fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/kthread.h>
  25 #include <linux/slab.h>
  26 #include <linux/ratelimit.h>
  27 #include "compat.h"
  28 #include "hash.h"
  29 #include "ctree.h"
  30 #include "disk-io.h"
  31 #include "print-tree.h"
  32 #include "transaction.h"
  33 #include "volumes.h"
  34 #include "locking.h"
  35 #include "free-space-cache.h"
  36
  37 /* control flags for do_chunk_alloc's force field
  38  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
  39  * if we really need one.
  40  *
  41  * CHUNK_ALLOC_FORCE means it must try to allocate one
  42  *
  43  * CHUNK_ALLOC_LIMITED means to only try and allocate one
  44  * if we have very few chunks already allocated.  This is
  45  * used as part of the clustering code to help make sure
  46  * we have a good pool of storage to cluster in, without
  47  * filling the FS with empty chunks
  48  *
  49  */
  50 enum {
  51         CHUNK_ALLOC_NO_FORCE = 0,
  52         CHUNK_ALLOC_FORCE = 1,
  53         CHUNK_ALLOC_LIMITED = 2,
  54 };
  55
  56 /*
  57  * Control how reservations are dealt with.
  58  *
  59  * RESERVE_FREE - freeing a reservation.
  60  * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
  61  *   ENOSPC accounting
  62  * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
  63  *   bytes_may_use as the ENOSPC accounting is done elsewhere
  64  */
  65 enum {
  66         RESERVE_FREE = 0,
  67         RESERVE_ALLOC = 1,
  68         RESERVE_ALLOC_NO_ACCOUNT = 2,
  69 };
  70
  71 static int update_block_group(struct btrfs_trans_handle *trans,
  72                               struct btrfs_root *root,
  73                               u64 bytenr, u64 num_bytes, int alloc);
  74 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  75                                 struct btrfs_root *root,
  76                                 u64 bytenr, u64 num_bytes, u64 parent,
  77                                 u64 root_objectid, u64 owner_objectid,
  78                                 u64 owner_offset, int refs_to_drop,
  79                                 struct btrfs_delayed_extent_op *extra_op);
  80 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  81                                     struct extent_buffer *leaf,
  82                                     struct btrfs_extent_item *ei);
  83 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  84                                       struct btrfs_root *root,
  85                                       u64 parent, u64 root_objectid,
  86                                       u64 flags, u64 owner, u64 offset,
  87                                       struct btrfs_key *ins, int ref_mod);
  88 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  89                                      struct btrfs_root *root,
  90                                      u64 parent, u64 root_objectid,
  91                                      u64 flags, struct btrfs_disk_key *key,
  92                                      int level, struct btrfs_key *ins);
  93 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  94                           struct btrfs_root *extent_root, u64 alloc_bytes,
  95                           u64 flags, int force);
  96 static int find_next_key(struct btrfs_path *path, int level,
  97                          struct btrfs_key *key);
  98 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  99                             int dump_block_groups);
 100 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
 101                                        u64 num_bytes, int reserve);
 102
 103 static noinline int
 104 block_group_cache_done(struct btrfs_block_group_cache *cache)
 105 {
 106         smp_mb();
 107         return cache->cached == BTRFS_CACHE_FINISHED;
 108 }
 109
 110 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 111 {
 112         return (cache->flags & bits) == bits;
 113 }
 114
 115 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
 116 {
 117         atomic_inc(&cache->count);
 118 }
 119
 120 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 121 {
 122         if (atomic_dec_and_test(&cache->count)) {
 123                 WARN_ON(cache->pinned > 0);
 124                 WARN_ON(cache->reserved > 0);
 125                 kfree(cache->free_space_ctl);
 126                 kfree(cache);
 127         }
 128 }
 129
 130 /*
 131  * this adds the block group to the fs_info rb tree for the block group
 132  * cache
 133  */
 134 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 135                                 struct btrfs_block_group_cache *block_group)
 136 {
 137         struct rb_node **p;
 138         struct rb_node *parent = NULL;
 139         struct btrfs_block_group_cache *cache;
 140
 141         spin_lock(&info->block_group_cache_lock);
 142         p = &info->block_group_cache_tree.rb_node;
 143
 144         while (*p) {
 145                 parent = *p;
 146                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 147                                  cache_node);
 148                 if (block_group->key.objectid < cache->key.objectid) {
 149                         p = &(*p)->rb_left;
 150                 } else if (block_group->key.objectid > cache->key.objectid) {
 151                         p = &(*p)->rb_right;
 152                 } else {
 153                         spin_unlock(&info->block_group_cache_lock);
 154                         return -EEXIST;
 155                 }
 156         }
 157
 158         rb_link_node(&block_group->cache_node, parent, p);
 159         rb_insert_color(&block_group->cache_node,
 160                         &info->block_group_cache_tree);
 161         spin_unlock(&info->block_group_cache_lock);
 162
 163         return 0;
 164 }
 165
 166 /*
 167  * This will return the block group at or after bytenr if contains is 0, else
 168  * it will return the block group that contains the bytenr
 169  */
 170 static struct btrfs_block_group_cache *
 171 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 172                               int contains)
 173 {
 174         struct btrfs_block_group_cache *cache, *ret = NULL;
 175         struct rb_node *n;
 176         u64 end, start;
 177
 178         spin_lock(&info->block_group_cache_lock);
 179         n = info->block_group_cache_tree.rb_node;
 180
 181         while (n) {
 182                 cache = rb_entry(n, struct btrfs_block_group_cache,
 183                                  cache_node);
 184                 end = cache->key.objectid + cache->key.offset - 1;
 185                 start = cache->key.objectid;
 186
 187                 if (bytenr < start) {
 188                         if (!contains && (!ret || start < ret->key.objectid))
 189                                 ret = cache;
 190                         n = n->rb_left;
 191                 } else if (bytenr > start) {
 192                         if (contains && bytenr <= end) {
 193                                 ret = cache;
 194                                 break;
 195                         }
 196                         n = n->rb_right;
 197                 } else {
 198                         ret = cache;
 199                         break;
 200                 }
 201         }
 202         if (ret)
 203                 btrfs_get_block_group(ret);
 204         spin_unlock(&info->block_group_cache_lock);
 205
 206         return ret;
 207 }
 208
 209 static int add_excluded_extent(struct btrfs_root *root,
 210                                u64 start, u64 num_bytes)
 211 {
 212         u64 end = start + num_bytes - 1;
 213         set_extent_bits(&root->fs_info->freed_extents[0],
 214                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 215         set_extent_bits(&root->fs_info->freed_extents[1],
 216                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 217         return 0;
 218 }
 219
 220 static void free_excluded_extents(struct btrfs_root *root,
 221                                   struct btrfs_block_group_cache *cache)
 222 {
 223         u64 start, end;
 224
 225         start = cache->key.objectid;
 226         end = start + cache->key.offset - 1;
 227
 228         clear_extent_bits(&root->fs_info->freed_extents[0],
 229                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 230         clear_extent_bits(&root->fs_info->freed_extents[1],
 231                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 232 }
 233
 234 static int exclude_super_stripes(struct btrfs_root *root,
 235                                  struct btrfs_block_group_cache *cache)
 236 {
 237         u64 bytenr;
 238         u64 *logical;
 239         int stripe_len;
 240         int i, nr, ret;
 241
 242         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 243                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 244                 cache->bytes_super += stripe_len;
 245                 ret = add_excluded_extent(root, cache->key.objectid,
 246                                           stripe_len);
 247                 BUG_ON(ret);
 248         }
 249
 250         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 251                 bytenr = btrfs_sb_offset(i);
 252                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 253                                        cache->key.objectid, bytenr,
 254                                        0, &logical, &nr, &stripe_len);
 255                 BUG_ON(ret);
 256
 257                 while (nr--) {
 258                         cache->bytes_super += stripe_len;
 259                         ret = add_excluded_extent(root, logical[nr],
 260                                                   stripe_len);
 261                         BUG_ON(ret);
 262                 }
 263
 264                 kfree(logical);
 265         }
 266         return 0;
 267 }
 268
 269 static struct btrfs_caching_control *
 270 get_caching_control(struct btrfs_block_group_cache *cache)
 271 {
 272         struct btrfs_caching_control *ctl;
 273
 274         spin_lock(&cache->lock);
 275         if (cache->cached != BTRFS_CACHE_STARTED) {
 276                 spin_unlock(&cache->lock);
 277                 return NULL;
 278         }
 279
 280         /* We're loading it the fast way, so we don't have a caching_ctl. */
 281         if (!cache->caching_ctl) {
 282                 spin_unlock(&cache->lock);
 283                 return NULL;
 284         }
 285
 286         ctl = cache->caching_ctl;
 287         atomic_inc(&ctl->count);
 288         spin_unlock(&cache->lock);
 289         return ctl;
 290 }
 291
 292 static void put_caching_control(struct btrfs_caching_control *ctl)
 293 {
 294         if (atomic_dec_and_test(&ctl->count))
 295                 kfree(ctl);
 296 }
 297
 298 /*
 299  * this is only called by cache_block_group, since we could have freed extents
 300  * we need to check the pinned_extents for any extents that can't be used yet
 301  * since their free space will be released as soon as the transaction commits.
 302  */
 303 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 304                               struct btrfs_fs_info *info, u64 start, u64 end)
 305 {
 306         u64 extent_start, extent_end, size, total_added = 0;
 307         int ret;
 308
 309         while (start < end) {
 310                 ret = find_first_extent_bit(info->pinned_extents, start,
 311                                             &extent_start, &extent_end,
 312                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 313                 if (ret)
 314                         break;
 315
 316                 if (extent_start <= start) {
 317                         start = extent_end + 1;
 318                 } else if (extent_start > start && extent_start < end) {
 319                         size = extent_start - start;
 320                         total_added += size;
 321                         ret = btrfs_add_free_space(block_group, start,
 322                                                    size);
 323                         BUG_ON(ret);
 324                         start = extent_end + 1;
 325                 } else {
 326                         break;
 327                 }
 328         }
 329
 330         if (start < end) {
 331                 size = end - start;
 332                 total_added += size;
 333                 ret = btrfs_add_free_space(block_group, start, size);
 334                 BUG_ON(ret);
 335         }
 336
 337         return total_added;
 338 }
 339
 340 static noinline void caching_thread(struct btrfs_work *work)
 341 {
 342         struct btrfs_block_group_cache *block_group;
 343         struct btrfs_fs_info *fs_info;
 344         struct btrfs_caching_control *caching_ctl;
 345         struct btrfs_root *extent_root;
 346         struct btrfs_path *path;
 347         struct extent_buffer *leaf;
 348         struct btrfs_key key;
 349         u64 total_found = 0;
 350         u64 last = 0;
 351         u32 nritems;
 352         int ret = 0;
 353
 354         caching_ctl = container_of(work, struct btrfs_caching_control, work);
 355         block_group = caching_ctl->block_group;
 356         fs_info = block_group->fs_info;
 357         extent_root = fs_info->extent_root;
 358
 359         path = btrfs_alloc_path();
 360         if (!path)
 361                 goto out;
 362
 363         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 364
 365         /*
 366          * We don't want to deadlock with somebody trying to allocate a new
 367          * extent for the extent root while also trying to search the extent
 368          * root to add free space.  So we skip locking and search the commit
 369          * root, since its read-only
 370          */
 371         path->skip_locking = 1;
 372         path->search_commit_root = 1;
 373         path->reada = 1;
 374
 375         key.objectid = last;
 376         key.offset = 0;
 377         key.type = BTRFS_EXTENT_ITEM_KEY;
 378 again:
 379         mutex_lock(&caching_ctl->mutex);
 380         /* need to make sure the commit_root doesn't disappear */
 381         down_read(&fs_info->extent_commit_sem);
 382
 383         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 384         if (ret < 0)
 385                 goto err;
 386
 387         leaf = path->nodes[0];
 388         nritems = btrfs_header_nritems(leaf);
 389
 390         while (1) {
 391                 if (btrfs_fs_closing(fs_info) > 1) {
 392                         last = (u64)-1;
 393                         break;
 394                 }
 395
 396                 if (path->slots[0] < nritems) {
 397                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 398                 } else {
 399                         ret = find_next_key(path, 0, &key);
 400                         if (ret)
 401                                 break;
 402
 403                         if (need_resched() ||
 404                             btrfs_next_leaf(extent_root, path)) {
 405                                 caching_ctl->progress = last;
 406                                 btrfs_release_path(path);
 407                                 up_read(&fs_info->extent_commit_sem);
 408                                 mutex_unlock(&caching_ctl->mutex);
 409                                 cond_resched();
 410                                 goto again;
 411                         }
 412                         leaf = path->nodes[0];
 413                         nritems = btrfs_header_nritems(leaf);
 414                         continue;
 415                 }
 416
 417                 if (key.objectid < block_group->key.objectid) {
 418                         path->slots[0]++;
 419                         continue;
 420                 }
 421
 422                 if (key.objectid >= block_group->key.objectid +
 423                     block_group->key.offset)
 424                         break;
 425
 426                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 427                         total_found += add_new_free_space(block_group,
 428                                                           fs_info, last,
 429                                                           key.objectid);
 430                         last = key.objectid + key.offset;
 431
 432                         if (total_found > (1024 * 1024 * 2)) {
 433                                 total_found = 0;
 434                                 wake_up(&caching_ctl->wait);
 435                         }
 436                 }
 437                 path->slots[0]++;
 438         }
 439         ret = 0;
 440
 441         total_found += add_new_free_space(block_group, fs_info, last,
 442                                           block_group->key.objectid +
 443                                           block_group->key.offset);
 444         caching_ctl->progress = (u64)-1;
 445
 446         spin_lock(&block_group->lock);
 447         block_group->caching_ctl = NULL;
 448         block_group->cached = BTRFS_CACHE_FINISHED;
 449         spin_unlock(&block_group->lock);
 450
 451 err:
 452         btrfs_free_path(path);
 453         up_read(&fs_info->extent_commit_sem);
 454
 455         free_excluded_extents(extent_root, block_group);
 456
 457         mutex_unlock(&caching_ctl->mutex);
 458 out:
 459         wake_up(&caching_ctl->wait);
 460
 461         put_caching_control(caching_ctl);
 462         btrfs_put_block_group(block_group);
 463 }
 464
 465 static int cache_block_group(struct btrfs_block_group_cache *cache,
 466                              struct btrfs_trans_handle *trans,
 467                              struct btrfs_root *root,
 468                              int load_cache_only)
 469 {
 470         DEFINE_WAIT(wait);
 471         struct btrfs_fs_info *fs_info = cache->fs_info;
 472         struct btrfs_caching_control *caching_ctl;
 473         int ret = 0;
 474
 475         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 476         BUG_ON(!caching_ctl);
 477
 478         INIT_LIST_HEAD(&caching_ctl->list);
 479         mutex_init(&caching_ctl->mutex);
 480         init_waitqueue_head(&caching_ctl->wait);
 481         caching_ctl->block_group = cache;
 482         caching_ctl->progress = cache->key.objectid;
 483         atomic_set(&caching_ctl->count, 1);
 484         caching_ctl->work.func = caching_thread;
 485
 486         spin_lock(&cache->lock);
 487         /*
 488          * This should be a rare occasion, but this could happen I think in the
 489          * case where one thread starts to load the space cache info, and then
 490          * some other thread starts a transaction commit which tries to do an
 491          * allocation while the other thread is still loading the space cache
 492          * info.  The previous loop should have kept us from choosing this block
 493          * group, but if we've moved to the state where we will wait on caching
 494          * block groups we need to first check if we're doing a fast load here,
 495          * so we can wait for it to finish, otherwise we could end up allocating
 496          * from a block group who's cache gets evicted for one reason or
 497          * another.
 498          */
 499         while (cache->cached == BTRFS_CACHE_FAST) {
 500                 struct btrfs_caching_control *ctl;
 501
 502                 ctl = cache->caching_ctl;
 503                 atomic_inc(&ctl->count);
 504                 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
 505                 spin_unlock(&cache->lock);
 506
 507                 schedule();
 508
 509                 finish_wait(&ctl->wait, &wait);
 510                 put_caching_control(ctl);
 511                 spin_lock(&cache->lock);
 512         }
 513
 514         if (cache->cached != BTRFS_CACHE_NO) {
 515                 spin_unlock(&cache->lock);
 516                 kfree(caching_ctl);
 517                 return 0;
 518         }
 519         WARN_ON(cache->caching_ctl);
 520         cache->caching_ctl = caching_ctl;
 521         cache->cached = BTRFS_CACHE_FAST;
 522         spin_unlock(&cache->lock);
 523
 524         /*
 525          * We can't do the read from on-disk cache during a commit since we need
 526          * to have the normal tree locking.  Also if we are currently trying to
 527          * allocate blocks for the tree root we can't do the fast caching since
 528          * we likely hold important locks.
 529          */
 530         if (trans && (!trans->transaction->in_commit) &&
 531             (root && root != root->fs_info->tree_root) &&
 532             btrfs_test_opt(root, SPACE_CACHE)) {
 533                 ret = load_free_space_cache(fs_info, cache);
 534
 535                 spin_lock(&cache->lock);
 536                 if (ret == 1) {
 537                         cache->caching_ctl = NULL;
 538                         cache->cached = BTRFS_CACHE_FINISHED;
 539                         cache->last_byte_to_unpin = (u64)-1;
 540                 } else {
 541                         if (load_cache_only) {
 542                                 cache->caching_ctl = NULL;
 543                                 cache->cached = BTRFS_CACHE_NO;
 544                         } else {
 545                                 cache->cached = BTRFS_CACHE_STARTED;
 546                         }
 547                 }
 548                 spin_unlock(&cache->lock);
 549                 wake_up(&caching_ctl->wait);
 550                 if (ret == 1) {
 551                         put_caching_control(caching_ctl);
 552                         free_excluded_extents(fs_info->extent_root, cache);
 553                         return 0;
 554                 }
 555         } else {
 556                 /*
 557                  * We are not going to do the fast caching, set cached to the
 558                  * appropriate value and wakeup any waiters.
 559                  */
 560                 spin_lock(&cache->lock);
 561                 if (load_cache_only) {
 562                         cache->caching_ctl = NULL;
 563                         cache->cached = BTRFS_CACHE_NO;
 564                 } else {
 565                         cache->cached = BTRFS_CACHE_STARTED;
 566                 }
 567                 spin_unlock(&cache->lock);
 568                 wake_up(&caching_ctl->wait);
 569         }
 570
 571         if (load_cache_only) {
 572                 put_caching_control(caching_ctl);
 573                 return 0;
 574         }
 575
 576         down_write(&fs_info->extent_commit_sem);
 577         atomic_inc(&caching_ctl->count);
 578         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 579         up_write(&fs_info->extent_commit_sem);
 580
 581         btrfs_get_block_group(cache);
 582
 583         btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
 584
 585         return ret;
 586 }
 587
 588 /*
 589  * return the block group that starts at or after bytenr
 590  */
 591 static struct btrfs_block_group_cache *
 592 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 593 {
 594         struct btrfs_block_group_cache *cache;
 595
 596         cache = block_group_cache_tree_search(info, bytenr, 0);
 597
 598         return cache;
 599 }
 600
 601 /*
 602  * return the block group that contains the given bytenr
 603  */
 604 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 605                                                  struct btrfs_fs_info *info,
 606                                                  u64 bytenr)
 607 {
 608         struct btrfs_block_group_cache *cache;
 609
 610         cache = block_group_cache_tree_search(info, bytenr, 1);
 611
 612         return cache;
 613 }
 614
 615 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 616                                                   u64 flags)
 617 {
 618         struct list_head *head = &info->space_info;
 619         struct btrfs_space_info *found;
 620
 621         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 622                  BTRFS_BLOCK_GROUP_METADATA;
 623
 624         rcu_read_lock();
 625         list_for_each_entry_rcu(found, head, list) {
 626                 if (found->flags & flags) {
 627                         rcu_read_unlock();
 628                         return found;
 629                 }
 630         }
 631         rcu_read_unlock();
 632         return NULL;
 633 }
 634
 635 /*
 636  * after adding space to the filesystem, we need to clear the full flags
 637  * on all the space infos.
 638  */
 639 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 640 {
 641         struct list_head *head = &info->space_info;
 642         struct btrfs_space_info *found;
 643
 644         rcu_read_lock();
 645         list_for_each_entry_rcu(found, head, list)
 646                 found->full = 0;
 647         rcu_read_unlock();
 648 }
 649
 650 static u64 div_factor(u64 num, int factor)
 651 {
 652         if (factor == 10)
 653                 return num;
 654         num *= factor;
 655         do_div(num, 10);
 656         return num;
 657 }
 658
 659 static u64 div_factor_fine(u64 num, int factor)
 660 {
 661         if (factor == 100)
 662                 return num;
 663         num *= factor;
 664         do_div(num, 100);
 665         return num;
 666 }
 667
 668 u64 btrfs_find_block_group(struct btrfs_root *root,
 669                            u64 search_start, u64 search_hint, int owner)
 670 {
 671         struct btrfs_block_group_cache *cache;
 672         u64 used;
 673         u64 last = max(search_hint, search_start);
 674         u64 group_start = 0;
 675         int full_search = 0;
 676         int factor = 9;
 677         int wrapped = 0;
 678 again:
 679         while (1) {
 680                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 681                 if (!cache)
 682                         break;
 683
 684                 spin_lock(&cache->lock);
 685                 last = cache->key.objectid + cache->key.offset;
 686                 used = btrfs_block_group_used(&cache->item);
 687
 688                 if ((full_search || !cache->ro) &&
 689                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 690                         if (used + cache->pinned + cache->reserved <
 691                             div_factor(cache->key.offset, factor)) {
 692                                 group_start = cache->key.objectid;
 693                                 spin_unlock(&cache->lock);
 694                                 btrfs_put_block_group(cache);
 695                                 goto found;
 696                         }
 697                 }
 698                 spin_unlock(&cache->lock);
 699                 btrfs_put_block_group(cache);
 700                 cond_resched();
 701         }
 702         if (!wrapped) {
 703                 last = search_start;
 704                 wrapped = 1;
 705                 goto again;
 706         }
 707         if (!full_search && factor < 10) {
 708                 last = search_start;
 709                 full_search = 1;
 710                 factor = 10;
 711                 goto again;
 712         }
 713 found:
 714         return group_start;
 715 }
 716
 717 /* simple helper to search for an existing extent at a given offset */
 718 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 719 {
 720         int ret;
 721         struct btrfs_key key;
 722         struct btrfs_path *path;
 723
 724         path = btrfs_alloc_path();
 725         if (!path)
 726                 return -ENOMEM;
 727
 728         key.objectid = start;
 729         key.offset = len;
 730         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 731         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 732                                 0, 0);
 733         btrfs_free_path(path);
 734         return ret;
 735 }
 736
 737 /*
 738  * helper function to lookup reference count and flags of extent.
 739  *
 740  * the head node for delayed ref is used to store the sum of all the
 741  * reference count modifications queued up in the rbtree. the head
 742  * node may also store the extent flags to set. This way you can check
 743  * to see what the reference count and extent flags would be if all of
 744  * the delayed refs are not processed.
