block_groups)) {
7335                 caching_ctl = list_entry(info->caching_block_groups.next,
7336                                          struct btrfs_caching_control, list);
7337                 list_del(&caching_ctl->list);
7338                 put_caching_control(caching_ctl);
7339         }
7340         up_write(&info->extent_commit_sem);
7341
7342         spin_lock(&info->block_group_cache_lock);
7343         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7344                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7345                                        cache_node);
7346                 rb_erase(&block_group->cache_node,
7347                          &info->block_group_cache_tree);
7348                 spin_unlock(&info->block_group_cache_lock);
7349
7350                 down_write(&block_group->space_info->groups_sem);
7351                 list_del(&block_group->list);
7352                 up_write(&block_group->space_info->groups_sem);
7353
7354                 if (block_group->cached == BTRFS_CACHE_STARTED)
7355                         wait_block_group_cache_done(block_group);
7356
7357                 /*
7358                  * We haven't cached this block group, which means we could
7359                  * possibly have excluded extents on this block group.
7360                  */
7361                 if (block_group->cached == BTRFS_CACHE_NO)
7362                         free_excluded_extents(info->extent_root, block_group);
7363
7364                 btrfs_remove_free_space_cache(block_group);
7365                 btrfs_put_block_group(block_group);
7366
7367                 spin_lock(&info->block_group_cache_lock);
7368         }
7369         spin_unlock(&info->block_group_cache_lock);
7370
7371         /* now that all the block groups are freed, go through and
7372          * free all the space_info structs.  This is only called during
7373          * the final stages of unmount, and so we know nobody is
7374          * using them.  We call synchronize_rcu() once before we start,
7375          * just to be on the safe side.
7376          */
7377         synchronize_rcu();
7378
7379         release_global_block_rsv(info);
7380
7381         while(!list_empty(&info->space_info)) {
7382                 space_info = list_entry(info->space_info.next,
7383                                         struct btrfs_space_info,
7384                                         list);
7385                 if (space_info->bytes_pinned > 0 ||
7386                     space_info->bytes_reserved > 0 ||
7387                     space_info->bytes_may_use > 0) {
7388                         WARN_ON(1);
7389                         dump_space_info(space_info, 0, 0);
7390                 }
7391                 list_del(&space_info->list);
7392                 kfree(space_info);
7393         }
7394         return 0;
7395 }
7396
7397 static void __link_block_group(struct btrfs_space_info *space_info,
7398                                struct btrfs_block_group_cache *cache)
7399 {
7400         int index = get_block_group_index(cache);
7401
7402         down_write(&space_info->groups_sem);
7403         list_add_tail(&cache->list, &space_info->block_groups[index]);
7404         up_write(&space_info->groups_sem);
7405 }
7406
7407 int btrfs_read_block_groups(struct btrfs_root *root)
7408 {
7409         struct btrfs_path *path;
7410         int ret;
7411         struct btrfs_block_group_cache *cache;
7412         struct btrfs_fs_info *info = root->fs_info;
7413         struct btrfs_space_info *space_info;
7414         struct btrfs_key key;
7415         struct btrfs_key found_key;
7416         struct extent_buffer *leaf;
7417         int need_clear = 0;
7418         u64 cache_gen;
7419
7420         root = info->extent_root;
7421         key.objectid = 0;
7422         key.offset = 0;
7423         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7424         path = btrfs_alloc_path();
7425         if (!path)
7426                 return -ENOMEM;
7427         path->reada = 1;
7428
7429         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7430         if (btrfs_test_opt(root, SPACE_CACHE) &&
7431             btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7432                 need_clear = 1;
7433         if (btrfs_test_opt(root, CLEAR_CACHE))
7434                 need_clear = 1;
7435
7436         while (1) {
7437                 ret = find_first_block_group(root, path, &key);
7438                 if (ret > 0)
7439                         break;
7440                 if (ret != 0)
7441                         goto error;
7442                 leaf = path->nodes[0];
7443                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7444                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7445                 if (!cache) {
7446                         ret = -ENOMEM;
7447                         goto error;
7448                 }
7449                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7450                                                 GFP_NOFS);
7451                 if (!cache->free_space_ctl) {
7452                         kfree(cache);
7453                         ret = -ENOMEM;
7454                         goto error;
7455                 }
7456
7457                 atomic_set(&cache->count, 1);
7458                 spin_lock_init(&cache->lock);
7459                 cache->fs_info = info;
7460                 INIT_LIST_HEAD(&cache->list);
7461                 INIT_LIST_HEAD(&cache->cluster_list);
7462
7463                 if (need_clear)
7464                         cache->disk_cache_state = BTRFS_DC_CLEAR;
7465
7466                 read_extent_buffer(leaf, &cache->item,
7467                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7468                                    sizeof(cache->item));
7469                 memcpy(&cache->key, &found_key, sizeof(found_key));
7470
7471                 key.objectid = found_key.objectid + found_key.offset;
7472                 btrfs_release_path(path);
7473                 cache->flags = btrfs_block_group_flags(&cache->item);
7474                 cache->sectorsize = root->sectorsize;
7475
7476                 btrfs_init_free_space_ctl(cache);
7477
7478                 /*
7479                  * We need to exclude the super stripes now so that the space
7480                  * info has super bytes accounted for, otherwise we'll think
7481                  * we have more space than we actually do.
7482                  */
7483                 exclude_super_stripes(root, cache);
7484
7485                 /*
7486                  * check for two cases, either we are full, and therefore
7487                  * don't need to bother with the caching work since we won't
7488                  * find any space, or we are empty, and we can just add all
7489                  * the space in and be done with it.  This saves us _alot_ of
7490                  * time, particularly in the full case.
7491                  */
7492                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7493                         cache->last_byte_to_unpin = (u64)-1;
7494                         cache->cached = BTRFS_CACHE_FINISHED;
7495                         free_excluded_extents(root, cache);
7496                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7497                         cache->last_byte_to_unpin = (u64)-1;
7498                         cache->cached = BTRFS_CACHE_FINISHED;
7499                         add_new_free_space(cache, root->fs_info,
7500                                            found_key.objectid,
7501                                            found_key.objectid +
7502                                            found_key.offset);
7503                         free_excluded_extents(root, cache);
7504                 }
7505
7506                 ret = update_space_info(info, cache->flags, found_key.offset,
7507                                         btrfs_block_group_used(&cache->item),
7508                                         &space_info);
7509                 BUG_ON(ret);
7510                 cache->space_info = space_info;
7511                 spin_lock(&cache->space_info->lock);
7512                 cache->space_info->bytes_readonly += cache->bytes_super;
7513                 spin_unlock(&cache->space_info->lock);
7514
7515                 __link_block_group(space_info, cache);
7516
7517                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7518                 BUG_ON(ret);
7519
7520                 set_avail_alloc_bits(root->fs_info, cache->flags);
7521                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7522                         set_block_group_ro(cache, 1);
7523         }
7524
7525         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7526                 if (!(get_alloc_profile(root, space_info->flags) &
7527                       (BTRFS_BLOCK_GROUP_RAID10 |
7528                        BTRFS_BLOCK_GROUP_RAID1 |
7529                        BTRFS_BLOCK_GROUP_DUP)))
7530                         continue;
7531                 /*
7532                  * avoid allocating from un-mirrored block group if there are
7533                  * mirrored block groups.
7534                  */
7535                 list_for_each_entry(cache, &space_info->block_groups[3], list)
7536                         set_block_group_ro(cache, 1);
7537                 list_for_each_entry(cache, &space_info->block_groups[4], list)
7538                         set_block_group_ro(cache, 1);
7539         }
7540
7541         init_global_block_rsv(info);
7542         ret = 0;
7543 error:
7544         btrfs_free_path(path);
7545         return ret;
7546 }
7547
7548 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7549                            struct btrfs_root *root, u64 bytes_used,
7550                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7551                            u64 size)
7552 {
7553         int ret;
7554         struct btrfs_root *extent_root;
7555         struct btrfs_block_group_cache *cache;
7556
7557         extent_root = root->fs_info->extent_root;
7558
7559         root->fs_info->last_trans_log_full_commit = trans->transid;
7560
7561         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7562         if (!cache)
7563                 return -ENOMEM;
7564         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7565                                         GFP_NOFS);
7566         if (!cache->free_space_ctl) {
7567                 kfree(cache);
7568                 return -ENOMEM;
7569         }
7570
7571         cache->key.objectid = chunk_offset;
7572         cache->key.offset = size;
7573         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7574         cache->sectorsize = root->sectorsize;
7575         cache->fs_info = root->fs_info;
7576
7577         atomic_set(&cache->count, 1);
7578         spin_lock_init(&cache->lock);
7579         INIT_LIST_HEAD(&cache->list);
7580         INIT_LIST_HEAD(&cache->cluster_list);
7581
7582         btrfs_init_free_space_ctl(cache);
7583
7584         btrfs_set_block_group_used(&cache->item, bytes_used);
7585         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7586         cache->flags = type;
7587         btrfs_set_block_group_flags(&cache->item, type);
7588
7589         cache->last_byte_to_unpin = (u64)-1;
7590         cache->cached = BTRFS_CACHE_FINISHED;
7591         exclude_super_stripes(root, cache);
7592
7593         add_new_free_space(cache, root->fs_info, chunk_offset,
7594                            chunk_offset + size);
7595
7596         free_excluded_extents(root, cache);
7597