 745  */
 746 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 747                              struct btrfs_root *root, u64 bytenr,
 748                              u64 num_bytes, u64 *refs, u64 *flags)
 749 {
 750         struct btrfs_delayed_ref_head *head;
 751         struct btrfs_delayed_ref_root *delayed_refs;
 752         struct btrfs_path *path;
 753         struct btrfs_extent_item *ei;
 754         struct extent_buffer *leaf;
 755         struct btrfs_key key;
 756         u32 item_size;
 757         u64 num_refs;
 758         u64 extent_flags;
 759         int ret;
 760
 761         path = btrfs_alloc_path();
 762         if (!path)
 763                 return -ENOMEM;
 764
 765         key.objectid = bytenr;
 766         key.type = BTRFS_EXTENT_ITEM_KEY;
 767         key.offset = num_bytes;
 768         if (!trans) {
 769                 path->skip_locking = 1;
 770                 path->search_commit_root = 1;
 771         }
 772 again:
 773         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 774                                 &key, path, 0, 0);
 775         if (ret < 0)
 776                 goto out_free;
 777
 778         if (ret == 0) {
 779                 leaf = path->nodes[0];
 780                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 781                 if (item_size >= sizeof(*ei)) {
 782                         ei = btrfs_item_ptr(leaf, path->slots[0],
 783                                             struct btrfs_extent_item);
 784                         num_refs = btrfs_extent_refs(leaf, ei);
 785                         extent_flags = btrfs_extent_flags(leaf, ei);
 786                 } else {
 787 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 788                         struct btrfs_extent_item_v0 *ei0;
 789                         BUG_ON(item_size != sizeof(*ei0));
 790                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 791                                              struct btrfs_extent_item_v0);
 792                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 793                         /* FIXME: this isn't correct for data */
 794                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 795 #else
 796                         BUG();
 797 #endif
 798                 }
 799                 BUG_ON(num_refs == 0);
 800         } else {
 801                 num_refs = 0;
 802                 extent_flags = 0;
 803                 ret = 0;
 804         }
 805
 806         if (!trans)
 807                 goto out;
 808
 809         delayed_refs = &trans->transaction->delayed_refs;
 810         spin_lock(&delayed_refs->lock);
 811         head = btrfs_find_delayed_ref_head(trans, bytenr);
 812         if (head) {
 813                 if (!mutex_trylock(&head->mutex)) {
 814                         atomic_inc(&head->node.refs);
 815                         spin_unlock(&delayed_refs->lock);
 816
 817                         btrfs_release_path(path);
 818
 819                         /*
 820                          * Mutex was contended, block until it's released and try
 821                          * again
 822                          */
 823                         mutex_lock(&head->mutex);
 824                         mutex_unlock(&head->mutex);
 825                         btrfs_put_delayed_ref(&head->node);
 826                         goto again;
 827                 }
 828                 if (head->extent_op && head->extent_op->update_flags)
 829                         extent_flags |= head->extent_op->flags_to_set;
 830                 else
 831                         BUG_ON(num_refs == 0);
 832
 833                 num_refs += head->node.ref_mod;
 834                 mutex_unlock(&head->mutex);
 835         }
 836         spin_unlock(&delayed_refs->lock);
 837 out:
 838         WARN_ON(num_refs == 0);
 839         if (refs)
 840                 *refs = num_refs;
 841         if (flags)
 842                 *flags = extent_flags;
 843 out_free:
 844         btrfs_free_path(path);
 845         return ret;
 846 }
 847
 848 /*
 849  * Back reference rules.  Back refs have three main goals:
 850  *
 851  * 1) differentiate between all holders of references to an extent so that
 852  *    when a reference is dropped we can make sure it was a valid reference
 853  *    before freeing the extent.
 854  *
 855  * 2) Provide enough information to quickly find the holders of an extent
 856  *    if we notice a given block is corrupted or bad.
 857  *
 858  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 859  *    maintenance.  This is actually the same as #2, but with a slightly
 860  *    different use case.
 861  *
 862  * There are two kinds of back refs. The implicit back refs is optimized
 863  * for pointers in non-shared tree blocks. For a given pointer in a block,
 864  * back refs of this kind provide information about the block's owner tree
 865  * and the pointer's key. These information allow us to find the block by
 866  * b-tree searching. The full back refs is for pointers in tree blocks not
 867  * referenced by their owner trees. The location of tree block is recorded
 868  * in the back refs. Actually the full back refs is generic, and can be
 869  * used in all cases the implicit back refs is used. The major shortcoming
 870  * of the full back refs is its overhead. Every time a tree block gets
 871  * COWed, we have to update back refs entry for all pointers in it.
 872  *
 873  * For a newly allocated tree block, we use implicit back refs for
 874  * pointers in it. This means most tree related operations only involve
 875  * implicit back refs. For a tree block created in old transaction, the
 876  * only way to drop a reference to it is COW it. So we can detect the
 877  * event that tree block loses its owner tree's reference and do the
 878  * back refs conversion.
 879  *
 880  * When a tree block is COW'd through a tree, there are four cases:
 881  *
 882  * The reference count of the block is one and the tree is the block's
 883  * owner tree. Nothing to do in this case.
 884  *
 885  * The reference count of the block is one and the tree is not the
 886  * block's owner tree. In this case, full back refs is used for pointers
 887  * in the block. Remove these full back refs, add implicit back refs for
 888  * every pointers in the new block.
 889  *
 890  * The reference count of the block is greater than one and the tree is
 891  * the block's owner tree. In this case, implicit back refs is used for
 892  * pointers in the block. Add full back refs for every pointers in the
 893  * block, increase lower level extents' reference counts. The original
 894  * implicit back refs are entailed to the new block.
 895  *
 896  * The reference count of the block is greater than one and the tree is
 897  * not the block's owner tree. Add implicit back refs for every pointer in
 898  * the new block, increase lower level extents' reference count.
 899  *
 900  * Back Reference Key composing:
 901  *
 902  * The key objectid corresponds to the first byte in the extent,
 903  * The key type is used to differentiate between types of back refs.
 904  * There are different meanings of the key offset for different types
 905  * of back refs.
 906  *
 907  * File extents can be referenced by:
 908  *
 909  * - multiple snapshots, subvolumes, or different generations in one subvol
 910  * - different files inside a single subvolume
 911  * - different offsets inside a file (bookend extents in file.c)
 912  *
 913  * The extent ref structure for the implicit back refs has fields for:
 914  *
 915  * - Objectid of the subvolume root
 916  * - objectid of the file holding the reference
 917  * - original offset in the file
 918  * - how many bookend extents
 919  *
 920  * The key offset for the implicit back refs is hash of the first
 921  * three fields.
 922  *
 923  * The extent ref structure for the full back refs has field for:
 924  *
 925  * - number of pointers in the tree leaf
 926  *
 927  * The key offset for the implicit back refs is the first byte of
 928  * the tree leaf
 929  *
 930  * When a file extent is allocated, The implicit back refs is used.
 931  * the fields are filled in:
 932  *
 933  *     (root_key.objectid, inode objectid, offset in file, 1)
 934  *
 935  * When a file extent is removed file truncation, we find the
 936  * corresponding implicit back refs and check the following fields:
 937  *
 938  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 939  *
 940  * Btree extents can be referenced by:
 941  *
 942  * - Different subvolumes
 943  *
 944  * Both the implicit back refs and the full back refs for tree blocks
 945  * only consist of key. The key offset for the implicit back refs is
 946  * objectid of block's owner tree. The key offset for the full back refs
 947  * is the first byte of parent block.
 948  *
 949  * When implicit back refs is used, information about the lowest key and
 950  * level of the tree block are required. These information are stored in
 951  * tree block info structure.
 952  */
 953
 954 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 955 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 956                                   struct btrfs_root *root,
 957                                   struct btrfs_path *path,
 958                                   u64 owner, u32 extra_size)
 959 {
 960         struct btrfs_extent_item *item;
 961         struct btrfs_extent_item_v0 *ei0;
 962         struct btrfs_extent_ref_v0 *ref0;
 963         struct btrfs_tree_block_info *bi;
 964         struct extent_buffer *leaf;
 965         struct btrfs_key key;
 966         struct btrfs_key found_key;
 967         u32 new_size = sizeof(*item);
 968         u64 refs;
 969         int ret;
 970
 971         leaf = path->nodes[0];
 972         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 973
 974         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 975         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 976                              struct btrfs_extent_item_v0);
 977         refs = btrfs_extent_refs_v0(leaf, ei0);
 978
 979         if (owner == (u64)-1) {
 980                 while (1) {
 981                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 982                                 ret = btrfs_next_leaf(root, path);
 983                                 if (ret < 0)
 984                                         return ret;
 985                                 BUG_ON(ret > 0);
 986                                 leaf = path->nodes[0];
 987                         }
 988                         btrfs_item_key_to_cpu(leaf, &found_key,
 989                                               path->slots[0]);
 990                         BUG_ON(key.objectid != found_key.objectid);
 991                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 992                                 path->slots[0]++;
 993                                 continue;
 994                         }
 995                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 996                                               struct btrfs_extent_ref_v0);
 997                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 998                         break;
 999                 }
1000         }
1001         btrfs_release_path(path);
1002
1003         if (owner < BTRFS_FIRST_FREE_OBJECTID)
1004                 new_size += sizeof(*bi);
1005
1006         new_size -= sizeof(*ei0);
1007         ret = btrfs_search_slot(trans, root, &key, path,
1008                                 new_size + extra_size, 1);
1009         if (ret < 0)
1010                 return ret;
1011         BUG_ON(ret);
1012
1013         ret = btrfs_extend_item(trans, root, path, new_size);
1014
1015         leaf = path->nodes[0];
1016         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1017         btrfs_set_extent_refs(leaf, item, refs);
1018         /* FIXME: get real generation */
1019         btrfs_set_extent_generation(leaf, item, 0);
1020         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1021                 btrfs_set_extent_flags(leaf, item,
1022                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
1023                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
1024                 bi = (struct btrfs_tree_block_info *)(item + 1);
1025                 /* FIXME: get first key of the block */
1026                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1027                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1028         } else {
1029                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1030         }
1031         btrfs_mark_buffer_dirty(leaf);
1032         return 0;
1033 }
1034 #endif
1035
1036 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1037 {
1038         u32 high_crc = ~(u32)0;
1039         u32 low_crc = ~(u32)0;
1040         __le64 lenum;
1041
1042         lenum = cpu_to_le64(root_objectid);
1043         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1044         lenum = cpu_to_le64(owner);
1045         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1046         lenum = cpu_to_le64(offset);
1047         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1048
1049         return ((u64)high_crc << 31) ^ (u64)low_crc;
1050 }
1051
1052 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1053                                      struct btrfs_extent_data_ref *ref)
1054 {
1055         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1056                                     btrfs_extent_data_ref_objectid(leaf, ref),
1057                                     btrfs_extent_data_ref_offset(leaf, ref));
1058 }
1059
1060 static int match_extent_data_ref(struct extent_buffer *leaf,
1061                                  struct btrfs_extent_data_ref *ref,
1062                                  u64 root_objectid, u64 owner, u64 offset)
1063 {
1064         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1065             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1066             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1067                 return 0;
1068         return 1;
1069 }
1070
1071 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1072                                            struct btrfs_root *root,
1073                                            struct btrfs_path *path,
1074                                            u64 bytenr, u64 parent,
1075                                            u64 root_objectid,
1076                                            u64 owner, u64 offset)
1077 {
1078         struct btrfs_key key;
1079         struct btrfs_extent_data_ref *ref;
1080         struct extent_buffer *leaf;
1081         u32 nritems;
1082         int ret;
1083         int recow;
1084         int err = -ENOENT;
1085
1086         key.objectid = bytenr;
1087         if (parent) {
1088                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1089                 key.offset = parent;
1090         } else {
1091                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1092                 key.offset = hash_extent_data_ref(root_objectid,
1093                                                   owner, offset);
1094         }
1095 again:
1096         recow = 0;
1097         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1098         if (ret < 0) {
1099                 err = ret;
1100                 goto fail;
1101         }
1102
1103         if (parent) {
1104                 if (!ret)
1105                         return 0;
1106 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1107                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1108                 btrfs_release_path(path);
1109                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1110                 if (ret < 0) {
1111                         err = ret;
1112                         goto fail;
1113                 }
1114                 if (!ret)
1115                         return 0;
1116 #endif
1117                 goto fail;
1118         }
1119
1120         leaf = path->nodes[0];
1121         nritems = btrfs_header_nritems(leaf);
1122         while (1) {
1123                 if (path->slots[0] >= nritems) {
1124                         ret = btrfs_next_leaf(root, path);
1125                         if (ret < 0)
1126                                 err = ret;
1127                         if (ret)
1128                                 goto fail;
1129
1130                         leaf = path->nodes[0];
1131                         nritems = btrfs_header_nritems(leaf);
1132                         recow = 1;
1133                 }
1134
1135                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1136                 if (key.objectid != bytenr ||
1137                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1138                         goto fail;
1139
1140                 ref = btrfs_item_ptr(leaf, path->slots[0],
1141                                      struct btrfs_extent_data_ref);
1142
1143                 if (match_extent_data_ref(leaf, ref, root_objectid,
1144                                           owner, offset)) {
1145                         if (recow) {
1146                                 btrfs_release_path(path);
1147                                 goto again;
1148                         }
1149                         err = 0;
1150                         break;
1151                 }
1152                 path->slots[0]++;
1153         }
1154 fail:
1155         return err;
1156 }
1157
1158 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1159                                            struct btrfs_root *root,
1160                                            struct btrfs_path *path,
1161                                            u64 bytenr, u64 parent,
1162                                            u64 root_objectid, u64 owner,
1163                                            u64 offset, int refs_to_add)
1164 {
1165         struct btrfs_key key;
1166         struct extent_buffer *leaf;
1167         u32 size;
1168         u32 num_refs;
1169         int ret;
1170
1171         key.objectid = bytenr;
1172         if (parent) {
1173                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1174                 key.offset = parent;
1175                 size = sizeof(struct btrfs_shared_data_ref);
1176         } else {
1177                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1178                 key.offset = hash_extent_data_ref(root_objectid,
1179                                                   owner, offset);
1180                 size = sizeof(struct btrfs_extent_data_ref);
1181         }
1182
1183         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1184         if (ret && ret != -EEXIST)
1185                 goto fail;
1186
1187         leaf = path->nodes[0];
1188         if (parent) {
1189                 struct btrfs_shared_data_ref *ref;
1190                 ref = btrfs_item_ptr(leaf, path->slots[0],
1191                                      struct btrfs_shared_data_ref);
1192                 if (ret == 0) {
1193                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1194                 } else {
1195                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1196                         num_refs += refs_to_add;
1197                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1198                 }
1199         } else {
1200                 struct btrfs_extent_data_ref *ref;
1201                 while (ret == -EEXIST) {
1202                         ref = btrfs_item_ptr(leaf, path->slots[0],
1203                                              struct btrfs_extent_data_ref);
1204                         if (match_extent_data_ref(leaf, ref, root_objectid,
1205                                                   owner, offset))
1206                                 break;
1207                         btrfs_release_path(path);
1208                         key.offset++;
1209                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1210                                                       size);
1211                         if (ret && ret != -EEXIST)
1212                                 goto fail;
1213
1214                         leaf = path->nodes[0];
1215                 }
1216                 ref = btrfs_item_ptr(leaf, path->slots[0],
1217                                      struct btrfs_extent_data_ref);
1218                 if (ret == 0) {
1219                         btrfs_set_extent_data_ref_root(leaf, ref,
1220                                                        root_objectid);
1221                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1222                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1223                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1224                 } else {
1225                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1226                         num_refs += refs_to_add;
1227                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1228                 }
1229         }
1230         btrfs_mark_buffer_dirty(leaf);
1231         ret = 0;
1232 fail:
1233         btrfs_release_path(path);
1234         return ret;
1235 }
1236
1237 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1238                                            struct btrfs_root *root,
1239                                            struct btrfs_path *path,
1240                                            int refs_to_drop)
1241 {
1242         struct btrfs_key key;
1243         struct btrfs_extent_data_ref *ref1 = NULL;
1244         struct btrfs_shared_data_ref *ref2 = NULL;
1245         struct extent_buffer *leaf;
1246         u32 num_refs = 0;
1247         int ret = 0;
1248
1249         leaf = path->nodes[0];
1250         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1251
1252         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1253                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1254                                       struct btrfs_extent_data_ref);
1255                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1256         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1257                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1258                                       struct btrfs_shared_data_ref);
1259                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1260 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1261         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1262                 struct btrfs_extent_ref_v0 *ref0;
1263                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1264                                       struct btrfs_extent_ref_v0);
1265                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1266 #endif
1267         } else {
1268                 BUG();
1269         }
1270
1271         BUG_ON(num_refs < refs_to_drop);
1272         num_refs -= refs_to_drop;
1273
1274         if (num_refs == 0) {
1275                 ret = btrfs_del_item(trans, root, path);
1276         } else {
1277                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1278                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1279                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1280                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1281 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1282                 else {
1283                         struct btrfs_extent_ref_v0 *ref0;
1284                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1285                                         struct btrfs_extent_ref_v0);
1286                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1287                 }
1288 #endif
1289                 btrfs_mark_buffer_dirty(leaf);
1290         }
1291         return ret;
1292 }
1293
1294 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1295                                           struct btrfs_path *path,
1296                                           struct btrfs_extent_inline_ref *iref)
1297 {
1298         struct btrfs_key key;
1299         struct extent_buffer *leaf;
1300         struct btrfs_extent_data_ref *ref1;
1301         struct btrfs_shared_data_ref *ref2;
1302         u32 num_refs = 0;
1303
1304         leaf = path->nodes[0];
1305         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1306         if (iref) {
1307                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1308                     BTRFS_EXTENT_DATA_REF_KEY) {
1309                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1310                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1311                 } else {
1312                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1313                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1314                 }
1315         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1316                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1317                                       struct btrfs_extent_data_ref);
1318                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1319         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1320                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1321                                       struct btrfs_shared_data_ref);
1322                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1323 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1324         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1325                 struct btrfs_extent_ref_v0 *ref0;
1326                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1327                                       struct btrfs_extent_ref_v0);
1328                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1329 #endif
1330         } else {
1331                 WARN_ON(1);
1332         }
1333         return num_refs;
1334 }
1335
1336 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1337                                           struct btrfs_root *root,
1338                                           struct btrfs_path *path,
1339                                           u64 bytenr, u64 parent,
1340                                           u64 root_objectid)
1341 {
1342         struct btrfs_key key;
1343         int ret;
1344
1345         key.objectid = bytenr;
1346         if (parent) {
1347                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1348                 key.offset = parent;
1349         } else {
1350                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1351                 key.offset = root_objectid;
1352         }
1353
1354         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1355         if (ret > 0)
1356                 ret = -ENOENT;
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358         if (ret == -ENOENT && parent) {
1359                 btrfs_release_path(path);
1360                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1361                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1362                 if (ret > 0)
1363                         ret = -ENOENT;
1364         }
1365 #endif
1366         return ret;
1367 }
1368
1369 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1370                                           struct btrfs_root *root,
1371                                           struct btrfs_path *path,
1372                                           u64 bytenr, u64 parent,
1373                                           u64 root_objectid)
1374 {
1375         struct btrfs_key key;
1376         int ret;
1377
1378         key.objectid = bytenr;
1379         if (parent) {
1380                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1381                 key.offset = parent;
1382         } else {
1383                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1384                 key.offset = root_objectid;
1385         }
1386
1387         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1388         btrfs_release_path(path);
1389         return ret;
1390 }
1391
1392 static inline int extent_ref_type(u64 parent, u64 owner)
1393 {
1394         int type;
1395         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1396                 if (parent > 0)
1397                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1398                 else
1399                         type = BTRFS_TREE_BLOCK_REF_KEY;
1400         } else {
1401                 if (parent > 0)
1402                         type = BTRFS_SHARED_DATA_REF_KEY;
1403                 else
1404                         type = BTRFS_EXTENT_DATA_REF_KEY;
1405         }
1406         return type;
1407 }
1408
1409 static int find_next_key(struct btrfs_path *path, int level,
1410                          struct btrfs_key *key)
1411
1412 {
1413         for (; level < BTRFS_MAX_LEVEL; level++) {
1414                 if (!path->nodes[level])
1415                         break;
1416                 if (path->slots[level] + 1 >=
1417                     btrfs_header_nritems(path->nodes[level]))
1418                         continue;
1419                 if (level == 0)
1420                         btrfs_item_key_to_cpu(path->nodes[level], key,
1421                                               path->slots[level] + 1);
1422                 else
1423                         btrfs_node_key_to_cpu(path->nodes[level], key,
1424                                               path->slots[level] + 1);
1425                 return 0;
1426         }
1427         return 1;
1428 }
1429
1430 /*
1431  * look for inline back ref. if back ref is found, *ref_ret is set
1432  * to the address of inline back ref, and 0 is returned.
1433  *
1434  * if back ref isn't found, *ref_ret is set to the address where it
1435  * should be inserted, and -ENOENT is returned.
1436  *
1437  * if insert is true and there are too many inline back refs, the path
1438  * points to the extent item, and -EAGAIN is returned.
1439  *
1440  * NOTE: inline back refs are ordered in the same way that back ref
1441  *       items in the tree are ordered.
1442  */
1443 static noinline_for_stack
1444 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1445                                  struct btrfs_root *root,
1446                                  struct btrfs_path *path,
1447                                  struct btrfs_extent_inline_ref **ref_ret,
1448                                  u64 bytenr, u64 num_bytes,
1449                                  u64 parent, u64 root_objectid,
1450                                  u64 owner, u64 offset, int insert)
1451 {
1452         struct btrfs_key key;
1453         struct extent_buffer *leaf;
1454         struct btrfs_extent_item *ei;
1455         struct btrfs_extent_inline_ref *iref;
1456         u64 flags;
1457         u64 item_size;
1458         unsigned long ptr;
1459         unsigned long end;
1460         int extra_size;
1461         int type;
1462         int want;
1463         int ret;
1464         int err = 0;
1465
1466         key.objectid = bytenr;
1467         key.type = BTRFS_EXTENT_ITEM_KEY;
1468         key.offset = num_bytes;
1469
1470         want = extent_ref_type(parent, owner);
1471         if (insert) {
1472                 extra_size = btrfs_extent_inline_ref_size(want);
1473                 path->keep_locks = 1;
1474         } else
1475                 extra_size = -1;
1476         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1477         if (ret < 0) {
1478                 err = ret;
1479                 goto out;
1480         }
1481         BUG_ON(ret);
1482
1483         leaf = path->nodes[0];
1484         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1485 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1486         if (item_size < sizeof(*ei)) {
1487                 if (!insert) {
1488                         err = -ENOENT;
1489                         goto out;
1490                 }
1491                 ret = convert_extent_item_v0(trans, root, path, owner,
1492                                              extra_size);
1493                 if (ret < 0) {
1494                         err = ret;
1495                         goto out;
1496                 }
1497                 leaf = path->nodes[0];
1498                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1499         }
1500 #endif
1501         BUG_ON(item_size < sizeof(*ei));
1502
1503         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1504         flags = btrfs_extent_flags(leaf, ei);
1505
1506         ptr = (unsigned long)(ei + 1);
1507         end = (unsigned long)ei + item_size;
1508
1509         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1510                 ptr += sizeof(struct btrfs_tree_block_info);
1511                 BUG_ON(ptr > end);
1512         } else {
1513                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1514         }
1515
1516         err = -ENOENT;
1517         while (1) {
1518                 if (ptr >= end) {
1519                         WARN_ON(ptr > end);
1520                         break;
1521                 }
1522                 iref = (struct btrfs_extent_inline_ref *)ptr;
1523                 type = btrfs_extent_inline_ref_type(leaf, iref);
1524                 if (want < type)
1525                         break;
1526                 if (want > type) {
1527                         ptr += btrfs_extent_inline_ref_size(type);
1528                         continue;
1529                 }
1530
1531                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1532                         struct btrfs_extent_data_ref *dref;
1533                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1534                         if (match_extent_data_ref(leaf, dref, root_objectid,
1535                                                   owner, offset)) {
1536                                 err = 0;
1537                                 break;
1538                         }
1539                         if (hash_extent_data_ref_item(leaf, dref) <
1540                             hash_extent_data_ref(root_objectid, owner, offset))
1541                                 break;
1542                 } else {
1543                         u64 ref_offset;
1544                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1545                         if (parent > 0) {
1546                                 if (parent == ref_offset) {
1547                                         err = 0;
1548                                         break;
1549                                 }
1550                                 if (ref_offset < parent)
1551                                         break;
1552                         } else {
1553                                 if (root_objectid == ref_offset) {
1554                                         err = 0;
1555                                         break;
1556                                 }
1557                                 if (ref_offset < root_objectid)
1558                                         break;
1559                         }
1560                 }
1561                 ptr += btrfs_extent_inline_ref_size(type);
1562         }
1563         if (err == -ENOENT && insert) {
1564                 if (item_size + extra_size >=
1565                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1566                         err = -EAGAIN;
1567                         goto out;
1568                 }
1569                 /*
1570                  * To add new inline back ref, we have to make sure
1571                  * there is no corresponding back ref item.
1572                  * For simplicity, we just do not add new inline back
1573                  * ref if there is any kind of item for this block
1574                  */
1575                 if (find_next_key(path, 0, &key) == 0 &&
1576                     key.objectid == bytenr &&
1577                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1578                         err = -EAGAIN;
1579                         goto out;
1580                 }
1581         }
1582         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1583 out:
1584         if (insert) {
1585                 path->keep_locks = 0;
1586                 btrfs_unlock_up_safe(path, 1);
1587         }
1588         return err;
1589 }
1590
1591 /*
1592  * helper to add new inline back ref
1593  */
1594 static noinline_for_stack
1595 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1596                                 struct btrfs_root *root,
1597                                 struct btrfs_path *path,
1598                                 struct btrfs_extent_inline_ref *iref,
1599                                 u64 parent, u64 root_objectid,
1600                                 u64 owner, u64 offset, int refs_to_add,
1601                                 struct btrfs_delayed_extent_op *extent_op)
1602 {
1603         struct extent_buffer *leaf;
1604         struct btrfs_extent_item *ei;
1605         unsigned long ptr;
1606         unsigned long end;
1607         unsigned long item_offset;
1608         u64 refs;
1609         int size;
1610         int type;
1611         int ret;
1612
1613         leaf = path->nodes[0];
1614         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1615         item_offset = (unsigned long)iref - (unsigned long)ei;
1616
1617         type = extent_ref_type(parent, owner);
1618         size = btrfs_extent_inline_ref_size(type);
1619
1620         ret = btrfs_extend_item(trans, root, path, size);
1621
1622         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1623         refs = btrfs_extent_refs(leaf, ei);
1624         refs += refs_to_add;
1625         btrfs_set_extent_refs(leaf, ei, refs);
1626         if (extent_op)
1627                 __run_delayed_extent_op(extent_op, leaf, ei);
1628
1629         ptr = (unsigned long)ei + item_offset;
1630         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1631         if (ptr < end - size)
1632                 memmove_extent_buffer(leaf, ptr + size, ptr,
1633                                       end - size - ptr);
1634
1635         iref = (struct btrfs_extent_inline_ref *)ptr;
1636         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1637         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1638                 struct btrfs_extent_data_ref *dref;
1639                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1640                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1641                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1642                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1643                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1644         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1645                 struct btrfs_shared_data_ref *sref;
1646                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1647                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1648                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1649         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1650                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1651         } else {
1652                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1653         }
1654         btrfs_mark_buffer_dirty(leaf);
1655         return 0;
1656 }
1657
1658 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1659                                  struct btrfs_root *root,
1660                                  struct btrfs_path *path,
1661                                  struct btrfs_extent_inline_ref **ref_ret,
1662                                  u64 bytenr, u64 num_bytes, u64 parent,
1663                                  u64 root_objectid, u64 owner, u64 offset)
1664 {
1665         int ret;
1666
1667         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1668                                            bytenr, num_bytes, parent,
1669                                            root_objectid, owner, offset, 0);
1670         if (ret != -ENOENT)
1671                 return ret;
1672
1673         btrfs_release_path(path);
1674         *ref_ret = NULL;
1675
1676         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1677                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1678                                             root_objectid);
1679         } else {
1680                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1681                                              root_objectid, owner, offset);
1682         }
1683         return ret;
1684 }
1685
1686 /*
1687  * helper to update/remove inline back ref
1688  */
1689 static noinline_for_stack
1690 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1691                                  struct btrfs_root *root,
1692                                  struct btrfs_path *path,
1693                                  struct btrfs_extent_inline_ref *iref,
1694                                  int refs_to_mod,
1695                                  struct btrfs_delayed_extent_op *extent_op)
1696 {
1697         struct extent_buffer *leaf;
1698         struct btrfs_extent_item *ei;
1699         struct btrfs_extent_data_ref *dref = NULL;
1700         struct btrfs_shared_data_ref *sref = NULL;
1701         unsigned long ptr;
1702         unsigned long end;
1703         u32 item_size;
1704         int size;
1705         int type;
1706         int ret;
1707         u64 refs;
1708
1709         leaf = path->nodes[0];
1710         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1711         refs = btrfs_extent_refs(leaf, ei);
1712         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1713         refs += refs_to_mod;
1714         btrfs_set_extent_refs(leaf, ei, refs);
1715         if (extent_op)
1716                 __run_delayed_extent_op(extent_op, leaf, ei);
1717
1718         type = btrfs_extent_inline_ref_type(leaf, iref);
1719
1720         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1721                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1722                 refs = btrfs_extent_data_ref_count(leaf, dref);
1723         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1724                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1725                 refs = btrfs_shared_data_ref_count(leaf, sref);
1726         } else {
1727                 refs = 1;
1728                 BUG_ON(refs_to_mod != -1);
1729         }
1730
1731         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1732         refs += refs_to_mod;
1733
1734         if (refs > 0) {
1735                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1736                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1737                 else
1738                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1739         } else {
1740                 size =  btrfs_extent_inline_ref_size(type);
1741                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1742                 ptr = (unsigned long)iref;
1743                 end = (unsigned long)ei + item_size;
1744                 if (ptr + size < end)
1745                         memmove_extent_buffer(leaf, ptr, ptr + size,
1746                                               end - ptr - size);
1747                 item_size -= size;
1748                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1749         }
1750         btrfs_mark_buffer_dirty(leaf);
1751         return 0;
1752 }
1753
1754 static noinline_for_stack
1755 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1756                                  struct btrfs_root *root,
1757                                  struct btrfs_path *path,
1758                                  u64 bytenr, u64 num_bytes, u64 parent,
1759                                  u64 root_objectid, u64 owner,
1760                                  u64 offset, int refs_to_add,
1761                                  struct btrfs_delayed_extent_op *extent_op)
1762 {
1763         struct btrfs_extent_inline_ref *iref;
1764         int ret;
1765
1766         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1767                                            bytenr, num_bytes, parent,
1768                                            root_objectid, owner, offset, 1);
1769         if (ret == 0) {
1770                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1771                 ret = update_inline_extent_backref(trans, root, path, iref,
1772                                                    refs_to_add, extent_op);
1773         } else if (ret == -ENOENT) {
1774                 ret = setup_inline_extent_backref(trans, root, path, iref,
1775                                                   parent, root_objectid,
1776                                                   owner, offset, refs_to_add,
1777                                                   extent_op);
1778         }
1779         return ret;
1780 }
1781
1782 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1783                                  struct btrfs_root *root,
1784                                  struct btrfs_path *path,
1785                                  u64 bytenr, u64 parent, u64 root_objectid,
1786                                  u64 owner, u64 offset, int refs_to_add)
1787 {
1788         int ret;
1789         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1790                 BUG_ON(refs_to_add != 1);
1791                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1792                                             parent, root_objectid);
1793         } else {
1794                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1795                                              parent, root_objectid,
1796                                              owner, offset, refs_to_add);
1797         }
1798         return ret;
1799 }
1800
1801 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1802                                  struct btrfs_root *root,
1803                                  struct btrfs_path *path,
1804                                  struct btrfs_extent_inline_ref *iref,
1805                                  int refs_to_drop, int is_data)
1806 {
1807         int ret;
1808
1809         BUG_ON(!is_data && refs_to_drop != 1);
1810         if (iref) {
1811                 ret = update_inline_extent_backref(trans, root, path, iref,
1812                                                    -refs_to_drop, NULL);
1813         } else if (is_data) {
1814                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1815         } else {
1816                 ret = btrfs_del_item(trans, root, path);
1817         }
1818         return ret;
1819 }
1820
1821 static int btrfs_issue_discard(struct block_device *bdev,
1822                                 u64 start, u64 len)
1823 {
1824         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1825 }
1826
1827 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1828                                 u64 num_bytes, u64 *actual_bytes)
1829 {
1830         int ret;
1831         u64 discarded_bytes = 0;
1832         struct btrfs_bio *bbio = NULL;
1833
1834
1835         /* Tell the block device(s) that the sectors can be discarded */
1836         ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1837                               bytenr, &num_bytes, &bbio, 0);
1838         if (!ret) {
1839                 struct btrfs_bio_stripe *stripe = bbio->stripes;
1840                 int i;
1841
1842
1843                 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1844                         if (!stripe->dev->can_discard)
1845                                 continue;
1846
1847                         ret = btrfs_issue_discard(stripe->dev->bdev,
1848                                                   stripe->physical,
1849                                                   stripe->length);
1850                         if (!ret)
1851                                 discarded_bytes += stripe->length;
1852                         else if (ret != -EOPNOTSUPP)
1853                                 break;
1854
1855                         /*
1856                          * Just in case we get back EOPNOTSUPP for some reason,
1857                          * just ignore the return value so we don't screw up
1858                          * people calling discard_extent.
1859                          */
1860                         ret = 0;
1861                 }
1862                 kfree(bbio);
1863         }
1864
1865         if (actual_bytes)
1866                 *actual_bytes = discarded_bytes;
1867
1868
1869         return ret;
1870 }
1871
1872 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1873                          struct btrfs_root *root,
1874                          u64 bytenr, u64 num_bytes, u64 parent,
1875                          u64 root_objectid, u64 owner, u64 offset)
1876 {
1877         int ret;
1878         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1879                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1880
1881         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1882                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1883                                         parent, root_objectid, (int)owner,
1884                                         BTRFS_ADD_DELAYED_REF, NULL);
1885         } else {
1886                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1887                                         parent, root_objectid, owner, offset,
1888                                         BTRFS_ADD_DELAYED_REF, NULL);
1889         }
1890         return ret;
1891 }
1892
1893 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1894                                   struct btrfs_root *root,
1895                                   u64 bytenr, u64 num_bytes,
1896                                   u64 parent, u64 root_objectid,
1897                                   u64 owner, u64 offset, int refs_to_add,
1898                                   struct btrfs_delayed_extent_op *extent_op)
1899 {
1900         struct btrfs_path *path;
1901         struct extent_buffer *leaf;
1902         struct btrfs_extent_item *item;
1903         u64 refs;
1904         int ret;
1905         int err = 0;
1906
1907         path = btrfs_alloc_path();
1908         if (!path)
1909                 return -ENOMEM;
1910
1911         path->reada = 1;
1912         path->leave_spinning = 1;
1913         /* this will setup the path even if it fails to insert the back ref */
1914         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1915                                            path, bytenr, num_bytes, parent,
1916                                            root_objectid, owner, offset,
1917                                            refs_to_add, extent_op);
1918         if (ret == 0)
1919                 goto out;
1920
1921         if (ret != -EAGAIN) {
1922                 err = ret;
1923                 goto out;
1924         }
1925
1926         leaf = path->nodes[0];
1927         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1928         refs = btrfs_extent_refs(leaf, item);
1929         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1930         if (extent_op)
1931                 __run_delayed_extent_op(extent_op, leaf, item);
1932
1933         btrfs_mark_buffer_dirty(leaf);
1934         btrfs_release_path(path);
1935
1936         path->reada = 1;
1937         path->leave_spinning = 1;
1938
1939         /* now insert the actual backref */
1940         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1941                                     path, bytenr, parent, root_objectid,
1942                                     owner, offset, refs_to_add);
1943         BUG_ON(ret);
1944 out:
1945         btrfs_free_path(path);
1946         return err;
1947 }
1948
1949 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1950                                 struct btrfs_root *root,
1951                                 struct btrfs_delayed_ref_node *node,
1952                                 struct btrfs_delayed_extent_op *extent_op,
1953                                 int insert_reserved)
1954 {
1955         int ret = 0;
1956         struct btrfs_delayed_data_ref *ref;
1957         struct btrfs_key ins;
1958         u64 parent = 0;
1959         u64 ref_root = 0;
1960         u64 flags = 0;
1961
1962         ins.objectid = node->bytenr;
1963         ins.offset = node->num_bytes;
1964         ins.type = BTRFS_EXTENT_ITEM_KEY;
1965
1966         ref = btrfs_delayed_node_to_data_ref(node);
1967         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1968                 parent = ref->parent;
1969         else
1970                 ref_root = ref->root;
1971
1972         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1973                 if (extent_op) {
1974                         BUG_ON(extent_op->update_key);
1975                         flags |= extent_op->flags_to_set;
1976                 }
1977                 ret = alloc_reserved_file_extent(trans, root,
1978                                                  parent, ref_root, flags,
1979                                                  ref->objectid, ref->offset,
1980                                                  &ins, node->ref_mod);
1981         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1982                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1983                                              node->num_bytes, parent,
1984                                              ref_root, ref->objectid,
1985                                              ref->offset, node->ref_mod,
1986                                              extent_op);
1987         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1988                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1989                                           node->num_bytes, parent,
1990                                           ref_root, ref->objectid,
1991                                           ref->offset, node->ref_mod,
1992                                           extent_op);
1993         } else {
1994                 BUG();
1995         }
1996         return ret;
1997 }
1998
1999 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2000                                     struct extent_buffer *leaf,
2001                                     struct btrfs_extent_item *ei)
2002 {
2003         u64 flags = btrfs_extent_flags(leaf, ei);
2004         if (extent_op->update_flags) {
2005                 flags |= extent_op->flags_to_set;
2006                 btrfs_set_extent_flags(leaf, ei, flags);
2007         }
2008
2009         if (extent_op->update_key) {
2010                 struct btrfs_tree_block_info *bi;
2011                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2012                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2013                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2014         }
2015 }
2016
2017 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2018                                  struct btrfs_root *root,
2019                                  struct btrfs_delayed_ref_node *node,
2020                                  struct btrfs_delayed_extent_op *extent_op)
2021 {
2022         struct btrfs_key key;
2023         struct btrfs_path *path;
2024         struct btrfs_extent_item *ei;
2025         struct extent_buffer *leaf;
2026         u32 item_size;
2027         int ret;
2028         int err = 0;
2029
2030         path = btrfs_alloc_path();
2031         if (!path)
2032                 return -ENOMEM;
2033
2034         key.objectid = node->bytenr;
2035         key.type = BTRFS_EXTENT_ITEM_KEY;
2036         key.offset = node->num_bytes;
2037
2038         path->reada = 1;
2039         path->leave_spinning = 1;
2040         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2041                                 path, 0, 1);
2042         if (ret < 0) {
2043                 err = ret;
2044                 goto out;
2045         }
2046         if (ret > 0) {
2047                 err = -EIO;
2048                 goto out;
2049         }
2050
2051         leaf = path->nodes[0];
2052         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2053 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2054         if (item_size < sizeof(*ei)) {
2055                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2056                                              path, (u64)-1, 0);
2057                 if (ret < 0) {
2058                         err = ret;
2059                         goto out;
2060                 }
2061                 leaf = path->nodes[0];
2062                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2063         }
2064 #endif
2065         BUG_ON(item_size < sizeof(*ei));
2066         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2067         __run_delayed_extent_op(extent_op, leaf, ei);
2068
2069         btrfs_mark_buffer_dirty(leaf);
2070 out:
2071         btrfs_free_path(path);
2072         return err;
2073 }
2074
2075 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2076                                 struct btrfs_root *root,
2077                                 struct btrfs_delayed_ref_node *node,
2078                                 struct btrfs_delayed_extent_op *extent_op,
2079                                 int insert_reserved)
2080 {
2081         int ret = 0;
2082         struct btrfs_delayed_tree_ref *ref;
2083         struct btrfs_key ins;
2084         u64 parent = 0;
2085         u64 ref_root = 0;
2086
2087         ins.objectid = node->bytenr;
2088         ins.offset = node->num_bytes;
2089         ins.type = BTRFS_EXTENT_ITEM_KEY;
2090
2091         ref = btrfs_delayed_node_to_tree_ref(node);
2092         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2093                 parent = ref->parent;
2094         else
2095                 ref_root = ref->root;
2096
2097         BUG_ON(node->ref_mod != 1);
2098         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2099                 BUG_ON(!extent_op || !extent_op->update_flags ||
2100                        !extent_op->update_key);
2101                 ret = alloc_reserved_tree_block(trans, root,
2102                                                 parent, ref_root,
2103                                                 extent_op->flags_to_set,
2104                                                 &extent_op->key,
2105                                                 ref->level, &ins);
2106         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2107                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2108                                              node->num_bytes, parent, ref_root,
2109                                              ref->level, 0, 1, extent_op);
2110         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2111                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2112                                           node->num_bytes, parent, ref_root,
2113                                           ref->level, 0, 1, extent_op);
2114         } else {
2115                 BUG();
2116         }
2117         return ret;
2118 }
2119
2120 /* helper function to actually process a single delayed ref entry */
2121 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2122                                struct btrfs_root *root,
2123                                struct btrfs_delayed_ref_node *node,
2124                                struct btrfs_delayed_extent_op *extent_op,
2125                                int insert_reserved)
2126 {
2127         int ret;
2128         if (btrfs_delayed_ref_is_head(node)) {
2129                 struct btrfs_delayed_ref_head *head;
2130                 /*
2131                  * we've hit the end of the chain and we were supposed
2132                  * to insert this extent into the tree.  But, it got
2133                  * deleted before we ever needed to insert it, so all
2134                  * we have to do is clean up the accounting
2135                  */
2136                 BUG_ON(extent_op);
2137                 head = btrfs_delayed_node_to_head(node);
2138                 if (insert_reserved) {
2139                         btrfs_pin_extent(root, node->bytenr,
2140                                          node->num_bytes, 1);
2141                         if (head->is_data) {
2142                                 ret = btrfs_del_csums(trans, root,
2143                                                       node->bytenr,
2144                                                       node->num_bytes);
2145                                 BUG_ON(ret);
2146                         }
2147                 }
2148                 mutex_unlock(&head->mutex);
2149                 return 0;
2150         }
2151
2152         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2153             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2154                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2155                                            insert_reserved);
2156         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2157                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2158                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2159                                            insert_reserved);
2160         else
2161                 BUG();
2162         return ret;
2163 }
2164
2165 static noinline struct btrfs_delayed_ref_node *
2166 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2167 {
2168         struct rb_node *node;
2169         struct btrfs_delayed_ref_node *ref;
2170         int action = BTRFS_ADD_DELAYED_REF;
2171 again:
2172         /*
2173          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2174          * this prevents ref count from going down to zero when
2175          * there still are pending delayed ref.
2176          */
2177         node = rb_prev(&head->node.rb_node);
2178         while (1) {
2179                 if (!node)
2180                         break;
2181                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2182                                 rb_node);
2183                 if (ref->bytenr != head->node.bytenr)
2184                         break;
2185                 if (ref->action == action)
2186                         return ref;
2187                 node = rb_prev(node);
2188         }
2189         if (action == BTRFS_ADD_DELAYED_REF) {
2190                 action = BTRFS_DROP_DELAYED_REF;
2191                 goto again;
2192         }
2193         return NULL;
2194 }
2195
2196 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2197                                        struct btrfs_root *root,
2198                                        struct list_head *cluster)
2199 {
2200         struct btrfs_delayed_ref_root *delayed_refs;
2201         struct btrfs_delayed_ref_node *ref;
2202         struct btrfs_delayed_ref_head *locked_ref = NULL;
2203         struct btrfs_delayed_extent_op *extent_op;
2204         int ret;
2205         int count = 0;
2206         int must_insert_reserved = 0;
2207
2208         delayed_refs = &trans->transaction->delayed_refs;
2209         while (1) {
2210                 if (!locked_ref) {
2211                         /* pick a new head ref from the cluster list */
2212                         if (list_empty(cluster))
2213                                 break;
2214
2215                         locked_ref = list_entry(cluster->next,
2216                                      struct btrfs_delayed_ref_head, cluster);
2217
2218                         /* grab the lock that says we are going to process
2219                          * all the refs for this head */
2220                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2221
2222                         /*
2223                          * we may have dropped the spin lock to get the head
2224                          * mutex lock, and that might have given someone else
2225                          * time to free the head.  If that's true, it has been
2226                          * removed from our list and we can move on.
2227                          */
2228                         if (ret == -EAGAIN) {
2229                                 locked_ref = NULL;
2230                                 count++;
2231                                 continue;
2232                         }
2233                 }
2234
2235                 /*
2236                  * record the must insert reserved flag before we
2237                  * drop the spin lock.
2238                  */
2239                 must_insert_reserved = locked_ref->must_insert_reserved;
2240                 locked_ref->must_insert_reserved = 0;
2241
2242                 extent_op = locked_ref->extent_op;
2243                 locked_ref->extent_op = NULL;
2244
2245                 /*
2246                  * locked_ref is the head node, so we have to go one
2247                  * node back for any delayed ref updates
2248                  */
2249                 ref = select_delayed_ref(locked_ref);
2250                 if (!ref) {
2251                         /* All delayed refs have been processed, Go ahead
2252                          * and send the head node to run_one_delayed_ref,
2253                          * so that any accounting fixes can happen
2254                          */
2255                         ref = &locked_ref->node;
2256
2257                         if (extent_op && must_insert_reserved) {
2258                                 kfree(extent_op);
2259                                 extent_op = NULL;
2260                         }
2261
2262                         if (extent_op) {
2263                                 spin_unlock(&delayed_refs->lock);
2264
2265                                 ret = run_delayed_extent_op(trans, root,
2266                                                             ref, extent_op);
2267                                 BUG_ON(ret);
2268                                 kfree(extent_op);
2269
2270                                 cond_resched();
2271                                 spin_lock(&delayed_refs->lock);
2272                                 continue;
2273                         }
2274
2275                         list_del_init(&locked_ref->cluster);
2276                         locked_ref = NULL;
2277                 }
2278
2279                 ref->in_tree = 0;
2280                 rb_erase(&ref->rb_node, &delayed_refs->root);
2281                 delayed_refs->num_entries--;
2282
2283                 spin_unlock(&delayed_refs->lock);
2284
2285                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2286                                           must_insert_reserved);
2287                 BUG_ON(ret);
2288
2289                 btrfs_put_delayed_ref(ref);
2290                 kfree(extent_op);
2291                 count++;
2292
2293                 cond_resched();
2294                 spin_lock(&delayed_refs->lock);
2295         }
2296         return count;
2297 }
2298
2299 /*
2300  * this starts processing the delayed reference count updates and
2301  * extent insertions we have queued up so far.  count can be
2302  * 0, which means to process everything in the tree at the start
2303  * of the run (but not newly added entries), or it can be some target
2304  * number you'd like to process.