7598         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7599                                 &cache->space_info);
7600         BUG_ON(ret);
7601         update_global_block_rsv(root->fs_info);
7602
7603         spin_lock(&cache->space_info->lock);
7604         cache->space_info->bytes_readonly += cache->bytes_super;
7605         spin_unlock(&cache->space_info->lock);
7606
7607         __link_block_group(cache->space_info, cache);
7608
7609         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7610         BUG_ON(ret);
7611
7612         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7613                                 sizeof(cache->item));
7614         BUG_ON(ret);
7615
7616         set_avail_alloc_bits(extent_root->fs_info, type);
7617
7618         return 0;
7619 }
7620
7621 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7622 {
7623         u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
7624
7625         /* chunk -> extended profile */
7626         if (extra_flags == 0)
7627                 extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
7628
7629         if (flags & BTRFS_BLOCK_GROUP_DATA)
7630                 fs_info->avail_data_alloc_bits &= ~extra_flags;
7631         if (flags & BTRFS_BLOCK_GROUP_METADATA)
7632                 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7633         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7634                 fs_info->avail_system_alloc_bits &= ~extra_flags;
7635 }
7636
7637 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7638                              struct btrfs_root *root, u64 group_start)
7639 {
7640         struct btrfs_path *path;
7641         struct btrfs_block_group_cache *block_group;
7642         struct btrfs_free_cluster *cluster;
7643         struct btrfs_root *tree_root = root->fs_info->tree_root;
7644         struct btrfs_key key;
7645         struct inode *inode;
7646         int ret;
7647         int index;
7648         int factor;
7649
7650         root = root->fs_info->extent_root;
7651
7652         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7653         BUG_ON(!block_group);
7654         BUG_ON(!block_group->ro);
7655
7656         /*
7657          * Free the reserved super bytes from this block group before
7658          * remove it.
7659          */
7660         free_excluded_extents(root, block_group);
7661
7662         memcpy(&key, &block_group->key, sizeof(key));
7663         index = get_block_group_index(block_group);
7664         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7665                                   BTRFS_BLOCK_GROUP_RAID1 |
7666                                   BTRFS_BLOCK_GROUP_RAID10))
7667                 factor = 2;
7668         else
7669                 factor = 1;
7670
7671         /* make sure this block group isn't part of an allocation cluster */
7672         cluster = &root->fs_info->data_alloc_cluster;
7673         spin_lock(&cluster->refill_lock);
7674         btrfs_return_cluster_to_free_space(block_group, cluster);
7675         spin_unlock(&cluster->refill_lock);
7676
7677         /*
7678          * make sure this block group isn't part of a metadata
7679          * allocation cluster
7680          */
7681         cluster = &root->fs_info->meta_alloc_cluster;
7682         spin_lock(&cluster->refill_lock);
7683         btrfs_return_cluster_to_free_space(block_group, cluster);
7684         spin_unlock(&cluster->refill_lock);
7685
7686         path = btrfs_alloc_path();
7687         if (!path) {
7688                 ret = -ENOMEM;
7689                 goto out;
7690         }
7691
7692         inode = lookup_free_space_inode(tree_root, block_group, path);
7693         if (!IS_ERR(inode)) {
7694                 ret = btrfs_orphan_add(trans, inode);
7695                 BUG_ON(ret);
7696                 clear_nlink(inode);
7697                 /* One for the block groups ref */
7698                 spin_lock(&block_group->lock);
7699                 if (block_group->iref) {
7700                         block_group->iref = 0;
7701                         block_group->inode = NULL;
7702                         spin_unlock(&block_group->lock);
7703                         iput(inode);
7704                 } else {
7705                         spin_unlock(&block_group->lock);
7706                 }
7707                 /* One for our lookup ref */
7708                 btrfs_add_delayed_iput(inode);
7709         }
7710
7711         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7712         key.offset = block_group->key.objectid;
7713         key.type = 0;
7714
7715         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7716         if (ret < 0)
7717                 goto out;
7718         if (ret > 0)
7719                 btrfs_release_path(path);
7720         if (ret == 0) {
7721                 ret = btrfs_del_item(trans, tree_root, path);
7722                 if (ret)
7723                         goto out;
7724                 btrfs_release_path(path);
7725         }
7726
7727         spin_lock(&root->fs_info->block_group_cache_lock);
7728         rb_erase(&block_group->cache_node,
7729                  &root->fs_info->block_group_cache_tree);
7730         spin_unlock(&root->fs_info->block_group_cache_lock);
7731