2305  */
2306 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2307                            struct btrfs_root *root, unsigned long count)
2308 {
2309         struct rb_node *node;
2310         struct btrfs_delayed_ref_root *delayed_refs;
2311         struct btrfs_delayed_ref_node *ref;
2312         struct list_head cluster;
2313         int ret;
2314         int run_all = count == (unsigned long)-1;
2315         int run_most = 0;
2316
2317         if (root == root->fs_info->extent_root)
2318                 root = root->fs_info->tree_root;
2319
2320         delayed_refs = &trans->transaction->delayed_refs;
2321         INIT_LIST_HEAD(&cluster);
2322 again:
2323         spin_lock(&delayed_refs->lock);
2324         if (count == 0) {
2325                 count = delayed_refs->num_entries * 2;
2326                 run_most = 1;
2327         }
2328         while (1) {
2329                 if (!(run_all || run_most) &&
2330                     delayed_refs->num_heads_ready < 64)
2331                         break;
2332
2333                 /*
2334                  * go find something we can process in the rbtree.  We start at
2335                  * the beginning of the tree, and then build a cluster
2336                  * of refs to process starting at the first one we are able to
2337                  * lock
2338                  */
2339                 ret = btrfs_find_ref_cluster(trans, &cluster,
2340                                              delayed_refs->run_delayed_start);
2341                 if (ret)
2342                         break;
2343
2344                 ret = run_clustered_refs(trans, root, &cluster);
2345                 BUG_ON(ret < 0);
2346
2347                 count -= min_t(unsigned long, ret, count);
2348
2349                 if (count == 0)
2350                         break;
2351         }
2352
2353         if (run_all) {
2354                 node = rb_first(&delayed_refs->root);
2355                 if (!node)
2356                         goto out;
2357                 count = (unsigned long)-1;
2358
2359                 while (node) {
2360                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2361                                        rb_node);
2362                         if (btrfs_delayed_ref_is_head(ref)) {
2363                                 struct btrfs_delayed_ref_head *head;
2364
2365                                 head = btrfs_delayed_node_to_head(ref);
2366                                 atomic_inc(&ref->refs);
2367
2368                                 spin_unlock(&delayed_refs->lock);
2369                                 /*
2370                                  * Mutex was contended, block until it's
2371                                  * released and try again
2372                                  */
2373                                 mutex_lock(&head->mutex);
2374                                 mutex_unlock(&head->mutex);
2375
2376                                 btrfs_put_delayed_ref(ref);
2377                                 cond_resched();
2378                                 goto again;
2379                         }
2380                         node = rb_next(node);
2381                 }
2382                 spin_unlock(&delayed_refs->lock);
2383                 schedule_timeout(1);
2384                 goto again;
2385         }
2386 out:
2387         spin_unlock(&delayed_refs->lock);
2388         return 0;
2389 }
2390
2391 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2392                                 struct btrfs_root *root,
2393                                 u64 bytenr, u64 num_bytes, u64 flags,
2394                                 int is_data)
2395 {
2396         struct btrfs_delayed_extent_op *extent_op;
2397         int ret;
2398
2399         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2400         if (!extent_op)
2401                 return -ENOMEM;
2402
2403         extent_op->flags_to_set = flags;
2404         extent_op->update_flags = 1;
2405         extent_op->update_key = 0;
2406         extent_op->is_data = is_data ? 1 : 0;
2407
2408         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2409         if (ret)
2410                 kfree(extent_op);
2411         return ret;
2412 }
2413
2414 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2415                                       struct btrfs_root *root,
2416                                       struct btrfs_path *path,
2417                                       u64 objectid, u64 offset, u64 bytenr)
2418 {
2419         struct btrfs_delayed_ref_head *head;
2420         struct btrfs_delayed_ref_node *ref;
2421         struct btrfs_delayed_data_ref *data_ref;
2422         struct btrfs_delayed_ref_root *delayed_refs;
2423         struct rb_node *node;
2424         int ret = 0;
2425
2426         ret = -ENOENT;
2427         delayed_refs = &trans->transaction->delayed_refs;
2428         spin_lock(&delayed_refs->lock);
2429         head = btrfs_find_delayed_ref_head(trans, bytenr);
2430         if (!head)
2431                 goto out;
2432
2433         if (!mutex_trylock(&head->mutex)) {
2434                 atomic_inc(&head->node.refs);
2435                 spin_unlock(&delayed_refs->lock);
2436
2437                 btrfs_release_path(path);
2438
2439                 /*
2440                  * Mutex was contended, block until it's released and let
2441                  * caller try again
2442                  */
2443                 mutex_lock(&head->mutex);
2444                 mutex_unlock(&head->mutex);
2445                 btrfs_put_delayed_ref(&head->node);
2446                 return -EAGAIN;
2447         }
2448
2449         node = rb_prev(&head->node.rb_node);
2450         if (!node)
2451                 goto out_unlock;
2452
2453         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2454
2455         if (ref->bytenr != bytenr)
2456                 goto out_unlock;
2457
2458         ret = 1;
2459         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2460                 goto out_unlock;
2461
2462         data_ref = btrfs_delayed_node_to_data_ref(ref);
2463
2464         node = rb_prev(node);
2465         if (node) {
2466                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2467                 if (ref->bytenr == bytenr)
2468                         goto out_unlock;
2469         }
2470
2471         if (data_ref->root != root->root_key.objectid ||
2472             data_ref->objectid != objectid || data_ref->offset != offset)
2473                 goto out_unlock;
2474
2475         ret = 0;
2476 out_unlock:
2477         mutex_unlock(&head->mutex);
2478 out:
2479         spin_unlock(&delayed_refs->lock);
2480         return ret;
2481 }
2482
2483 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2484                                         struct btrfs_root *root,
2485                                         struct btrfs_path *path,
2486                                         u64 objectid, u64 offset, u64 bytenr)
2487 {
2488         struct btrfs_root *extent_root = root->fs_info->extent_root;
2489         struct extent_buffer *leaf;
2490         struct btrfs_extent_data_ref *ref;
2491         struct btrfs_extent_inline_ref *iref;
2492         struct btrfs_extent_item *ei;
2493         struct btrfs_key key;
2494         u32 item_size;
2495         int ret;
2496
2497         key.objectid = bytenr;
2498         key.offset = (u64)-1;
2499         key.type = BTRFS_EXTENT_ITEM_KEY;
2500
2501         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2502         if (ret < 0)
2503                 goto out;
2504         BUG_ON(ret == 0);
2505
2506         ret = -ENOENT;
2507         if (path->slots[0] == 0)
2508                 goto out;
2509
2510         path->slots[0]--;
2511         leaf = path->nodes[0];
2512         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2513
2514         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2515                 goto out;
2516
2517         ret = 1;
2518         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2519 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2520         if (item_size < sizeof(*ei)) {
2521                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2522                 goto out;
2523         }
2524 #endif
2525         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2526
2527         if (item_size != sizeof(*ei) +
2528             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2529                 goto out;
2530
2531         if (btrfs_extent_generation(leaf, ei) <=
2532             btrfs_root_last_snapshot(&root->root_item))
2533                 goto out;
2534
2535         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2536         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2537             BTRFS_EXTENT_DATA_REF_KEY)
2538                 goto out;
2539
2540         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2541         if (btrfs_extent_refs(leaf, ei) !=
2542             btrfs_extent_data_ref_count(leaf, ref) ||
2543             btrfs_extent_data_ref_root(leaf, ref) !=
2544             root->root_key.objectid ||
2545             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2546             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2547                 goto out;
2548
2549         ret = 0;
2550 out:
2551         return ret;
2552 }
2553
2554 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2555                           struct btrfs_root *root,
2556                           u64 objectid, u64 offset, u64 bytenr)
2557 {
2558         struct btrfs_path *path;
2559         int ret;
2560         int ret2;
2561
2562         path = btrfs_alloc_path();
2563         if (!path)
2564                 return -ENOENT;
2565
2566         do {
2567                 ret = check_committed_ref(trans, root, path, objectid,
2568                                           offset, bytenr);
2569                 if (ret && ret != -ENOENT)
2570                         goto out;
2571
2572                 ret2 = check_delayed_ref(trans, root, path, objectid,
2573                                          offset, bytenr);
2574         } while (ret2 == -EAGAIN);
2575
2576         if (ret2 && ret2 != -ENOENT) {
2577                 ret = ret2;
2578                 goto out;
2579         }
2580
2581         if (ret != -ENOENT || ret2 != -ENOENT)
2582                 ret = 0;
2583 out:
2584         btrfs_free_path(path);
2585         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2586                 WARN_ON(ret > 0);
2587         return ret;
2588 }
2589
2590 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2591                            struct btrfs_root *root,
2592                            struct extent_buffer *buf,
2593                            int full_backref, int inc)
2594 {
2595         u64 bytenr;
2596         u64 num_bytes;
2597         u64 parent;
2598         u64 ref_root;
2599         u32 nritems;
2600         struct btrfs_key key;
2601         struct btrfs_file_extent_item *fi;
2602         int i;
2603         int level;
2604         int ret = 0;
2605         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2606                             u64, u64, u64, u64, u64, u64);
2607
2608         ref_root = btrfs_header_owner(buf);
2609         nritems = btrfs_header_nritems(buf);
2610         level = btrfs_header_level(buf);
2611
2612         if (!root->ref_cows && level == 0)
2613                 return 0;
2614
2615         if (inc)
2616                 process_func = btrfs_inc_extent_ref;
2617         else
2618                 process_func = btrfs_free_extent;
2619
2620         if (full_backref)
2621                 parent = buf->start;
2622         else
2623                 parent = 0;
2624
2625         for (i = 0; i < nritems; i++) {
2626                 if (level == 0) {
2627                         btrfs_item_key_to_cpu(buf, &key, i);
2628                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2629                                 continue;
2630                         fi = btrfs_item_ptr(buf, i,
2631                                             struct btrfs_file_extent_item);
2632                         if (btrfs_file_extent_type(buf, fi) ==
2633                             BTRFS_FILE_EXTENT_INLINE)
2634                                 continue;
2635                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2636                         if (bytenr == 0)
2637                                 continue;
2638
2639                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2640                         key.offset -= btrfs_file_extent_offset(buf, fi);
2641                         ret = process_func(trans, root, bytenr, num_bytes,
2642                                            parent, ref_root, key.objectid,
2643                                            key.offset);
2644                         if (ret)
2645                                 goto fail;
2646                 } else {
2647                         bytenr = btrfs_node_blockptr(buf, i);
2648                         num_bytes = btrfs_level_size(root, level - 1);
2649                         ret = process_func(trans, root, bytenr, num_bytes,
2650                                            parent, ref_root, level - 1, 0);
2651                         if (ret)
2652                                 goto fail;
2653                 }
2654         }
2655         return 0;
2656 fail:
2657         BUG();
2658         return ret;
2659 }
2660
2661 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2662                   struct extent_buffer *buf, int full_backref)
2663 {
2664         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2665 }
2666
2667 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2668                   struct extent_buffer *buf, int full_backref)
2669 {
2670         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2671 }
2672
2673 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2674                                  struct btrfs_root *root,
2675                                  struct btrfs_path *path,
2676                                  struct btrfs_block_group_cache *cache)
2677 {
2678         int ret;
2679         struct btrfs_root *extent_root = root->fs_info->extent_root;
2680         unsigned long bi;
2681         struct extent_buffer *leaf;
2682
2683         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2684         if (ret < 0)
2685                 goto fail;
2686         BUG_ON(ret);
2687
2688         leaf = path->nodes[0];
2689         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2690         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2691         btrfs_mark_buffer_dirty(leaf);
2692         btrfs_release_path(path);
2693 fail:
2694         if (ret)
2695                 return ret;
2696         return 0;
2697
2698 }
2699
2700 static struct btrfs_block_group_cache *
2701 next_block_group(struct btrfs_root *root,
2702                  struct btrfs_block_group_cache *cache)
2703 {
2704         struct rb_node *node;
2705         spin_lock(&root->fs_info->block_group_cache_lock);
2706         node = rb_next(&cache->cache_node);
2707         btrfs_put_block_group(cache);
2708         if (node) {
2709                 cache = rb_entry(node, struct btrfs_block_group_cache,
2710                                  cache_node);
2711                 btrfs_get_block_group(cache);
2712         } else
2713                 cache = NULL;
2714         spin_unlock(&root->fs_info->block_group_cache_lock);
2715         return cache;
2716 }
2717
2718 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2719                             struct btrfs_trans_handle *trans,
2720                             struct btrfs_path *path)
2721 {
2722         struct btrfs_root *root = block_group->fs_info->tree_root;
2723         struct inode *inode = NULL;
2724         u64 alloc_hint = 0;
2725         int dcs = BTRFS_DC_ERROR;
2726         int num_pages = 0;
2727         int retries = 0;
2728         int ret = 0;
2729
2730         /*
2731          * If this block group is smaller than 100 megs don't bother caching the
2732          * block group.
2733          */
2734         if (block_group->key.offset < (100 * 1024 * 1024)) {
2735                 spin_lock(&block_group->lock);
2736                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2737                 spin_unlock(&block_group->lock);
2738                 return 0;
2739         }
2740
2741 again:
2742         inode = lookup_free_space_inode(root, block_group, path);
2743         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2744                 ret = PTR_ERR(inode);
2745                 btrfs_release_path(path);
2746                 goto out;
2747         }
2748
2749         if (IS_ERR(inode)) {
2750                 BUG_ON(retries);
2751                 retries++;
2752
2753                 if (block_group->ro)
2754                         goto out_free;
2755
2756                 ret = create_free_space_inode(root, trans, block_group, path);
2757                 if (ret)
2758                         goto out_free;
2759                 goto again;
2760         }
2761
2762         /* We've already setup this transaction, go ahead and exit */
2763         if (block_group->cache_generation == trans->transid &&
2764             i_size_read(inode)) {
2765                 dcs = BTRFS_DC_SETUP;
2766                 goto out_put;
2767         }
2768
2769         /*
2770          * We want to set the generation to 0, that way if anything goes wrong
2771          * from here on out we know not to trust this cache when we load up next
2772          * time.
2773          */
2774         BTRFS_I(inode)->generation = 0;
2775         ret = btrfs_update_inode(trans, root, inode);
2776         WARN_ON(ret);
2777
2778         if (i_size_read(inode) > 0) {
2779                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2780                                                       inode);
2781                 if (ret)
2782                         goto out_put;
2783         }
2784
2785         spin_lock(&block_group->lock);
2786         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2787                 /* We're not cached, don't bother trying to write stuff out */
2788                 dcs = BTRFS_DC_WRITTEN;
2789                 spin_unlock(&block_group->lock);
2790                 goto out_put;
2791         }
2792         spin_unlock(&block_group->lock);
2793
2794         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2795         if (!num_pages)
2796                 num_pages = 1;
2797
2798         /*
2799          * Just to make absolutely sure we have enough space, we're going to
2800          * preallocate 12 pages worth of space for each block group.  In
2801          * practice we ought to use at most 8, but we need extra space so we can
2802          * add our header and have a terminator between the extents and the
2803          * bitmaps.
2804          */
2805         num_pages *= 16;
2806         num_pages *= PAGE_CACHE_SIZE;
2807
2808         ret = btrfs_check_data_free_space(inode, num_pages);
2809         if (ret)
2810                 goto out_put;
2811
2812         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2813                                               num_pages, num_pages,
2814                                               &alloc_hint);
2815         if (!ret)
2816                 dcs = BTRFS_DC_SETUP;
2817         btrfs_free_reserved_data_space(inode, num_pages);
2818
2819 out_put:
2820         iput(inode);
2821 out_free:
2822         btrfs_release_path(path);
2823 out:
2824         spin_lock(&block_group->lock);
2825         if (!ret && dcs == BTRFS_DC_SETUP)
2826                 block_group->cache_generation = trans->transid;
2827         block_group->disk_cache_state = dcs;
2828         spin_unlock(&block_group->lock);
2829
2830         return ret;
2831 }
2832
2833 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2834                                    struct btrfs_root *root)
2835 {
2836         struct btrfs_block_group_cache *cache;
2837         int err = 0;
2838         struct btrfs_path *path;
2839         u64 last = 0;
2840
2841         path = btrfs_alloc_path();
2842         if (!path)
2843                 return -ENOMEM;
2844
2845 again:
2846         while (1) {
2847                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2848                 while (cache) {
2849                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2850                                 break;
2851                         cache = next_block_group(root, cache);
2852                 }
2853                 if (!cache) {
2854                         if (last == 0)
2855                                 break;
2856                         last = 0;
2857                         continue;
2858                 }
2859                 err = cache_save_setup(cache, trans, path);
2860                 last = cache->key.objectid + cache->key.offset;
2861                 btrfs_put_block_group(cache);
2862         }
2863
2864         while (1) {
2865                 if (last == 0) {
2866                         err = btrfs_run_delayed_refs(trans, root,
2867                                                      (unsigned long)-1);
2868                         BUG_ON(err);
2869                 }
2870
2871                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2872                 while (cache) {
2873                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2874                                 btrfs_put_block_group(cache);
2875                                 goto again;
2876                         }
2877
2878                         if (cache->dirty)
2879                                 break;
2880                         cache = next_block_group(root, cache);
2881                 }
2882                 if (!cache) {
2883                         if (last == 0)
2884                                 break;
2885                         last = 0;
2886                         continue;
2887                 }
2888
2889                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2890                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2891                 cache->dirty = 0;
2892                 last = cache->key.objectid + cache->key.offset;
2893
2894                 err = write_one_cache_group(trans, root, path, cache);
2895                 BUG_ON(err);
2896                 btrfs_put_block_group(cache);
2897         }
2898
2899         while (1) {
2900                 /*
2901                  * I don't think this is needed since we're just marking our
2902                  * preallocated extent as written, but just in case it can't
2903                  * hurt.
2904                  */
2905                 if (last == 0) {
2906                         err = btrfs_run_delayed_refs(trans, root,
2907                                                      (unsigned long)-1);
2908                         BUG_ON(err);
2909                 }
2910
2911                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2912                 while (cache) {
2913                         /*
2914                          * Really this shouldn't happen, but it could if we
2915                          * couldn't write the entire preallocated extent and
2916                          * splitting the extent resulted in a new block.
2917                          */
2918                         if (cache->dirty) {
2919                                 btrfs_put_block_group(cache);
2920                                 goto again;
2921                         }
2922                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2923                                 break;
2924                         cache = next_block_group(root, cache);
2925                 }
2926                 if (!cache) {
2927                         if (last == 0)
2928                                 break;
2929                         last = 0;
2930                         continue;
2931                 }
2932
2933                 btrfs_write_out_cache(root, trans, cache, path);
2934
2935                 /*
2936                  * If we didn't have an error then the cache state is still
2937                  * NEED_WRITE, so we can set it to WRITTEN.
2938                  */
2939                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2940                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2941                 last = cache->key.objectid + cache->key.offset;
2942                 btrfs_put_block_group(cache);
2943         }
2944
2945         btrfs_free_path(path);
2946         return 0;
2947 }
2948
2949 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2950 {
2951         struct btrfs_block_group_cache *block_group;
2952         int readonly = 0;
2953
2954         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2955         if (!block_group || block_group->ro)
2956                 readonly = 1;
2957         if (block_group)
2958                 btrfs_put_block_group(block_group);
2959         return readonly;
2960 }
2961
2962 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2963                              u64 total_bytes, u64 bytes_used,
2964                              struct btrfs_space_info **space_info)
2965 {
2966         struct btrfs_space_info *found;
2967         int i;
2968         int factor;
2969
2970         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2971                      BTRFS_BLOCK_GROUP_RAID10))
2972                 factor = 2;
2973         else
2974                 factor = 1;
2975
2976         found = __find_space_info(info, flags);
2977         if (found) {
2978                 spin_lock(&found->lock);
2979                 found->total_bytes += total_bytes;
2980                 found->disk_total += total_bytes * factor;
2981                 found->bytes_used += bytes_used;
2982                 found->disk_used += bytes_used * factor;
2983                 found->full = 0;
2984                 spin_unlock(&found->lock);
2985                 *space_info = found;
2986                 return 0;
2987         }
2988         found = kzalloc(sizeof(*found), GFP_NOFS);
2989         if (!found)
2990                 return -ENOMEM;
2991
2992         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2993                 INIT_LIST_HEAD(&found->block_groups[i]);
2994         init_rwsem(&found->groups_sem);
2995         spin_lock_init(&found->lock);
2996         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2997                                 BTRFS_BLOCK_GROUP_SYSTEM |
2998                                 BTRFS_BLOCK_GROUP_METADATA);
2999         found->total_bytes = total_bytes;
3000         found->disk_total = total_bytes * factor;
3001         found->bytes_used = bytes_used;
3002         found->disk_used = bytes_used * factor;
3003         found->bytes_pinned = 0;
3004         found->bytes_reserved = 0;
3005         found->bytes_readonly = 0;
3006         found->bytes_may_use = 0;
3007         found->full = 0;
3008         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3009         found->chunk_alloc = 0;
3010         found->flush = 0;
3011         init_waitqueue_head(&found->wait);
3012         *space_info = found;
3013         list_add_rcu(&found->list, &info->space_info);
3014         return 0;
3015 }
3016
3017 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3018 {
3019         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3020                                    BTRFS_BLOCK_GROUP_RAID1 |
3021                                    BTRFS_BLOCK_GROUP_RAID10 |
3022                                    BTRFS_BLOCK_GROUP_DUP);
3023         if (extra_flags) {
3024                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3025                         fs_info->avail_data_alloc_bits |= extra_flags;
3026                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3027                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3028                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3029                         fs_info->avail_system_alloc_bits |= extra_flags;
3030         }
3031 }
3032
3033 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3034 {
3035         /*
3036          * we add in the count of missing devices because we want
3037          * to make sure that any RAID levels on a degraded FS
3038          * continue to be honored.
3039          */
3040         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3041                 root->fs_info->fs_devices->missing_devices;
3042
3043         if (num_devices == 1)
3044                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3045         if (num_devices < 4)
3046                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3047
3048         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3049             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3050                       BTRFS_BLOCK_GROUP_RAID10))) {
3051                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3052         }
3053
3054         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3055             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3056                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3057         }
3058
3059         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3060             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3061              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3062              (flags & BTRFS_BLOCK_GROUP_DUP)))
3063                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3064         return flags;
3065 }
3066
3067 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3068 {
3069         if (flags & BTRFS_BLOCK_GROUP_DATA)
3070                 flags |= root->fs_info->avail_data_alloc_bits;
3071         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3072                 flags |= root->fs_info->avail_system_alloc_bits;
3073         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3074                 flags |= root->fs_info->avail_metadata_alloc_bits;
3075
3076         return btrfs_reduce_alloc_profile(root, flags);
3077 }
3078
3079 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3080 {
3081         u64 flags;
3082
3083         if (data)
3084                 flags = BTRFS_BLOCK_GROUP_DATA;
3085         else if (root == root->fs_info->chunk_root)
3086                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3087         else
3088                 flags = BTRFS_BLOCK_GROUP_METADATA;
3089
3090         return get_alloc_profile(root, flags);
3091 }
3092
3093 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3094 {
3095         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3096                                                        BTRFS_BLOCK_GROUP_DATA);
3097 }
3098
3099 /*
3100  * This will check the space that the inode allocates from to make sure we have
3101  * enough space for bytes.
3102  */
3103 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3104 {
3105         struct btrfs_space_info *data_sinfo;
3106         struct btrfs_root *root = BTRFS_I(inode)->root;
3107         u64 used;
3108         int ret = 0, committed = 0, alloc_chunk = 1;
3109
3110         /* make sure bytes are sectorsize aligned */
3111         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3112
3113         if (root == root->fs_info->tree_root ||
3114             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3115                 alloc_chunk = 0;
3116                 committed = 1;
3117         }
3118
3119         data_sinfo = BTRFS_I(inode)->space_info;
3120         if (!data_sinfo)
3121                 goto alloc;
3122
3123 again:
3124         /* make sure we have enough space to handle the data first */
3125         spin_lock(&data_sinfo->lock);
3126         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3127                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3128                 data_sinfo->bytes_may_use;
3129
3130         if (used + bytes > data_sinfo->total_bytes) {
3131                 struct btrfs_trans_handle *trans;
3132
3133                 /*
3134                  * if we don't have enough free bytes in this space then we need
3135                  * to alloc a new chunk.
3136                  */
3137                 if (!data_sinfo->full && alloc_chunk) {
3138                         u64 alloc_target;
3139
3140                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3141                         spin_unlock(&data_sinfo->lock);
3142 alloc:
3143                         alloc_target = btrfs_get_alloc_profile(root, 1);
3144                         trans = btrfs_join_transaction(root);
3145                         if (IS_ERR(trans))
3146                                 return PTR_ERR(trans);
3147
3148                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3149                                              bytes + 2 * 1024 * 1024,
3150                                              alloc_target,
3151                                              CHUNK_ALLOC_NO_FORCE);
3152                         btrfs_end_transaction(trans, root);
3153                         if (ret < 0) {
3154                                 if (ret != -ENOSPC)
3155                                         return ret;
3156                                 else
3157                                         goto commit_trans;
3158                         }
3159
3160                         if (!data_sinfo) {
3161                                 btrfs_set_inode_space_info(root, inode);
3162                                 data_sinfo = BTRFS_I(inode)->space_info;
3163                         }
3164                         goto again;
3165                 }
3166
3167                 /*
3168                  * If we have less pinned bytes than we want to allocate then
3169                  * don't bother committing the transaction, it won't help us.
3170                  */
3171                 if (data_sinfo->bytes_pinned < bytes)
3172                         committed = 1;
3173                 spin_unlock(&data_sinfo->lock);
3174
3175                 /* commit the current transaction and try again */
3176 commit_trans:
3177                 if (!committed &&
3178                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3179                         committed = 1;
3180                         trans = btrfs_join_transaction(root);
3181                         if (IS_ERR(trans))
3182                                 return PTR_ERR(trans);
3183                         ret = btrfs_commit_transaction(trans, root);
3184                         if (ret)
3185                                 return ret;
3186                         goto again;
3187                 }
3188
3189                 return -ENOSPC;
3190         }
3191         data_sinfo->bytes_may_use += bytes;
3192         spin_unlock(&data_sinfo->lock);
3193
3194         return 0;
3195 }
3196
3197 /*
3198  * Called if we need to clear a data reservation for this inode.
3199  */
3200 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3201 {
3202         struct btrfs_root *root = BTRFS_I(inode)->root;
3203         struct btrfs_space_info *data_sinfo;
3204
3205         /* make sure bytes are sectorsize aligned */
3206         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3207
3208         data_sinfo = BTRFS_I(inode)->space_info;
3209         spin_lock(&data_sinfo->lock);
3210         data_sinfo->bytes_may_use -= bytes;
3211         spin_unlock(&data_sinfo->lock);
3212 }
3213
3214 static void force_metadata_allocation(struct btrfs_fs_info *info)
3215 {
3216         struct list_head *head = &info->space_info;
3217         struct btrfs_space_info *found;
3218
3219         rcu_read_lock();
3220         list_for_each_entry_rcu(found, head, list) {
3221                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3222                         found->force_alloc = CHUNK_ALLOC_FORCE;
3223         }
3224         rcu_read_unlock();
3225 }
3226
3227 static int should_alloc_chunk(struct btrfs_root *root,
3228                               struct btrfs_space_info *sinfo, u64 alloc_bytes,
3229                               int force)
3230 {
3231         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3232         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3233         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3234         u64 thresh;
3235
3236         if (force == CHUNK_ALLOC_FORCE)
3237                 return 1;
3238
3239         /*
3240          * We need to take into account the global rsv because for all intents
3241          * and purposes it's used space.  Don't worry about locking the
3242          * global_rsv, it doesn't change except when the transaction commits.
3243          */
3244         num_allocated += global_rsv->size;
3245
3246         /*
3247          * in limited mode, we want to have some free space up to
3248          * about 1% of the FS size.
3249          */
3250         if (force == CHUNK_ALLOC_LIMITED) {
3251                 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3252                 thresh = max_t(u64, 64 * 1024 * 1024,
3253                                div_factor_fine(thresh, 1));
3254
3255                 if (num_bytes - num_allocated < thresh)
3256                         return 1;
3257         }
3258
3259         /*
3260          * we have two similar checks here, one based on percentage
3261          * and once based on a hard number of 256MB.  The idea
3262          * is that if we have a good amount of free
3263          * room, don't allocate a chunk.  A good mount is
3264          * less than 80% utilized of the chunks we have allocated,
3265          * or more than 256MB free
3266          */
3267         if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3268                 return 0;
3269
3270         if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3271                 return 0;
3272
3273         thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3274
3275         /* 256MB or 5% of the FS */
3276         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3277
3278         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3279                 return 0;
3280         return 1;
3281 }
3282
3283 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3284                           struct btrfs_root *extent_root, u64 alloc_bytes,
3285                           u64 flags, int force)
3286 {
3287         struct btrfs_space_info *space_info;
3288         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3289         int wait_for_alloc = 0;
3290         int ret = 0;
3291
3292         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3293
3294         space_info = __find_space_info(extent_root->fs_info, flags);
3295         if (!space_info) {
3296                 ret = update_space_info(extent_root->fs_info, flags,
3297                                         0, 0, &space_info);
3298                 BUG_ON(ret);
3299         }
3300         BUG_ON(!space_info);
3301
3302 again:
3303         spin_lock(&space_info->lock);
3304         if (space_info->force_alloc)
3305                 force = space_info->force_alloc;
3306         if (space_info->full) {
3307                 spin_unlock(&space_info->lock);
3308                 return 0;
3309         }
3310
3311         if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3312                 spin_unlock(&space_info->lock);
3313                 return 0;
3314         } else if (space_info->chunk_alloc) {
3315                 wait_for_alloc = 1;
3316         } else {
3317                 space_info->chunk_alloc = 1;
3318         }
3319
3320         spin_unlock(&space_info->lock);
3321
3322         mutex_lock(&fs_info->chunk_mutex);
3323
3324         /*
3325          * The chunk_mutex is held throughout the entirety of a chunk
3326          * allocation, so once we've acquired the chunk_mutex we know that the
3327          * other guy is done and we need to recheck and see if we should
3328          * allocate.
3329          */
3330         if (wait_for_alloc) {
3331                 mutex_unlock(&fs_info->chunk_mutex);
3332                 wait_for_alloc = 0;
3333                 goto again;
3334         }
3335
3336         /*
3337          * If we have mixed data/metadata chunks we want to make sure we keep
3338          * allocating mixed chunks instead of individual chunks.
3339          */
3340         if (btrfs_mixed_space_info(space_info))
3341                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3342
3343         /*
3344          * if we're doing a data chunk, go ahead and make sure that
3345          * we keep a reasonable number of metadata chunks allocated in the
3346          * FS as well.
3347          */
3348         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3349                 fs_info->data_chunk_allocations++;
3350                 if (!(fs_info->data_chunk_allocations %
3351                       fs_info->metadata_ratio))
3352                         force_metadata_allocation(fs_info);
3353         }
3354
3355         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3356         if (ret < 0 && ret != -ENOSPC)
3357                 goto out;
3358
3359         spin_lock(&space_info->lock);
3360         if (ret)
3361                 space_info->full = 1;
3362         else
3363                 ret = 1;
3364
3365         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3366         space_info->chunk_alloc = 0;
3367         spin_unlock(&space_info->lock);
3368 out:
3369         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3370         return ret;
3371 }
3372
3373 /*
3374  * shrink metadata reservation for delalloc
3375  */
3376 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3377                            bool wait_ordered)
3378 {
3379         struct btrfs_block_rsv *block_rsv;
3380         struct btrfs_space_info *space_info;
3381         struct btrfs_trans_handle *trans;
3382         u64 reserved;
3383         u64 max_reclaim;
3384         u64 reclaimed = 0;
3385         long time_left;
3386         unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3387         int loops = 0;
3388         unsigned long progress;
3389
3390         trans = (struct btrfs_trans_handle *)current->journal_info;
3391         block_rsv = &root->fs_info->delalloc_block_rsv;
3392         space_info = block_rsv->space_info;
3393
3394         smp_mb();
3395         reserved = space_info->bytes_may_use;
3396         progress = space_info->reservation_progress;
3397
3398         if (reserved == 0)
3399                 return 0;
3400
3401         smp_mb();
3402         if (root->fs_info->delalloc_bytes == 0) {
3403                 if (trans)
3404                         return 0;
3405                 btrfs_wait_ordered_extents(root, 0, 0);
3406                 return 0;
3407         }
3408
3409         max_reclaim = min(reserved, to_reclaim);
3410         nr_pages = max_t(unsigned long, nr_pages,
3411                          max_reclaim >> PAGE_CACHE_SHIFT);
3412         while (loops < 1024) {
3413                 /* have the flusher threads jump in and do some IO */
3414                 smp_mb();
3415                 nr_pages = min_t(unsigned long, nr_pages,
3416                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3417                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3418
3419                 spin_lock(&space_info->lock);
3420                 if (reserved > space_info->bytes_may_use)
3421                         reclaimed += reserved - space_info->bytes_may_use;
3422                 reserved = space_info->bytes_may_use;
3423                 spin_unlock(&space_info->lock);
3424
3425                 loops++;
3426
3427                 if (reserved == 0 || reclaimed >= max_reclaim)
3428                         break;
3429
3430                 if (trans && trans->transaction->blocked)
3431                         return -EAGAIN;
3432
3433                 if (wait_ordered && !trans) {
3434                         btrfs_wait_ordered_extents(root, 0, 0);
3435                 } else {
3436                         time_left = schedule_timeout_interruptible(1);
3437
3438                         /* We were interrupted, exit */
3439                         if (time_left)
3440                                 break;
3441                 }
3442
3443                 /* we've kicked the IO a few times, if anything has been freed,
3444                  * exit.  There is no sense in looping here for a long time
3445                  * when we really need to commit the transaction, or there are
3446                  * just too many writers without enough free space
3447                  */
3448
3449                 if (loops > 3) {
3450                         smp_mb();
3451                         if (progress != space_info->reservation_progress)
3452                                 break;
3453                 }
3454
3455         }
3456
3457         return reclaimed >= to_reclaim;
3458 }
3459
3460 /**
3461  * maybe_commit_transaction - possibly commit the transaction if its ok to
3462  * @root - the root we're allocating for
3463  * @bytes - the number of bytes we want to reserve
3464  * @force - force the commit
3465  *
3466  * This will check to make sure that committing the transaction will actually
3467  * get us somewhere and then commit the transaction if it does.  Otherwise it
3468  * will return -ENOSPC.
3469  */
3470 static int may_commit_transaction(struct btrfs_root *root,
3471                                   struct btrfs_space_info *space_info,
3472                                   u64 bytes, int force)
3473 {
3474         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3475         struct btrfs_trans_handle *trans;
3476
3477         trans = (struct btrfs_trans_handle *)current->journal_info;
3478         if (trans)
3479                 return -EAGAIN;
3480
3481         if (force)
3482                 goto commit;
3483
3484         /* See if there is enough pinned space to make this reservation */
3485         spin_lock(&space_info->lock);
3486         if (space_info->bytes_pinned >= bytes) {
3487                 spin_unlock(&space_info->lock);
3488                 goto commit;
3489         }
3490         spin_unlock(&space_info->lock);
3491
3492         /*
3493          * See if there is some space in the delayed insertion reservation for
3494          * this reservation.
3495          */
3496         if (space_info != delayed_rsv->space_info)
3497                 return -ENOSPC;
3498
3499         spin_lock(&delayed_rsv->lock);
3500         if (delayed_rsv->size < bytes) {
3501                 spin_unlock(&delayed_rsv->lock);
3502                 return -ENOSPC;
3503         }
3504         spin_unlock(&delayed_rsv->lock);
3505
3506 commit:
3507         trans = btrfs_join_transaction(root);
3508         if (IS_ERR(trans))
3509                 return -ENOSPC;
3510
3511         return btrfs_commit_transaction(trans, root);
3512 }
3513
3514 /**
3515  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3516  * @root - the root we're allocating for
3517  * @block_rsv - the block_rsv we're allocating for
3518  * @orig_bytes - the number of bytes we want
3519  * @flush - wether or not we can flush to make our reservation
3520  *
3521  * This will reserve orgi_bytes number of bytes from the space info associated
3522  * with the block_rsv.  If there is not enough space it will make an attempt to
3523  * flush out space to make room.  It will do this by flushing delalloc if
3524  * possible or committing the transaction.  If flush is 0 then no attempts to
3525  * regain reservations will be made and this will fail if there is not enough
3526  * space already.
3527  */
3528 static int reserve_metadata_bytes(struct btrfs_root *root,
3529                                   struct btrfs_block_rsv *block_rsv,
3530                                   u64 orig_bytes, int flush)
3531 {
3532         struct btrfs_space_info *space_info = block_rsv->space_info;
3533         u64 used;
3534         u64 num_bytes = orig_bytes;
3535         int retries = 0;
3536         int ret = 0;
3537         bool committed = false;
3538         bool flushing = false;
3539         bool wait_ordered = false;
3540
3541 again:
3542         ret = 0;
3543         spin_lock(&space_info->lock);
3544         /*
3545          * We only want to wait if somebody other than us is flushing and we are
3546          * actually alloed to flush.
3547          */
3548         while (flush && !flushing && space_info->flush) {
3549                 spin_unlock(&space_info->lock);
3550                 /*
3551                  * If we have a trans handle we can't wait because the flusher
3552                  * may have to commit the transaction, which would mean we would
3553                  * deadlock since we are waiting for the flusher to finish, but
3554                  * hold the current transaction open.
3555                  */
3556                 if (current->journal_info)
3557                         return -EAGAIN;
3558                 ret = wait_event_interruptible(space_info->wait,
3559                                                !space_info->flush);
3560                 /* Must have been interrupted, return */
3561                 if (ret)
3562                         return -EINTR;
3563
3564                 spin_lock(&space_info->lock);
3565         }
3566
3567         ret = -ENOSPC;
3568         used = space_info->bytes_used + space_info->bytes_reserved +
3569                 space_info->bytes_pinned + space_info->bytes_readonly +
3570                 space_info->bytes_may_use;
3571
3572         /*
3573          * The idea here is that we've not already over-reserved the block group
3574          * then we can go ahead and save our reservation first and then start
3575          * flushing if we need to.  Otherwise if we've already overcommitted
3576          * lets start flushing stuff first and then come back and try to make
3577          * our reservation.
3578          */
3579         if (used <= space_info->total_bytes) {
3580                 if (used + orig_bytes <= space_info->total_bytes) {
3581                         space_info->bytes_may_use += orig_bytes;
3582                         ret = 0;
3583                 } else {
3584                         /*
3585                          * Ok set num_bytes to orig_bytes since we aren't
3586                          * overocmmitted, this way we only try and reclaim what
3587                          * we need.
3588                          */
3589                         num_bytes = orig_bytes;
3590                 }
3591         } else {
3592                 /*
3593                  * Ok we're over committed, set num_bytes to the overcommitted
3594                  * amount plus the amount of bytes that we need for this
3595                  * reservation.
3596                  */
3597                 wait_ordered = true;
3598                 num_bytes = used - space_info->total_bytes +
3599                         (orig_bytes * (retries + 1));
3600         }
3601
3602         if (ret) {
3603                 u64 profile = btrfs_get_alloc_profile(root, 0);
3604                 u64 avail;
3605
3606                 /*
3607                  * If we have a lot of space that's pinned, don't bother doing
3608                  * the overcommit dance yet and just commit the transaction.
3609                  */
3610                 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3611                 do_div(avail, 10);
3612                 if (space_info->bytes_pinned >= avail && flush && !committed) {
3613                         space_info->flush = 1;
3614                         flushing = true;
3615                         spin_unlock(&space_info->lock);
3616                         ret = may_commit_transaction(root, space_info,
3617                                                      orig_bytes, 1);
3618                         if (ret)
3619                                 goto out;
3620                         committed = true;
3621                         goto again;
3622                 }
3623
3624                 spin_lock(&root->fs_info->free_chunk_lock);
3625                 avail = root->fs_info->free_chunk_space;
3626
3627                 /*
3628                  * If we have dup, raid1 or raid10 then only half of the free
3629                  * space is actually useable.
3630                  */
3631                 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3632                                BTRFS_BLOCK_GROUP_RAID1 |
3633                                BTRFS_BLOCK_GROUP_RAID10))
3634                         avail >>= 1;
3635
3636                 /*
3637                  * If we aren't flushing don't let us overcommit too much, say
3638                  * 1/8th of the space.  If we can flush, let it overcommit up to
3639                  * 1/2 of the space.
3640                  */
3641                 if (flush)
3642                         avail >>= 3;
3643                 else
3644                         avail >>= 1;
3645                  spin_unlock(&root->fs_info->free_chunk_lock);
3646
3647                 if (used + num_bytes < space_info->total_bytes + avail) {
3648                         space_info->bytes_may_use += orig_bytes;
3649                         ret = 0;
3650                 } else {
3651                         wait_ordered = true;
3652                 }
3653         }
3654
3655         /*
3656          * Couldn't make our reservation, save our place so while we're trying
3657          * to reclaim space we can actually use it instead of somebody else
3658          * stealing it from us.
3659          */
3660         if (ret && flush) {
3661                 flushing = true;
3662                 space_info->flush = 1;
3663         }
3664
3665         spin_unlock(&space_info->lock);
3666
3667         if (!ret || !flush)
3668                 goto out;
3669
3670         /*
3671          * We do synchronous shrinking since we don't actually unreserve
3672          * metadata until after the IO is completed.
3673          */
3674         ret = shrink_delalloc(root, num_bytes, wait_ordered);
3675         if (ret < 0)
3676                 goto out;
3677
3678         ret = 0;
3679
3680         /*
3681          * So if we were overcommitted it's possible that somebody else flushed
3682          * out enough space and we simply didn't have enough space to reclaim,
3683          * so go back around and try again.
3684          */
3685         if (retries < 2) {
3686                 wait_ordered = true;
3687                 retries++;
3688                 goto again;
3689         }
3690
3691         ret = -ENOSPC;
3692         if (committed)
3693                 goto out;
3694
3695         ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3696         if (!ret) {
3697                 committed = true;
3698                 goto again;
3699         }
3700
3701 out:
3702         if (flushing) {
3703                 spin_lock(&space_info->lock);
3704                 space_info->flush = 0;
3705                 wake_up_all(&space_info->wait);
3706                 spin_unlock(&space_info->lock);
3707         }
3708         return ret;
3709 }
3710
3711 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3712                                              struct btrfs_root *root)
3713 {
3714         struct btrfs_block_rsv *block_rsv = NULL;
3715
3716         if (root->ref_cows || root == root->fs_info->csum_root)
3717                 block_rsv = trans->block_rsv;
3718
3719         if (!block_rsv)
3720                 block_rsv = root->block_rsv;
3721
3722         if (!block_rsv)
3723                 block_rsv = &root->fs_info->empty_block_rsv;
3724
3725         return block_rsv;
3726 }
3727
3728 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3729                                u64 num_bytes)
3730 {
3731         int ret = -ENOSPC;
3732         spin_lock(&block_rsv->lock);
3733         if (block_rsv->reserved >= num_bytes) {
3734                 block_rsv->reserved -= num_bytes;
3735                 if (block_rsv->reserved < block_rsv->size)
3736                         block_rsv->full = 0;
3737                 ret = 0;
3738         }
3739         spin_unlock(&block_rsv->lock);
3740         return ret;
3741 }
3742
3743 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3744                                 u64 num_bytes, int update_size)
3745 {
3746         spin_lock(&block_rsv->lock);
3747         block_rsv->reserved += num_bytes;
3748         if (update_size)
3749                 block_rsv->size += num_bytes;
3750         else if (block_rsv->reserved >= block_rsv->size)
3751                 block_rsv->full = 1;
3752         spin_unlock(&block_rsv->lock);
3753 }
3754
3755 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3756                                     struct btrfs_block_rsv *dest, u64 num_bytes)
3757 {
3758         struct btrfs_space_info *space_info = block_rsv->space_info;
3759
3760         spin_lock(&block_rsv->lock);
3761         if (num_bytes == (u64)-1)
3762                 num_bytes = block_rsv->size;
3763         block_rsv->size -= num_bytes;
3764         if (block_rsv->reserved >= block_rsv->size) {
3765                 num_bytes = block_rsv->reserved - block_rsv->size;
3766                 block_rsv->reserved = block_rsv->size;
3767                 block_rsv->full = 1;
3768         } else {
3769                 num_bytes = 0;
3770         }
3771         spin_unlock(&block_rsv->lock);
3772
3773         if (num_bytes > 0) {
3774                 if (dest) {
3775                         spin_lock(&dest->lock);
3776                         if (!dest->full) {
3777                                 u64 bytes_to_add;
3778
3779                                 bytes_to_add = dest->size - dest->reserved;
3780                                 bytes_to_add = min(num_bytes, bytes_to_add);
3781                                 dest->reserved += bytes_to_add;
3782                                 if (dest->reserved >= dest->size)
3783                                         dest->full = 1;
3784                                 num_bytes -= bytes_to_add;
3785                         }
3786                         spin_unlock(&dest->lock);
3787                 }
3788                 if (num_bytes) {
3789                         spin_lock(&space_info->lock);
3790                         space_info->bytes_may_use -= num_bytes;
3791                         space_info->reservation_progress++;
3792                         spin_unlock(&space_info->lock);
3793                 }
3794         }
3795 }
3796
3797 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3798                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3799 {
3800         int ret;
3801
3802         ret = block_rsv_use_bytes(src, num_bytes);
3803         if (ret)
3804                 return ret;
3805
3806         block_rsv_add_bytes(dst, num_bytes, 1);
3807         return 0;
3808 }
3809
3810 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3811 {
3812         memset(rsv, 0, sizeof(*rsv));
3813         spin_lock_init(&rsv->lock);
3814 }
3815
3816 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3817 {
3818         struct btrfs_block_rsv *block_rsv;
3819         struct btrfs_fs_info *fs_info = root->fs_info;
3820
3821         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3822         if (!block_rsv)
3823                 return NULL;
3824
3825         btrfs_init_block_rsv(block_rsv);
3826         block_rsv->space_info = __find_space_info(fs_info,
3827                                                   BTRFS_BLOCK_GROUP_METADATA);
3828         return block_rsv;
3829 }
3830
3831 void btrfs_free_block_rsv(struct btrfs_root *root,
3832                           struct btrfs_block_rsv *rsv)
3833 {
3834         btrfs_block_rsv_release(root, rsv, (u64)-1);
3835         kfree(rsv);
3836 }
3837
3838 static inline int __block_rsv_add(struct btrfs_root *root,
3839                                   struct btrfs_block_rsv *block_rsv,
3840                                   u64 num_bytes, int flush)
3841 {
3842         int ret;
3843
3844         if (num_bytes == 0)
3845                 return 0;
3846
3847         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3848         if (!ret) {
3849                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3850                 return 0;
3851         }
3852
3853         return ret;
3854 }
3855
3856 int btrfs_block_rsv_add(struct btrfs_root *root,
3857                         struct btrfs_block_rsv *block_rsv,
3858                         u64 num_bytes)
3859 {
3860         return __block_rsv_add(root, block_rsv, num_bytes, 1);
3861 }
3862
3863 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3864                                 struct btrfs_block_rsv *block_rsv,
3865                                 u64 num_bytes)
3866 {
3867         return __block_rsv_add(root, block_rsv, num_bytes, 0);
3868 }
3869
3870 int btrfs_block_rsv_check(struct btrfs_root *root,
3871                           struct btrfs_block_rsv *block_rsv, int min_factor)
3872 {
3873         u64 num_bytes = 0;
3874         int ret = -ENOSPC;
3875
3876         if (!block_rsv)
3877                 return 0;
3878
3879         spin_lock(&block_rsv->lock);
3880         num_bytes = div_factor(block_rsv->size, min_factor);
3881         if (block_rsv->reserved >= num_bytes)
3882                 ret = 0;
3883         spin_unlock(&block_rsv->lock);
3884
3885         return ret;
3886 }
3887
3888 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
3889                                            struct btrfs_block_rsv *block_rsv,
3890                                            u64 min_reserved, int flush)
3891 {
3892         u64 num_bytes = 0;
3893         int ret = -ENOSPC;
3894
3895         if (!block_rsv)
3896                 return 0;
3897
3898         spin_lock(&block_rsv->lock);
3899         num_bytes = min_reserved;
3900         if (block_rsv->reserved >= num_bytes)
3901                 ret = 0;
3902         else
3903                 num_bytes -= block_rsv->reserved;
3904         spin_unlock(&block_rsv->lock);
3905
3906         if (!ret)
3907                 return 0;
3908
3909         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3910         if (!ret) {
3911                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3912                 return 0;
3913         }
3914
3915         return ret;
3916 }
3917
3918 int btrfs_block_rsv_refill(struct btrfs_root *root,
3919                            struct btrfs_block_rsv *block_rsv,
3920                            u64 min_reserved)
3921 {
3922         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
3923 }
3924
3925 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
3926                                    struct btrfs_block_rsv *block_rsv,
3927                                    u64 min_reserved)
3928 {
3929         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
3930 }
3931
3932 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3933                             struct btrfs_block_rsv *dst_rsv,
3934                             u64 num_bytes)
3935 {
3936         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3937 }
3938
3939 void btrfs_block_rsv_release(struct btrfs_root *root,
3940                              struct btrfs_block_rsv *block_rsv,
3941                              u64 num_bytes)
3942 {
3943         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3944         if (global_rsv->full || global_rsv == block_rsv ||
3945             block_rsv->space_info != global_rsv->space_info)
3946                 global_rsv = NULL;
3947         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3948 }
3949
3950 /*
3951  * helper to calculate size of global block reservation.