7732         down_write(&block_group->space_info->groups_sem);
7733         /*
7734          * we must use list_del_init so people can check to see if they
7735          * are still on the list after taking the semaphore
7736          */
7737         list_del_init(&block_group->list);
7738         if (list_empty(&block_group->space_info->block_groups[index]))
7739                 clear_avail_alloc_bits(root->fs_info, block_group->flags);
7740         up_write(&block_group->space_info->groups_sem);
7741
7742         if (block_group->cached == BTRFS_CACHE_STARTED)
7743                 wait_block_group_cache_done(block_group);
7744
7745         btrfs_remove_free_space_cache(block_group);
7746
7747         spin_lock(&block_group->space_info->lock);
7748         block_group->space_info->total_bytes -= block_group->key.offset;
7749         block_group->space_info->bytes_readonly -= block_group->key.offset;
7750         block_group->space_info->disk_total -= block_group->key.offset * factor;
7751         spin_unlock(&block_group->space_info->lock);
7752
7753         memcpy(&key, &block_group->key, sizeof(key));
7754
7755         btrfs_clear_space_info_full(root->fs_info);
7756
7757         btrfs_put_block_group(block_group);
7758         btrfs_put_block_group(block_group);
7759
7760         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7761         if (ret > 0)
7762                 ret = -EIO;
7763         if (ret < 0)
7764                 goto out;
7765
7766         ret = btrfs_del_item(trans, root, path);
7767 out:
7768         btrfs_free_path(path);
7769         return ret;
7770 }
7771
7772 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7773 {
7774         struct btrfs_space_info *space_info;
7775         struct btrfs_super_block *disk_super;
7776         u64 features;
7777         u64 flags;
7778         int mixed = 0;
7779         int ret;
7780
7781         disk_super = fs_info->super_copy;
7782         if (!btrfs_super_root(disk_super))
7783                 return 1;
7784
7785         features = btrfs_super_incompat_flags(disk_super);
7786         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7787                 mixed = 1;
7788
7789         flags = BTRFS_BLOCK_GROUP_SYSTEM;
7790         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7791         if (ret)
7792                 goto out;
7793
7794         if (mixed) {
7795                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7796                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7797         } else {
7798                 flags = BTRFS_BLOCK_GROUP_METADATA;
7799                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7800                 if (ret)
7801                         goto out;
7802
7803                 flags = BTRFS_BLOCK_GROUP_DATA;
7804                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7805         }
7806 out:
7807         return ret;
7808 }
7809
7810 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7811 {
7812         return unpin_extent_range(root, start, end);
7813 }
7814
7815 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7816                                u64 num_bytes, u64 *actual_bytes)
7817 {
7818         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7819 }
7820
7821 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7822 {
7823         struct btrfs_fs_info *fs_info = root->fs_info;
7824         struct btrfs_block_group_cache *cache = NULL;
7825         u64 group_trimmed;
7826         u64 start;
7827         u64 end;
7828         u64 trimmed = 0;
7829         int ret = 0;
7830
7831         cache = btrfs_lookup_block_group(fs_info, range->start);
7832
7833         while (cache) {
7834                 if (cache->key.objectid >= (range->start + range->len)) {
7835                         btrfs_put_block_group(cache);
7836                         break;
7837                 }
7838
7839                 start = max(range->start, cache->key.objectid);
7840                 end = min(range->start + range->len,
7841                                 cache->key.objectid + cache->key.offset);
7842
7843                 if (end - start >= range->minlen) {
7844                         if (!block_group_cache_done(cache)) {
7845                                 ret = cache_block_group(cache, NULL, root, 0);
7846                                 if (!ret)
7847                                         wait_block_group_cache_done(cache);
7848                         }
7849                         ret = btrfs_trim_block_group(cache,
7850                                                      &group_trimmed,
7851                                                      start,
7852                                                      end,
7853                                                      range->minlen);
7854
7855                         trimmed += group_trimmed;
7856                         if (ret) {
7857                                 btrfs_put_block_group(cache);
7858                                 break;
7859                         }
7860                 }
7861
7862                 cache = next_block_group(fs_info->tree_root, cache);
7863         }
7864
7865         range->len = trimmed;
7866         return ret;
7867 }