3952  * the desired value is sum of space used by extent tree,
3953  * checksum tree and root tree
3954  */
3955 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3956 {
3957         struct btrfs_space_info *sinfo;
3958         u64 num_bytes;
3959         u64 meta_used;
3960         u64 data_used;
3961         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
3962
3963         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3964         spin_lock(&sinfo->lock);
3965         data_used = sinfo->bytes_used;
3966         spin_unlock(&sinfo->lock);
3967
3968         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3969         spin_lock(&sinfo->lock);
3970         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3971                 data_used = 0;
3972         meta_used = sinfo->bytes_used;
3973         spin_unlock(&sinfo->lock);
3974
3975         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3976                     csum_size * 2;
3977         num_bytes += div64_u64(data_used + meta_used, 50);
3978
3979         if (num_bytes * 3 > meta_used)
3980                 num_bytes = div64_u64(meta_used, 3);
3981
3982         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3983 }
3984
3985 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3986 {
3987         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3988         struct btrfs_space_info *sinfo = block_rsv->space_info;
3989         u64 num_bytes;
3990
3991         num_bytes = calc_global_metadata_size(fs_info);
3992
3993         spin_lock(&block_rsv->lock);
3994         spin_lock(&sinfo->lock);
3995
3996         block_rsv->size = num_bytes;
3997
3998         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3999                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4000                     sinfo->bytes_may_use;
4001
4002         if (sinfo->total_bytes > num_bytes) {
4003                 num_bytes = sinfo->total_bytes - num_bytes;
4004                 block_rsv->reserved += num_bytes;
4005                 sinfo->bytes_may_use += num_bytes;
4006         }
4007
4008         if (block_rsv->reserved >= block_rsv->size) {
4009                 num_bytes = block_rsv->reserved - block_rsv->size;
4010                 sinfo->bytes_may_use -= num_bytes;
4011                 sinfo->reservation_progress++;
4012                 block_rsv->reserved = block_rsv->size;
4013                 block_rsv->full = 1;
4014         }
4015
4016         spin_unlock(&sinfo->lock);
4017         spin_unlock(&block_rsv->lock);
4018 }
4019
4020 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4021 {
4022         struct btrfs_space_info *space_info;
4023
4024         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4025         fs_info->chunk_block_rsv.space_info = space_info;
4026
4027         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4028         fs_info->global_block_rsv.space_info = space_info;
4029         fs_info->delalloc_block_rsv.space_info = space_info;
4030         fs_info->trans_block_rsv.space_info = space_info;
4031         fs_info->empty_block_rsv.space_info = space_info;
4032         fs_info->delayed_block_rsv.space_info = space_info;
4033
4034         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4035         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4036         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4037         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4038         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4039
4040         update_global_block_rsv(fs_info);
4041 }
4042
4043 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4044 {
4045         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
4046         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4047         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4048         WARN_ON(fs_info->trans_block_rsv.size > 0);
4049         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4050         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4051         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4052         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4053         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4054 }
4055
4056 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4057                                   struct btrfs_root *root)
4058 {
4059         if (!trans->bytes_reserved)
4060                 return;
4061
4062         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4063         trans->bytes_reserved = 0;
4064 }
4065
4066 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4067                                   struct inode *inode)
4068 {
4069         struct btrfs_root *root = BTRFS_I(inode)->root;
4070         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4071         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4072
4073         /*
4074          * We need to hold space in order to delete our orphan item once we've
4075          * added it, so this takes the reservation so we can release it later
4076          * when we are truly done with the orphan item.
4077          */
4078         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4079         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4080 }
4081
4082 void btrfs_orphan_release_metadata(struct inode *inode)
4083 {
4084         struct btrfs_root *root = BTRFS_I(inode)->root;
4085         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4086         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4087 }
4088
4089 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4090                                 struct btrfs_pending_snapshot *pending)
4091 {
4092         struct btrfs_root *root = pending->root;
4093         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4094         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4095         /*
4096          * two for root back/forward refs, two for directory entries
4097          * and one for root of the snapshot.
4098          */
4099         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4100         dst_rsv->space_info = src_rsv->space_info;
4101         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4102 }
4103
4104 /**
4105  * drop_outstanding_extent - drop an outstanding extent
4106  * @inode: the inode we're dropping the extent for
4107  *
4108  * This is called when we are freeing up an outstanding extent, either called
4109  * after an error or after an extent is written.  This will return the number of
4110  * reserved extents that need to be freed.  This must be called with
4111  * BTRFS_I(inode)->lock held.
4112  */
4113 static unsigned drop_outstanding_extent(struct inode *inode)
4114 {
4115         unsigned drop_inode_space = 0;
4116         unsigned dropped_extents = 0;
4117
4118         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4119         BTRFS_I(inode)->outstanding_extents--;
4120
4121         if (BTRFS_I(inode)->outstanding_extents == 0 &&
4122             BTRFS_I(inode)->delalloc_meta_reserved) {
4123                 drop_inode_space = 1;
4124                 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4125         }
4126
4127         /*
4128          * If we have more or the same amount of outsanding extents than we have
4129          * reserved then we need to leave the reserved extents count alone.
4130          */
4131         if (BTRFS_I(inode)->outstanding_extents >=
4132             BTRFS_I(inode)->reserved_extents)
4133                 return drop_inode_space;
4134
4135         dropped_extents = BTRFS_I(inode)->reserved_extents -
4136                 BTRFS_I(inode)->outstanding_extents;
4137         BTRFS_I(inode)->reserved_extents -= dropped_extents;
4138         return dropped_extents + drop_inode_space;
4139 }
4140
4141 /**
4142  * calc_csum_metadata_size - return the amount of metada space that must be
4143  *      reserved/free'd for the given bytes.
4144  * @inode: the inode we're manipulating
4145  * @num_bytes: the number of bytes in question
4146  * @reserve: 1 if we are reserving space, 0 if we are freeing space
4147  *
4148  * This adjusts the number of csum_bytes in the inode and then returns the
4149  * correct amount of metadata that must either be reserved or freed.  We
4150  * calculate how many checksums we can fit into one leaf and then divide the
4151  * number of bytes that will need to be checksumed by this value to figure out
4152  * how many checksums will be required.  If we are adding bytes then the number
4153  * may go up and we will return the number of additional bytes that must be
4154  * reserved.  If it is going down we will return the number of bytes that must
4155  * be freed.
4156  *
4157  * This must be called with BTRFS_I(inode)->lock held.
4158  */
4159 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4160                                    int reserve)
4161 {
4162         struct btrfs_root *root = BTRFS_I(inode)->root;
4163         u64 csum_size;
4164         int num_csums_per_leaf;
4165         int num_csums;
4166         int old_csums;
4167
4168         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4169             BTRFS_I(inode)->csum_bytes == 0)
4170                 return 0;
4171
4172         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4173         if (reserve)
4174                 BTRFS_I(inode)->csum_bytes += num_bytes;
4175         else
4176                 BTRFS_I(inode)->csum_bytes -= num_bytes;
4177         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4178         num_csums_per_leaf = (int)div64_u64(csum_size,
4179                                             sizeof(struct btrfs_csum_item) +
4180                                             sizeof(struct btrfs_disk_key));
4181         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4182         num_csums = num_csums + num_csums_per_leaf - 1;
4183         num_csums = num_csums / num_csums_per_leaf;
4184
4185         old_csums = old_csums + num_csums_per_leaf - 1;
4186         old_csums = old_csums / num_csums_per_leaf;
4187
4188         /* No change, no need to reserve more */
4189         if (old_csums == num_csums)
4190                 return 0;
4191
4192         if (reserve)
4193                 return btrfs_calc_trans_metadata_size(root,
4194                                                       num_csums - old_csums);
4195
4196         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4197 }
4198
4199 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4200 {
4201         struct btrfs_root *root = BTRFS_I(inode)->root;
4202         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4203         u64 to_reserve = 0;
4204         u64 csum_bytes;
4205         unsigned nr_extents = 0;
4206         int extra_reserve = 0;
4207         int flush = 1;
4208         int ret;
4209
4210         /* Need to be holding the i_mutex here if we aren't free space cache */
4211         if (btrfs_is_free_space_inode(root, inode))
4212                 flush = 0;
4213         else
4214                 WARN_ON(!mutex_is_locked(&inode->i_mutex));
4215
4216         if (flush && btrfs_transaction_in_commit(root->fs_info))
4217                 schedule_timeout(1);
4218
4219         num_bytes = ALIGN(num_bytes, root->sectorsize);
4220
4221         spin_lock(&BTRFS_I(inode)->lock);
4222         BTRFS_I(inode)->outstanding_extents++;
4223
4224         if (BTRFS_I(inode)->outstanding_extents >
4225             BTRFS_I(inode)->reserved_extents)
4226                 nr_extents = BTRFS_I(inode)->outstanding_extents -
4227                         BTRFS_I(inode)->reserved_extents;
4228
4229         /*
4230          * Add an item to reserve for updating the inode when we complete the
4231          * delalloc io.
4232          */
4233         if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4234                 nr_extents++;
4235                 extra_reserve = 1;
4236         }
4237
4238         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4239         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4240         csum_bytes = BTRFS_I(inode)->csum_bytes;
4241         spin_unlock(&BTRFS_I(inode)->lock);
4242
4243         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4244         if (ret) {
4245                 u64 to_free = 0;
4246                 unsigned dropped;
4247
4248                 spin_lock(&BTRFS_I(inode)->lock);
4249                 dropped = drop_outstanding_extent(inode);
4250                 /*
4251                  * If the inodes csum_bytes is the same as the original
4252                  * csum_bytes then we know we haven't raced with any free()ers
4253                  * so we can just reduce our inodes csum bytes and carry on.
4254                  * Otherwise we have to do the normal free thing to account for
4255                  * the case that the free side didn't free up its reserve
4256                  * because of this outstanding reservation.
4257                  */
4258                 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4259                         calc_csum_metadata_size(inode, num_bytes, 0);
4260                 else
4261                         to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4262                 spin_unlock(&BTRFS_I(inode)->lock);
4263                 if (dropped)
4264                         to_free += btrfs_calc_trans_metadata_size(root, dropped);
4265
4266                 if (to_free)
4267                         btrfs_block_rsv_release(root, block_rsv, to_free);
4268                 return ret;
4269         }
4270
4271         spin_lock(&BTRFS_I(inode)->lock);
4272         if (extra_reserve) {
4273                 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4274                 nr_extents--;
4275         }
4276         BTRFS_I(inode)->reserved_extents += nr_extents;
4277         spin_unlock(&BTRFS_I(inode)->lock);
4278
4279         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4280
4281         return 0;
4282 }
4283
4284 /**
4285  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4286  * @inode: the inode to release the reservation for
4287  * @num_bytes: the number of bytes we're releasing
4288  *
4289  * This will release the metadata reservation for an inode.  This can be called
4290  * once we complete IO for a given set of bytes to release their metadata
4291  * reservations.
4292  */
4293 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4294 {
4295         struct btrfs_root *root = BTRFS_I(inode)->root;
4296         u64 to_free = 0;
4297         unsigned dropped;
4298
4299         num_bytes = ALIGN(num_bytes, root->sectorsize);
4300         spin_lock(&BTRFS_I(inode)->lock);
4301         dropped = drop_outstanding_extent(inode);
4302
4303         to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4304         spin_unlock(&BTRFS_I(inode)->lock);
4305         if (dropped > 0)
4306                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4307
4308         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4309                                 to_free);
4310 }
4311
4312 /**
4313  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4314  * @inode: inode we're writing to
4315  * @num_bytes: the number of bytes we want to allocate
4316  *
4317  * This will do the following things
4318  *
4319  * o reserve space in the data space info for num_bytes
4320  * o reserve space in the metadata space info based on number of outstanding
4321  *   extents and how much csums will be needed
4322  * o add to the inodes ->delalloc_bytes
4323  * o add it to the fs_info's delalloc inodes list.
4324  *
4325  * This will return 0 for success and -ENOSPC if there is no space left.
4326  */
4327 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4328 {
4329         int ret;
4330
4331         ret = btrfs_check_data_free_space(inode, num_bytes);
4332         if (ret)
4333                 return ret;
4334
4335         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4336         if (ret) {
4337                 btrfs_free_reserved_data_space(inode, num_bytes);
4338                 return ret;
4339         }
4340
4341         return 0;
4342 }
4343
4344 /**
4345  * btrfs_delalloc_release_space - release data and metadata space for delalloc
4346  * @inode: inode we're releasing space for
4347  * @num_bytes: the number of bytes we want to free up
4348  *
4349  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
4350  * called in the case that we don't need the metadata AND data reservations
4351  * anymore.  So if there is an error or we insert an inline extent.
4352  *
4353  * This function will release the metadata space that was not used and will
4354  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4355  * list if there are no delalloc bytes left.
4356  */
4357 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4358 {
4359         btrfs_delalloc_release_metadata(inode, num_bytes);
4360         btrfs_free_reserved_data_space(inode, num_bytes);
4361 }
4362
4363 static int update_block_group(struct btrfs_trans_handle *trans,
4364                               struct btrfs_root *root,
4365                               u64 bytenr, u64 num_bytes, int alloc)
4366 {
4367         struct btrfs_block_group_cache *cache = NULL;
4368         struct btrfs_fs_info *info = root->fs_info;
4369         u64 total = num_bytes;
4370         u64 old_val;
4371         u64 byte_in_group;
4372         int factor;
4373
4374         /* block accounting for super block */
4375         spin_lock(&info->delalloc_lock);
4376         old_val = btrfs_super_bytes_used(info->super_copy);
4377         if (alloc)
4378                 old_val += num_bytes;
4379         else
4380                 old_val -= num_bytes;
4381         btrfs_set_super_bytes_used(info->super_copy, old_val);
4382         spin_unlock(&info->delalloc_lock);
4383
4384         while (total) {
4385                 cache = btrfs_lookup_block_group(info, bytenr);
4386                 if (!cache)
4387                         return -1;
4388                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4389                                     BTRFS_BLOCK_GROUP_RAID1 |
4390                                     BTRFS_BLOCK_GROUP_RAID10))
4391                         factor = 2;
4392                 else
4393                         factor = 1;
4394                 /*
4395                  * If this block group has free space cache written out, we
4396                  * need to make sure to load it if we are removing space.  This
4397                  * is because we need the unpinning stage to actually add the
4398                  * space back to the block group, otherwise we will leak space.
4399                  */
4400                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4401                         cache_block_group(cache, trans, NULL, 1);
4402
4403                 byte_in_group = bytenr - cache->key.objectid;
4404                 WARN_ON(byte_in_group > cache->key.offset);
4405
4406                 spin_lock(&cache->space_info->lock);
4407                 spin_lock(&cache->lock);
4408
4409                 if (btrfs_test_opt(root, SPACE_CACHE) &&
4410                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4411                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4412
4413                 cache->dirty = 1;
4414                 old_val = btrfs_block_group_used(&cache->item);
4415                 num_bytes = min(total, cache->key.offset - byte_in_group);
4416                 if (alloc) {
4417                         old_val += num_bytes;
4418                         btrfs_set_block_group_used(&cache->item, old_val);
4419                         cache->reserved -= num_bytes;
4420                         cache->space_info->bytes_reserved -= num_bytes;
4421                         cache->space_info->bytes_used += num_bytes;
4422                         cache->space_info->disk_used += num_bytes * factor;
4423                         spin_unlock(&cache->lock);
4424                         spin_unlock(&cache->space_info->lock);
4425                 } else {
4426                         old_val -= num_bytes;
4427                         btrfs_set_block_group_used(&cache->item, old_val);
4428                         cache->pinned += num_bytes;
4429                         cache->space_info->bytes_pinned += num_bytes;
4430                         cache->space_info->bytes_used -= num_bytes;
4431                         cache->space_info->disk_used -= num_bytes * factor;
4432                         spin_unlock(&cache->lock);
4433                         spin_unlock(&cache->space_info->lock);
4434
4435                         set_extent_dirty(info->pinned_extents,
4436                                          bytenr, bytenr + num_bytes - 1,
4437                                          GFP_NOFS | __GFP_NOFAIL);
4438                 }
4439                 btrfs_put_block_group(cache);
4440                 total -= num_bytes;
4441                 bytenr += num_bytes;
4442         }
4443         return 0;
4444 }
4445
4446 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4447 {
4448         struct btrfs_block_group_cache *cache;
4449         u64 bytenr;
4450
4451         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4452         if (!cache)
4453                 return 0;
4454
4455         bytenr = cache->key.objectid;
4456         btrfs_put_block_group(cache);
4457
4458         return bytenr;
4459 }
4460
4461 static int pin_down_extent(struct btrfs_root *root,
4462                            struct btrfs_block_group_cache *cache,
4463                            u64 bytenr, u64 num_bytes, int reserved)
4464 {
4465         spin_lock(&cache->space_info->lock);
4466         spin_lock(&cache->lock);
4467         cache->pinned += num_bytes;
4468         cache->space_info->bytes_pinned += num_bytes;
4469         if (reserved) {
4470                 cache->reserved -= num_bytes;
4471                 cache->space_info->bytes_reserved -= num_bytes;
4472         }
4473         spin_unlock(&cache->lock);
4474         spin_unlock(&cache->space_info->lock);
4475
4476         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4477                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4478         return 0;
4479 }
4480
4481 /*
4482  * this function must be called within transaction
4483  */
4484 int btrfs_pin_extent(struct btrfs_root *root,
4485                      u64 bytenr, u64 num_bytes, int reserved)
4486 {
4487         struct btrfs_block_group_cache *cache;
4488
4489         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4490         BUG_ON(!cache);
4491
4492         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4493
4494         btrfs_put_block_group(cache);
4495         return 0;
4496 }
4497
4498 /*
4499  * this function must be called within transaction
4500  */
4501 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4502                                     struct btrfs_root *root,
4503                                     u64 bytenr, u64 num_bytes)
4504 {
4505         struct btrfs_block_group_cache *cache;
4506
4507         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4508         BUG_ON(!cache);
4509
4510         /*
4511          * pull in the free space cache (if any) so that our pin
4512          * removes the free space from the cache.  We have load_only set
4513          * to one because the slow code to read in the free extents does check
4514          * the pinned extents.
4515          */
4516         cache_block_group(cache, trans, root, 1);
4517
4518         pin_down_extent(root, cache, bytenr, num_bytes, 0);
4519
4520         /* remove us from the free space cache (if we're there at all) */
4521         btrfs_remove_free_space(cache, bytenr, num_bytes);
4522         btrfs_put_block_group(cache);
4523         return 0;
4524 }
4525
4526 /**
4527  * btrfs_update_reserved_bytes - update the block_group and space info counters
4528  * @cache:      The cache we are manipulating
4529  * @num_bytes:  The number of bytes in question
4530  * @reserve:    One of the reservation enums
4531  *
4532  * This is called by the allocator when it reserves space, or by somebody who is
4533  * freeing space that was never actually used on disk.  For example if you
4534  * reserve some space for a new leaf in transaction A and before transaction A
4535  * commits you free that leaf, you call this with reserve set to 0 in order to
4536  * clear the reservation.
4537  *
4538  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4539  * ENOSPC accounting.  For data we handle the reservation through clearing the
4540  * delalloc bits in the io_tree.  We have to do this since we could end up
4541  * allocating less disk space for the amount of data we have reserved in the
4542  * case of compression.
4543  *
4544  * If this is a reservation and the block group has become read only we cannot
4545  * make the reservation and return -EAGAIN, otherwise this function always
4546  * succeeds.
4547  */
4548 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4549                                        u64 num_bytes, int reserve)
4550 {
4551         struct btrfs_space_info *space_info = cache->space_info;
4552         int ret = 0;
4553         spin_lock(&space_info->lock);
4554         spin_lock(&cache->lock);
4555         if (reserve != RESERVE_FREE) {
4556                 if (cache->ro) {
4557                         ret = -EAGAIN;
4558                 } else {
4559                         cache->reserved += num_bytes;
4560                         space_info->bytes_reserved += num_bytes;
4561                         if (reserve == RESERVE_ALLOC) {
4562                                 BUG_ON(space_info->bytes_may_use < num_bytes);
4563                                 space_info->bytes_may_use -= num_bytes;
4564                         }
4565                 }
4566         } else {
4567                 if (cache->ro)
4568                         space_info->bytes_readonly += num_bytes;
4569                 cache->reserved -= num_bytes;
4570                 space_info->bytes_reserved -= num_bytes;
4571                 space_info->reservation_progress++;
4572         }
4573         spin_unlock(&cache->lock);
4574         spin_unlock(&space_info->lock);
4575         return ret;
4576 }
4577
4578 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4579                                 struct btrfs_root *root)
4580 {
4581         struct btrfs_fs_info *fs_info = root->fs_info;
4582         struct btrfs_caching_control *next;
4583         struct btrfs_caching_control *caching_ctl;
4584         struct btrfs_block_group_cache *cache;
4585
4586         down_write(&fs_info->extent_commit_sem);
4587
4588         list_for_each_entry_safe(caching_ctl, next,
4589                                  &fs_info->caching_block_groups, list) {
4590                 cache = caching_ctl->block_group;
4591                 if (block_group_cache_done(cache)) {
4592                         cache->last_byte_to_unpin = (u64)-1;
4593                         list_del_init(&caching_ctl->list);
4594                         put_caching_control(caching_ctl);
4595                 } else {
4596                         cache->last_byte_to_unpin = caching_ctl->progress;
4597                 }
4598         }
4599
4600         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4601                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4602         else
4603                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4604
4605         up_write(&fs_info->extent_commit_sem);
4606
4607         update_global_block_rsv(fs_info);
4608         return 0;
4609 }
4610
4611 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4612 {
4613         struct btrfs_fs_info *fs_info = root->fs_info;
4614         struct btrfs_block_group_cache *cache = NULL;
4615         u64 len;
4616
4617         while (start <= end) {
4618                 if (!cache ||
4619                     start >= cache->key.objectid + cache->key.offset) {
4620                         if (cache)
4621                                 btrfs_put_block_group(cache);
4622                         cache = btrfs_lookup_block_group(fs_info, start);
4623                         BUG_ON(!cache);
4624                 }
4625
4626                 len = cache->key.objectid + cache->key.offset - start;
4627                 len = min(len, end + 1 - start);
4628
4629                 if (start < cache->last_byte_to_unpin) {
4630                         len = min(len, cache->last_byte_to_unpin - start);
4631                         btrfs_add_free_space(cache, start, len);
4632                 }
4633
4634                 start += len;
4635
4636                 spin_lock(&cache->space_info->lock);
4637                 spin_lock(&cache->lock);
4638                 cache->pinned -= len;
4639                 cache->space_info->bytes_pinned -= len;
4640                 if (cache->ro)
4641                         cache->space_info->bytes_readonly += len;
4642                 spin_unlock(&cache->lock);
4643                 spin_unlock(&cache->space_info->lock);
4644         }
4645
4646         if (cache)
4647                 btrfs_put_block_group(cache);
4648         return 0;
4649 }
4650
4651 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4652                                struct btrfs_root *root)
4653 {
4654         struct btrfs_fs_info *fs_info = root->fs_info;
4655         struct extent_io_tree *unpin;
4656         u64 start;
4657         u64 end;
4658         int ret;
4659
4660         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4661                 unpin = &fs_info->freed_extents[1];
4662         else
4663                 unpin = &fs_info->freed_extents[0];
4664
4665         while (1) {
4666                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4667                                             EXTENT_DIRTY);
4668                 if (ret)
4669                         break;
4670
4671                 if (btrfs_test_opt(root, DISCARD))
4672                         ret = btrfs_discard_extent(root, start,
4673                                                    end + 1 - start, NULL);
4674
4675                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4676                 unpin_extent_range(root, start, end);
4677                 cond_resched();
4678         }
4679
4680         return 0;
4681 }
4682
4683 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4684                                 struct btrfs_root *root,
4685                                 u64 bytenr, u64 num_bytes, u64 parent,
4686                                 u64 root_objectid, u64 owner_objectid,
4687                                 u64 owner_offset, int refs_to_drop,
4688                                 struct btrfs_delayed_extent_op *extent_op)
4689 {
4690         struct btrfs_key key;
4691         struct btrfs_path *path;
4692         struct btrfs_fs_info *info = root->fs_info;
4693         struct btrfs_root *extent_root = info->extent_root;
4694         struct extent_buffer *leaf;
4695         struct btrfs_extent_item *ei;
4696         struct btrfs_extent_inline_ref *iref;
4697         int ret;
4698         int is_data;
4699         int extent_slot = 0;
4700         int found_extent = 0;
4701         int num_to_del = 1;
4702         u32 item_size;
4703         u64 refs;
4704
4705         path = btrfs_alloc_path();
4706         if (!path)
4707                 return -ENOMEM;
4708
4709         path->reada = 1;
4710         path->leave_spinning = 1;
4711
4712         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4713         BUG_ON(!is_data && refs_to_drop != 1);
4714
4715         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4716                                     bytenr, num_bytes, parent,
4717                                     root_objectid, owner_objectid,
4718                                     owner_offset);
4719         if (ret == 0) {
4720                 extent_slot = path->slots[0];
4721                 while (extent_slot >= 0) {
4722                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4723                                               extent_slot);
4724                         if (key.objectid != bytenr)
4725                                 break;
4726                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4727                             key.offset == num_bytes) {
4728                                 found_extent = 1;
4729                                 break;
4730                         }
4731                         if (path->slots[0] - extent_slot > 5)
4732                                 break;
4733                         extent_slot--;
4734                 }
4735 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4736                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4737                 if (found_extent && item_size < sizeof(*ei))
4738                         found_extent = 0;
4739 #endif
4740                 if (!found_extent) {
4741                         BUG_ON(iref);
4742                         ret = remove_extent_backref(trans, extent_root, path,
4743                                                     NULL, refs_to_drop,
4744                                                     is_data);
4745                         BUG_ON(ret);
4746                         btrfs_release_path(path);
4747                         path->leave_spinning = 1;
4748
4749                         key.objectid = bytenr;
4750                         key.type = BTRFS_EXTENT_ITEM_KEY;
4751                         key.offset = num_bytes;
4752
4753                         ret = btrfs_search_slot(trans, extent_root,
4754                                                 &key, path, -1, 1);
4755                         if (ret) {
4756                                 printk(KERN_ERR "umm, got %d back from search"
4757                                        ", was looking for %llu\n", ret,
4758                                        (unsigned long long)bytenr);
4759                                 if (ret > 0)
4760                                         btrfs_print_leaf(extent_root,
4761                                                          path->nodes[0]);
4762                         }
4763                         BUG_ON(ret);
4764                         extent_slot = path->slots[0];
4765                 }
4766         } else {
4767                 btrfs_print_leaf(extent_root, path->nodes[0]);
4768                 WARN_ON(1);
4769                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4770                        "parent %llu root %llu  owner %llu offset %llu\n",
4771                        (unsigned long long)bytenr,
4772                        (unsigned long long)parent,
4773                        (unsigned long long)root_objectid,
4774                        (unsigned long long)owner_objectid,
4775                        (unsigned long long)owner_offset);
4776         }
4777
4778         leaf = path->nodes[0];
4779         item_size = btrfs_item_size_nr(leaf, extent_slot);
4780 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4781         if (item_size < sizeof(*ei)) {
4782                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4783                 ret = convert_extent_item_v0(trans, extent_root, path,
4784                                              owner_objectid, 0);
4785                 BUG_ON(ret < 0);
4786
4787                 btrfs_release_path(path);
4788                 path->leave_spinning = 1;
4789
4790                 key.objectid = bytenr;
4791                 key.type = BTRFS_EXTENT_ITEM_KEY;
4792                 key.offset = num_bytes;
4793
4794                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4795                                         -1, 1);
4796                 if (ret) {
4797                         printk(KERN_ERR "umm, got %d back from search"
4798                                ", was looking for %llu\n", ret,
4799                                (unsigned long long)bytenr);
4800                         btrfs_print_leaf(extent_root, path->nodes[0]);
4801                 }
4802                 BUG_ON(ret);
4803                 extent_slot = path->slots[0];
4804                 leaf = path->nodes[0];
4805                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4806         }
4807 #endif
4808         BUG_ON(item_size < sizeof(*ei));
4809         ei = btrfs_item_ptr(leaf, extent_slot,
4810                             struct btrfs_extent_item);
4811         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4812                 struct btrfs_tree_block_info *bi;
4813                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4814                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4815                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4816         }
4817
4818         refs = btrfs_extent_refs(leaf, ei);
4819         BUG_ON(refs < refs_to_drop);
4820         refs -= refs_to_drop;
4821
4822         if (refs > 0) {
4823                 if (extent_op)
4824                         __run_delayed_extent_op(extent_op, leaf, ei);
4825                 /*
4826                  * In the case of inline back ref, reference count will
4827                  * be updated by remove_extent_backref
4828                  */
4829                 if (iref) {
4830                         BUG_ON(!found_extent);
4831                 } else {
4832                         btrfs_set_extent_refs(leaf, ei, refs);
4833                         btrfs_mark_buffer_dirty(leaf);
4834                 }
4835                 if (found_extent) {
4836                         ret = remove_extent_backref(trans, extent_root, path,
4837                                                     iref, refs_to_drop,
4838                                                     is_data);
4839                         BUG_ON(ret);
4840                 }
4841         } else {
4842                 if (found_extent) {
4843                         BUG_ON(is_data && refs_to_drop !=
4844                                extent_data_ref_count(root, path, iref));
4845                         if (iref) {
4846                                 BUG_ON(path->slots[0] != extent_slot);
4847                         } else {
4848                                 BUG_ON(path->slots[0] != extent_slot + 1);
4849                                 path->slots[0] = extent_slot;
4850                                 num_to_del = 2;
4851                         }
4852                 }
4853
4854                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4855                                       num_to_del);
4856                 BUG_ON(ret);
4857                 btrfs_release_path(path);
4858
4859                 if (is_data) {
4860                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4861                         BUG_ON(ret);
4862                 } else {
4863                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4864                              bytenr >> PAGE_CACHE_SHIFT,
4865                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4866                 }
4867
4868                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4869                 BUG_ON(ret);
4870         }
4871         btrfs_free_path(path);
4872         return ret;
4873 }
4874
4875 /*
4876  * when we free an block, it is possible (and likely) that we free the last
4877  * delayed ref for that extent as well.  This searches the delayed ref tree for
4878  * a given extent, and if there are no other delayed refs to be processed, it
4879  * removes it from the tree.
4880  */
4881 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4882                                       struct btrfs_root *root, u64 bytenr)
4883 {
4884         struct btrfs_delayed_ref_head *head;
4885         struct btrfs_delayed_ref_root *delayed_refs;
4886         struct btrfs_delayed_ref_node *ref;
4887         struct rb_node *node;
4888         int ret = 0;
4889
4890         delayed_refs = &trans->transaction->delayed_refs;
4891         spin_lock(&delayed_refs->lock);
4892         head = btrfs_find_delayed_ref_head(trans, bytenr);
4893         if (!head)
4894                 goto out;
4895
4896         node = rb_prev(&head->node.rb_node);
4897         if (!node)
4898                 goto out;
4899
4900         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4901
4902         /* there are still entries for this ref, we can't drop it */
4903         if (ref->bytenr == bytenr)
4904                 goto out;
4905
4906         if (head->extent_op) {
4907                 if (!head->must_insert_reserved)
4908                         goto out;
4909                 kfree(head->extent_op);
4910                 head->extent_op = NULL;
4911         }
4912
4913         /*
4914          * waiting for the lock here would deadlock.  If someone else has it
4915          * locked they are already in the process of dropping it anyway
4916          */
4917         if (!mutex_trylock(&head->mutex))
4918                 goto out;
4919
4920         /*
4921          * at this point we have a head with no other entries.  Go
4922          * ahead and process it.
4923          */
4924         head->node.in_tree = 0;
4925         rb_erase(&head->node.rb_node, &delayed_refs->root);
4926
4927         delayed_refs->num_entries--;
4928
4929         /*
4930          * we don't take a ref on the node because we're removing it from the
4931          * tree, so we just steal the ref the tree was holding.
4932          */
4933         delayed_refs->num_heads--;
4934         if (list_empty(&head->cluster))
4935                 delayed_refs->num_heads_ready--;
4936
4937         list_del_init(&head->cluster);
4938         spin_unlock(&delayed_refs->lock);
4939
4940         BUG_ON(head->extent_op);
4941         if (head->must_insert_reserved)
4942                 ret = 1;
4943
4944         mutex_unlock(&head->mutex);
4945         btrfs_put_delayed_ref(&head->node);
4946         return ret;
4947 out:
4948         spin_unlock(&delayed_refs->lock);
4949         return 0;
4950 }
4951
4952 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4953                            struct btrfs_root *root,
4954                            struct extent_buffer *buf,
4955                            u64 parent, int last_ref)
4956 {
4957         struct btrfs_block_group_cache *cache = NULL;
4958         int ret;
4959
4960         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4961                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4962                                                 parent, root->root_key.objectid,
4963                                                 btrfs_header_level(buf),
4964                                                 BTRFS_DROP_DELAYED_REF, NULL);
4965                 BUG_ON(ret);
4966         }
4967
4968         if (!last_ref)
4969                 return;
4970
4971         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4972
4973         if (btrfs_header_generation(buf) == trans->transid) {
4974                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4975                         ret = check_ref_cleanup(trans, root, buf->start);
4976                         if (!ret)
4977                                 goto out;
4978                 }
4979
4980                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4981                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4982                         goto out;
4983                 }
4984
4985                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4986
4987                 btrfs_add_free_space(cache, buf->start, buf->len);
4988                 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
4989         }
4990 out:
4991         /*
4992          * Deleting the buffer, clear the corrupt flag since it doesn't matter
4993          * anymore.
4994          */
4995         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4996         btrfs_put_block_group(cache);
4997 }
4998
4999 int btrfs_free_extent(struct btrfs_trans_handle *trans,
5000                       struct btrfs_root *root,
5001                       u64 bytenr, u64 num_bytes, u64 parent,
5002                       u64 root_objectid, u64 owner, u64 offset)
5003 {
5004         int ret;
5005
5006         /*
5007          * tree log blocks never actually go into the extent allocation
5008          * tree, just update pinning info and exit early.
5009          */
5010         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5011                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5012                 /* unlocks the pinned mutex */
5013                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5014                 ret = 0;
5015         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5016                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
5017                                         parent, root_objectid, (int)owner,
5018                                         BTRFS_DROP_DELAYED_REF, NULL);
5019                 BUG_ON(ret);
5020         } else {
5021                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
5022                                         parent, root_objectid, owner,
5023                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
5024                 BUG_ON(ret);
5025         }
5026         return ret;
5027 }
5028
5029 static u64 stripe_align(struct btrfs_root *root, u64 val)
5030 {
5031         u64 mask = ((u64)root->stripesize - 1);
5032         u64 ret = (val + mask) & ~mask;
5033         return ret;
5034 }
5035
5036 /*
5037  * when we wait for progress in the block group caching, its because
5038  * our allocation attempt failed at least once.  So, we must sleep
5039  * and let some progress happen before we try again.
5040  *
5041  * This function will sleep at least once waiting for new free space to
5042  * show up, and then it will check the block group free space numbers
5043  * for our min num_bytes.  Another option is to have it go ahead
5044  * and look in the rbtree for a free extent of a given size, but this
5045  * is a good start.
5046  */
5047 static noinline int
5048 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5049                                 u64 num_bytes)
5050 {
5051         struct btrfs_caching_control *caching_ctl;
5052         DEFINE_WAIT(wait);
5053
5054         caching_ctl = get_caching_control(cache);
5055         if (!caching_ctl)
5056                 return 0;
5057
5058         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5059                    (cache->free_space_ctl->free_space >= num_bytes));
5060
5061         put_caching_control(caching_ctl);
5062         return 0;
5063 }
5064
5065 static noinline int
5066 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5067 {
5068         struct btrfs_caching_control *caching_ctl;
5069         DEFINE_WAIT(wait);
5070
5071         caching_ctl = get_caching_control(cache);
5072         if (!caching_ctl)
5073                 return 0;
5074
5075         wait_event(caching_ctl->wait, block_group_cache_done(cache));
5076
5077         put_caching_control(caching_ctl);
5078         return 0;
5079 }
5080
5081 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5082 {
5083         int index;
5084         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5085                 index = 0;
5086         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5087                 index = 1;
5088         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5089                 index = 2;
5090         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5091                 index = 3;
5092         else
5093                 index = 4;
5094         return index;
5095 }
5096
5097 enum btrfs_loop_type {
5098         LOOP_FIND_IDEAL = 0,
5099         LOOP_CACHING_NOWAIT = 1,
5100         LOOP_CACHING_WAIT = 2,
5101         LOOP_ALLOC_CHUNK = 3,
5102         LOOP_NO_EMPTY_SIZE = 4,
5103 };
5104
5105 /*
5106  * walks the btree of allocated extents and find a hole of a given size.
5107  * The key ins is changed to record the hole:
5108  * ins->objectid == block start
5109  * ins->flags = BTRFS_EXTENT_ITEM_KEY
5110  * ins->offset == number of blocks
5111  * Any available blocks before search_start are skipped.
5112  */
5113 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5114                                      struct btrfs_root *orig_root,
5115                                      u64 num_bytes, u64 empty_size,
5116                                      u64 search_start, u64 search_end,
5117                                      u64 hint_byte, struct btrfs_key *ins,
5118                                      u64 data)
5119 {
5120         int ret = 0;
5121         struct btrfs_root *root = orig_root->fs_info->extent_root;
5122         struct btrfs_free_cluster *last_ptr = NULL;
5123         struct btrfs_block_group_cache *block_group = NULL;
5124         struct btrfs_block_group_cache *used_block_group;
5125         int empty_cluster = 2 * 1024 * 1024;
5126         int allowed_chunk_alloc = 0;
5127         int done_chunk_alloc = 0;
5128         struct btrfs_space_info *space_info;
5129         int loop = 0;
5130         int index = 0;
5131         int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5132                 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5133         bool found_uncached_bg = false;
5134         bool failed_cluster_refill = false;
5135         bool failed_alloc = false;
5136         bool use_cluster = true;
5137         bool have_caching_bg = false;
5138         u64 ideal_cache_percent = 0;
5139         u64 ideal_cache_offset = 0;
5140
5141         WARN_ON(num_bytes < root->sectorsize);
5142         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5143         ins->objectid = 0;
5144         ins->offset = 0;
5145
5146         space_info = __find_space_info(root->fs_info, data);
5147         if (!space_info) {
5148                 printk(KERN_ERR "No space info for %llu\n", data);
5149                 return -ENOSPC;
5150         }
5151
5152         /*
5153          * If the space info is for both data and metadata it means we have a
5154          * small filesystem and we can't use the clustering stuff.
5155          */
5156         if (btrfs_mixed_space_info(space_info))
5157                 use_cluster = false;
5158
5159         if (orig_root->ref_cows || empty_size)
5160                 allowed_chunk_alloc = 1;
5161
5162         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5163                 last_ptr = &root->fs_info->meta_alloc_cluster;
5164                 if (!btrfs_test_opt(root, SSD))
5165                         empty_cluster = 64 * 1024;
5166         }
5167
5168         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5169             btrfs_test_opt(root, SSD)) {
5170                 last_ptr = &root->fs_info->data_alloc_cluster;
5171         }
5172
5173         if (last_ptr) {
5174                 spin_lock(&last_ptr->lock);
5175                 if (last_ptr->block_group)
5176                         hint_byte = last_ptr->window_start;
5177                 spin_unlock(&last_ptr->lock);
5178         }
5179
5180         search_start = max(search_start, first_logical_byte(root, 0));
5181         search_start = max(search_start, hint_byte);
5182
5183         if (!last_ptr)
5184                 empty_cluster = 0;
5185
5186         if (search_start == hint_byte) {
5187 ideal_cache:
5188                 block_group = btrfs_lookup_block_group(root->fs_info,
5189                                                        search_start);
5190                 used_block_group = block_group;
5191                 /*
5192                  * we don't want to use the block group if it doesn't match our
5193                  * allocation bits, or if its not cached.
5194                  *
5195                  * However if we are re-searching with an ideal block group
5196                  * picked out then we don't care that the block group is cached.
5197                  */
5198                 if (block_group && block_group_bits(block_group, data) &&
5199                     (block_group->cached != BTRFS_CACHE_NO ||
5200                      search_start == ideal_cache_offset)) {
5201                         down_read(&space_info->groups_sem);
5202                         if (list_empty(&block_group->list) ||
5203                             block_group->ro) {
5204                                 /*
5205                                  * someone is removing this block group,
5206                                  * we can't jump into the have_block_group
5207                                  * target because our list pointers are not
5208                                  * valid
5209                                  */
5210                                 btrfs_put_block_group(block_group);
5211                                 up_read(&space_info->groups_sem);
5212                         } else {
5213                                 index = get_block_group_index(block_group);
5214                                 goto have_block_group;
5215                         }
5216                 } else if (block_group) {
5217                         btrfs_put_block_group(block_group);
5218                 }
5219         }
5220 search:
5221         have_caching_bg = false;
5222         down_read(&space_info->groups_sem);
5223         list_for_each_entry(block_group, &space_info->block_groups[index],
5224                             list) {
5225                 u64 offset;
5226                 int cached;
5227
5228                 used_block_group = block_group;
5229                 btrfs_get_block_group(block_group);
5230                 search_start = block_group->key.objectid;
5231
5232                 /*
5233                  * this can happen if we end up cycling through all the
5234                  * raid types, but we want to make sure we only allocate
5235                  * for the proper type.
5236                  */
5237                 if (!block_group_bits(block_group, data)) {
5238                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
5239                                 BTRFS_BLOCK_GROUP_RAID1 |
5240                                 BTRFS_BLOCK_GROUP_RAID10;
5241
5242                         /*
5243                          * if they asked for extra copies and this block group
5244                          * doesn't provide them, bail.  This does allow us to
5245                          * fill raid0 from raid1.
5246                          */
5247                         if ((data & extra) && !(block_group->flags & extra))
5248                                 goto loop;
5249                 }
5250
5251 have_block_group:
5252                 cached = block_group_cache_done(block_group);
5253                 if (unlikely(!cached)) {
5254                         u64 free_percent;
5255
5256                         found_uncached_bg = true;
5257                         ret = cache_block_group(block_group, trans,
5258                                                 orig_root, 1);
5259                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5260                                 goto alloc;
5261
5262                         free_percent = btrfs_block_group_used(&block_group->item);
5263                         free_percent *= 100;
5264                         free_percent = div64_u64(free_percent,
5265                                                  block_group->key.offset);
5266                         free_percent = 100 - free_percent;
5267                         if (free_percent > ideal_cache_percent &&
5268                             likely(!block_group->ro)) {
5269                                 ideal_cache_offset = block_group->key.objectid;
5270                                 ideal_cache_percent = free_percent;
5271                         }
5272
5273                         /*
5274                          * The caching workers are limited to 2 threads, so we
5275                          * can queue as much work as we care to.
5276                          */
5277                         if (loop > LOOP_FIND_IDEAL) {
5278                                 ret = cache_block_group(block_group, trans,
5279                                                         orig_root, 0);
5280                                 BUG_ON(ret);
5281                         }
5282
5283                         /*
5284                          * If loop is set for cached only, try the next block
5285                          * group.
5286                          */
5287                         if (loop == LOOP_FIND_IDEAL)
5288                                 goto loop;
5289                 }
5290
5291 alloc:
5292                 if (unlikely(block_group->ro))
5293                         goto loop;
5294
5295                 spin_lock(&block_group->free_space_ctl->tree_lock);
5296                 if (cached &&
5297                     block_group->free_space_ctl->free_space <
5298                     num_bytes + empty_cluster + empty_size) {
5299                         spin_unlock(&block_group->free_space_ctl->tree_lock);
5300                         goto loop;
5301                 }
5302                 spin_unlock(&block_group->free_space_ctl->tree_lock);
5303
5304                 /*
5305                  * Ok we want to try and use the cluster allocator, so
5306                  * lets look there
5307                  */
5308                 if (last_ptr) {
5309                         /*
5310                          * the refill lock keeps out other
5311                          * people trying to start a new cluster
5312                          */
5313                         spin_lock(&last_ptr->refill_lock);
5314                         used_block_group = last_ptr->block_group;
5315                         if (used_block_group != block_group &&
5316                             (!used_block_group ||
5317                              used_block_group->ro ||
5318                              !block_group_bits(used_block_group, data))) {
5319                                 used_block_group = block_group;
5320                                 goto refill_cluster;
5321                         }
5322
5323                         if (used_block_group != block_group)
5324                                 btrfs_get_block_group(used_block_group);
5325
5326                         offset = btrfs_alloc_from_cluster(used_block_group,
5327                           last_ptr, num_bytes, used_block_group->key.objectid);
5328                         if (offset) {
5329                                 /* we have a block, we're done */
5330                                 spin_unlock(&last_ptr->refill_lock);
5331                                 goto checks;
5332                         }
5333
5334                         WARN_ON(last_ptr->block_group != used_block_group);
5335                         if (used_block_group != block_group) {
5336                                 btrfs_put_block_group(used_block_group);
5337                                 used_block_group = block_group;
5338                         }
5339 refill_cluster:
5340                         BUG_ON(used_block_group != block_group);
5341                         /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5342                          * set up a new clusters, so lets just skip it
5343                          * and let the allocator find whatever block
5344                          * it can find.  If we reach this point, we
5345                          * will have tried the cluster allocator
5346                          * plenty of times and not have found
5347                          * anything, so we are likely way too
5348                          * fragmented for the clustering stuff to find
5349                          * anything.  */
5350                         if (loop >= LOOP_NO_EMPTY_SIZE) {
5351                                 spin_unlock(&last_ptr->refill_lock);
5352                                 goto unclustered_alloc;
5353                         }
5354
5355                         /*
5356                          * this cluster didn't work out, free it and
5357                          * start over
5358                          */
5359                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5360
5361                         /* allocate a cluster in this block group */
5362                         ret = btrfs_find_space_cluster(trans, root,
5363                                                block_group, last_ptr,
5364                                                search_start, num_bytes,
5365                                                empty_cluster + empty_size);
5366                         if (ret == 0) {
5367                                 /*
5368                                  * now pull our allocation out of this
5369                                  * cluster
5370                                  */
5371                                 offset = btrfs_alloc_from_cluster(block_group,
5372                                                   last_ptr, num_bytes,
5373                                                   search_start);
5374                                 if (offset) {
5375                                         /* we found one, proceed */
5376                                         spin_unlock(&last_ptr->refill_lock);
5377                                         goto checks;
5378                                 }
5379                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5380                                    && !failed_cluster_refill) {
5381                                 spin_unlock(&last_ptr->refill_lock);
5382
5383                                 failed_cluster_refill = true;
5384                                 wait_block_group_cache_progress(block_group,
5385                                        num_bytes + empty_cluster + empty_size);
5386                                 goto have_block_group;
5387                         }
5388
5389                         /*
5390                          * at this point we either didn't find a cluster
5391                          * or we weren't able to allocate a block from our
5392                          * cluster.  Free the cluster we've been trying
5393                          * to use, and go to the next block group
5394                          */
5395                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5396                         spin_unlock(&last_ptr->refill_lock);
5397                         goto loop;
5398                 }
5399
5400 unclustered_alloc:
5401                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5402                                                     num_bytes, empty_size);
5403                 /*
5404                  * If we didn't find a chunk, and we haven't failed on this
5405                  * block group before, and this block group is in the middle of
5406                  * caching and we are ok with waiting, then go ahead and wait
5407                  * for progress to be made, and set failed_alloc to true.
5408                  *
5409                  * If failed_alloc is true then we've already waited on this
5410                  * block group once and should move on to the next block group.
5411                  */
5412                 if (!offset && !failed_alloc && !cached &&
5413                     loop > LOOP_CACHING_NOWAIT) {
5414                         wait_block_group_cache_progress(block_group,
5415                                                 num_bytes + empty_size);
5416                         failed_alloc = true;
5417                         goto have_block_group;
5418                 } else if (!offset) {
5419                         if (!cached)
5420                                 have_caching_bg = true;
5421                         goto loop;
5422                 }
5423 checks:
5424                 search_start = stripe_align(root, offset);
5425                 /* move on to the next group */
5426                 if (search_start + num_bytes >= search_end) {
5427                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5428                         goto loop;
5429                 }
5430
5431                 /* move on to the next group */
5432                 if (search_start + num_bytes >
5433                     used_block_group->key.objectid + used_block_group->key.offset) {
5434                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5435                         goto loop;
5436                 }
5437
5438                 ins->objectid = search_start;
5439                 ins->offset = num_bytes;
5440
5441                 if (offset < search_start)
5442                         btrfs_add_free_space(used_block_group, offset,
5443                                              search_start - offset);
5444                 BUG_ON(offset > search_start);
5445
5446                 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5447                                                   alloc_type);
5448                 if (ret == -EAGAIN) {
5449                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5450                         goto loop;
5451                 }
5452
5453                 /* we are all good, lets return */
5454                 ins->objectid = search_start;
5455                 ins->offset = num_bytes;
5456
5457                 if (offset < search_start)
5458                         btrfs_add_free_space(used_block_group, offset,
5459                                              search_start - offset);
5460                 BUG_ON(offset > search_start);
5461                 if (used_block_group != block_group)
5462                         btrfs_put_block_group(used_block_group);
5463                 btrfs_put_block_group(block_group);
5464                 break;
5465 loop:
5466                 failed_cluster_refill = false;
5467                 failed_alloc = false;
5468                 BUG_ON(index != get_block_group_index(block_group));
5469                 if (used_block_group != block_group)
5470                         btrfs_put_block_group(used_block_group);
5471                 btrfs_put_block_group(block_group);
5472         }
5473         up_read(&space_info->groups_sem);
5474
5475         if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5476                 goto search;
5477
5478         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5479                 goto search;
5480
5481         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5482          *                      for them to make caching progress.  Also
5483          *                      determine the best possible bg to cache
5484          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5485          *                      caching kthreads as we move along
5486          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5487          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5488          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5489          *                      again
5490          */
5491         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5492                 index = 0;
5493                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5494                         found_uncached_bg = false;
5495                         loop++;
5496                         if (!ideal_cache_percent)
5497                                 goto search;
5498
5499                         /*
5500                          * 1 of the following 2 things have happened so far
5501                          *
5502                          * 1) We found an ideal block group for caching that
5503                          * is mostly full and will cache quickly, so we might
5504                          * as well wait for it.
5505                          *
5506                          * 2) We searched for cached only and we didn't find
5507                          * anything, and we didn't start any caching kthreads
5508                          * either, so chances are we will loop through and
5509                          * start a couple caching kthreads, and then come back
5510                          * around and just wait for them.  This will be slower
5511                          * because we will have 2 caching kthreads reading at
5512                          * the same time when we could have just started one
5513                          * and waited for it to get far enough to give us an
5514                          * allocation, so go ahead and go to the wait caching
5515                          * loop.
5516                          */
5517                         loop = LOOP_CACHING_WAIT;
5518                         search_start = ideal_cache_offset;
5519                         ideal_cache_percent = 0;
5520                         goto ideal_cache;
5521                 } else if (loop == LOOP_FIND_IDEAL) {
5522                         /*
5523                          * Didn't find a uncached bg, wait on anything we find
5524                          * next.
5525                          */
5526                         loop = LOOP_CACHING_WAIT;
5527                         goto search;
5528                 }
5529
5530                 loop++;
5531
5532                 if (loop == LOOP_ALLOC_CHUNK) {
5533                        if (allowed_chunk_alloc) {
5534                                 ret = do_chunk_alloc(trans, root, num_bytes +
5535                                                      2 * 1024 * 1024, data,
5536                                                      CHUNK_ALLOC_LIMITED);
5537                                 allowed_chunk_alloc = 0;
5538                                 if (ret == 1)
5539                                         done_chunk_alloc = 1;
5540                         } else if (!done_chunk_alloc &&
5541                                    space_info->force_alloc ==
5542                                    CHUNK_ALLOC_NO_FORCE) {
5543                                 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5544                         }
5545
5546                        /*
5547                         * We didn't allocate a chunk, go ahead and drop the
5548                         * empty size and loop again.
5549                         */
5550                        if (!done_chunk_alloc)
5551                                loop = LOOP_NO_EMPTY_SIZE;
5552                 }
5553
5554                 if (loop == LOOP_NO_EMPTY_SIZE) {
5555                         empty_size = 0;
5556                         empty_cluster = 0;
5557                 }
5558
5559                 goto search;
5560         } else if (!ins->objectid) {
5561                 ret = -ENOSPC;
5562         } else if (ins->objectid) {
5563                 ret = 0;
5564         }
5565
5566         return ret;
5567 }
5568
5569 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5570                             int dump_block_groups)
5571 {
5572         struct btrfs_block_group_cache *cache;
5573         int index = 0;
5574
5575         spin_lock(&info->lock);
5576         printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5577                (unsigned long long)info->flags,
5578                (unsigned long long)(info->total_bytes - info->bytes_used -
5579                                     info->bytes_pinned - info->bytes_reserved -
5580                                     info->bytes_readonly),
5581                (info->full) ? "" : "not ");
5582         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5583                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5584                (unsigned long long)info->total_bytes,
5585                (unsigned long long)info->bytes_used,
5586                (unsigned long long)info->bytes_pinned,
5587                (unsigned long long)info->bytes_reserved,
5588                (unsigned long long)info->bytes_may_use,
5589                (unsigned long long)info->bytes_readonly);
5590         spin_unlock(&info->lock);
5591
5592         if (!dump_block_groups)
5593                 return;
5594
5595         down_read(&info->groups_sem);
5596 again:
5597         list_for_each_entry(cache, &info->block_groups[index], list) {
5598                 spin_lock(&cache->lock);
5599                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5600                        "%llu pinned %llu reserved\n",
5601                        (unsigned long long)cache->key.objectid,
5602                        (unsigned long long)cache->key.offset,
5603                        (unsigned long long)btrfs_block_group_used(&cache->item),
5604                        (unsigned long long)cache->pinned,
5605                        (unsigned long long)cache->reserved);
5606                 btrfs_dump_free_space(cache, bytes);
5607                 spin_unlock(&cache->lock);
5608         }
5609         if (++index < BTRFS_NR_RAID_TYPES)
5610                 goto again;
5611         up_read(&info->groups_sem);
5612 }
5613
5614 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5615                          struct btrfs_root *root,
5616                          u64 num_bytes, u64 min_alloc_size,
5617                          u64 empty_size, u64 hint_byte,
5618                          u64 search_end, struct btrfs_key *ins,
5619                          u64 data)
5620 {
5621         int ret;
5622         u64 search_start = 0;
5623
5624         data = btrfs_get_alloc_profile(root, data);
5625 again:
5626         /*
5627          * the only place that sets empty_size is btrfs_realloc_node, which
5628          * is not called recursively on allocations
5629          */
5630         if (empty_size || root->ref_cows)
5631                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5632                                      num_bytes + 2 * 1024 * 1024, data,
5633                                      CHUNK_ALLOC_NO_FORCE);
5634
5635         WARN_ON(num_bytes < root->sectorsize);
5636         ret = find_free_extent(trans, root, num_bytes, empty_size,
5637                                search_start, search_end, hint_byte,
5638                                ins, data);
5639
5640         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5641                 num_bytes = num_bytes >> 1;
5642                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5643                 num_bytes = max(num_bytes, min_alloc_size);
5644                 do_chunk_alloc(trans, root->fs_info->extent_root,
5645                                num_bytes, data, CHUNK_ALLOC_FORCE);
5646                 goto again;
5647         }
5648         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5649                 struct btrfs_space_info *sinfo;
5650
5651                 sinfo = __find_space_info(root->fs_info, data);
5652                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5653                        "wanted %llu\n", (unsigned long long)data,
5654                        (unsigned long long)num_bytes);
5655                 dump_space_info(sinfo, num_bytes, 1);
5656         }
5657
5658         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5659
5660         return ret;
5661 }
5662
5663 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5664                                         u64 start, u64 len, int pin)
5665 {
5666         struct btrfs_block_group_cache *cache;
5667         int ret = 0;
5668
5669         cache = btrfs_lookup_block_group(root->fs_info, start);
5670         if (!cache) {
5671                 printk(KERN_ERR "Unable to find block group for %llu\n",
5672                        (unsigned long long)start);
5673                 return -ENOSPC;
5674         }
5675
5676         if (btrfs_test_opt(root, DISCARD))
5677                 ret = btrfs_discard_extent(root, start, len, NULL);
5678
5679         if (pin)
5680                 pin_down_extent(root, cache, start, len, 1);
5681         else {
5682                 btrfs_add_free_space(cache, start, len);
5683                 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5684         }
5685         btrfs_put_block_group(cache);
5686
5687         trace_btrfs_reserved_extent_free(root, start, len);
5688
5689         return ret;
5690 }
5691
5692 int btrfs_free_reserved_extent(struct btrfs_root *root,
5693                                         u64 start, u64 len)
5694 {
5695         return __btrfs_free_reserved_extent(root, start, len, 0);
5696 }
5697
5698 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5699                                        u64 start, u64 len)
5700 {
5701         return __btrfs_free_reserved_extent(root, start, len, 1);
5702 }
5703
5704 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5705                                       struct btrfs_root *root,
5706                                       u64 parent, u64 root_objectid,
5707                                       u64 flags, u64 owner, u64 offset,
5708                                       struct btrfs_key *ins, int ref_mod)
5709 {
5710         int ret;
5711         struct btrfs_fs_info *fs_info = root->fs_info;
5712         struct btrfs_extent_item *extent_item;
5713         struct btrfs_extent_inline_ref *iref;
5714         struct btrfs_path *path;
5715         struct extent_buffer *leaf;
5716         int type;
5717         u32 size;
5718
5719         if (parent > 0)
5720                 type = BTRFS_SHARED_DATA_REF_KEY;
5721         else
5722                 type = BTRFS_EXTENT_DATA_REF_KEY;
5723
5724         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5725
5726         path = btrfs_alloc_path();
5727         if (!path)
5728                 return -ENOMEM;
5729
5730         path->leave_spinning = 1;
5731         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5732                                       ins, size);
5733         BUG_ON(ret);
5734
5735         leaf = path->nodes[0];
5736         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5737                                      struct btrfs_extent_item);
5738         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5739         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5740         btrfs_set_extent_flags(leaf, extent_item,
5741                                flags | BTRFS_EXTENT_FLAG_DATA);
5742
5743         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5744         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5745         if (parent > 0) {
5746                 struct btrfs_shared_data_ref *ref;
5747                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5748                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5749                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5750         } else {
5751                 struct btrfs_extent_data_ref *ref;
5752                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5753                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5754                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5755                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5756                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5757         }
5758
5759         btrfs_mark_buffer_dirty(path->nodes[0]);
5760         btrfs_free_path(path);
5761
5762         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5763         if (ret) {
5764                 printk(KERN_ERR "btrfs update block group failed for %llu "
5765                        "%llu\n", (unsigned long long)ins->objectid,
5766                        (unsigned long long)ins->offset);
5767                 BUG();
5768         }
5769         return ret;
5770 }
5771
5772 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5773                                      struct btrfs_root *root,
5774                                      u64 parent, u64 root_objectid,
5775                                      u64 flags, struct btrfs_disk_key *key,
5776                                      int level, struct btrfs_key *ins)
5777 {
5778         int ret;
5779         struct btrfs_fs_info *fs_info = root->fs_info;
5780         struct btrfs_extent_item *extent_item;
5781         struct btrfs_tree_block_info *block_info;
5782         struct btrfs_extent_inline_ref *iref;
5783         struct btrfs_path *path;
5784         struct extent_buffer *leaf;
5785         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5786
5787         path = btrfs_alloc_path();
5788         if (!path)
5789                 return -ENOMEM;
5790
5791         path->leave_spinning = 1;
5792         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5793                                       ins, size);
5794         BUG_ON(ret);
5795
5796         leaf = path->nodes[0];
5797         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5798                                      struct btrfs_extent_item);
5799         btrfs_set_extent_refs(leaf, extent_item, 1);
5800         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5801         btrfs_set_extent_flags(leaf, extent_item,
5802                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5803         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5804
5805         btrfs_set_tree_block_key(leaf, block_info, key);
5806         btrfs_set_tree_block_level(leaf, block_info, level);
5807
5808         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5809         if (parent > 0) {
5810                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5811                 btrfs_set_extent_inline_ref_type(leaf, iref,
5812                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5813                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5814         } else {
5815                 btrfs_set_extent_inline_ref_type(leaf, iref,
5816                                                  BTRFS_TREE_BLOCK_REF_KEY);
5817                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5818         }
5819
5820         btrfs_mark_buffer_dirty(leaf);
5821         btrfs_free_path(path);
5822
5823         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5824         if (ret) {
5825                 printk(KERN_ERR "btrfs update block group failed for %llu "
5826                        "%llu\n", (unsigned long long)ins->objectid,
5827                        (unsigned long long)ins->offset);
5828                 BUG();
5829         }
5830         return ret;
5831 }
5832
5833 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5834                                      struct btrfs_root *root,
5835                                      u64 root_objectid, u64 owner,
5836                                      u64 offset, struct btrfs_key *ins)
5837 {
5838         int ret;
5839
5840         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5841
5842         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5843                                          0, root_objectid, owner, offset,
5844                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5845         return ret;
5846 }
5847
5848 /*
5849  * this is used by the tree logging recovery code.  It records that
5850  * an extent has been allocated and makes sure to clear the free
5851  * space cache bits as well
5852  */
5853 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5854                                    struct btrfs_root *root,
5855                                    u64 root_objectid, u64 owner, u64 offset,
5856                                    struct btrfs_key *ins)
5857 {
5858         int ret;
5859         struct btrfs_block_group_cache *block_group;
5860         struct btrfs_caching_control *caching_ctl;
5861         u64 start = ins->objectid;
5862         u64 num_bytes = ins->offset;
5863
5864         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5865         cache_block_group(block_group, trans, NULL, 0);
5866         caching_ctl = get_caching_control(block_group);
5867
5868         if (!caching_ctl) {
5869                 BUG_ON(!block_group_cache_done(block_group));
5870                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5871                 BUG_ON(ret);
5872         } else {
5873                 mutex_lock(&caching_ctl->mutex);
5874
5875                 if (start >= caching_ctl->progress) {
5876                         ret = add_excluded_extent(root, start, num_bytes);
5877                         BUG_ON(ret);
5878                 } else if (start + num_bytes <= caching_ctl->progress) {
5879                         ret = btrfs_remove_free_space(block_group,
5880                                                       start, num_bytes);
5881                         BUG_ON(ret);
5882                 } else {
5883                         num_bytes = caching_ctl->progress - start;
5884                         ret = btrfs_remove_free_space(block_group,
5885                                                       start, num_bytes);
5886                         BUG_ON(ret);
5887
5888                         start = caching_ctl->progress;
5889                         num_bytes = ins->objectid + ins->offset -
5890                                     caching_ctl->progress;
5891                         ret = add_excluded_extent(root, start, num_bytes);
5892                         BUG_ON(ret);
5893                 }
5894
5895                 mutex_unlock(&caching_ctl->mutex);
5896                 put_caching_control(caching_ctl);
5897         }
5898
5899         ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5900                                           RESERVE_ALLOC_NO_ACCOUNT);
5901         BUG_ON(ret);
5902         btrfs_put_block_group(block_group);
5903         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5904                                          0, owner, offset, ins, 1);
5905         return ret;
5906 }
5907
5908 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5909                                             struct btrfs_root *root,
5910                                             u64 bytenr, u32 blocksize,
5911                                             int level)
5912 {
5913         struct extent_buffer *buf;
5914
5915         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5916         if (!buf)
5917                 return ERR_PTR(-ENOMEM);
5918         btrfs_set_header_generation(buf, trans->transid);
5919         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5920         btrfs_tree_lock(buf);
5921         clean_tree_block(trans, root, buf);
5922
5923         btrfs_set_lock_blocking(buf);
5924         btrfs_set_buffer_uptodate(buf);
5925
5926         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5927                 /*
5928                  * we allow two log transactions at a time, use different
5929                  * EXENT bit to differentiate dirty pages.
5930                  */
5931                 if (root->log_transid % 2 == 0)
5932                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5933                                         buf->start + buf->len - 1, GFP_NOFS);
5934                 else
5935                         set_extent_new(&root->dirty_log_pages, buf->start,
5936                                         buf->start + buf->len - 1, GFP_NOFS);
5937         } else {
5938                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5939                          buf->start + buf->len - 1, GFP_NOFS);
5940         }
5941         trans->blocks_used++;
5942         /* this returns a buffer locked for blocking */
5943         return buf;
5944 }
5945
5946 static struct btrfs_block_rsv *
5947 use_block_rsv(struct btrfs_trans_handle *trans,
5948               struct btrfs_root *root, u32 blocksize)
5949 {
5950         struct btrfs_block_rsv *block_rsv;
5951         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5952         int ret;
5953
5954         block_rsv = get_block_rsv(trans, root);
5955
5956         if (block_rsv->size == 0) {
5957                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5958                 /*
5959                  * If we couldn't reserve metadata bytes try and use some from
5960                  * the global reserve.
5961                  */
5962                 if (ret && block_rsv != global_rsv) {
5963                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5964                         if (!ret)
5965                                 return global_rsv;
5966                         return ERR_PTR(ret);
5967                 } else if (ret) {
5968                         return ERR_PTR(ret);
5969                 }
5970                 return block_rsv;
5971         }
5972
5973         ret = block_rsv_use_bytes(block_rsv, blocksize);
5974         if (!ret)
5975                 return block_rsv;
5976         if (ret) {
5977                 static DEFINE_RATELIMIT_STATE(_rs,
5978                                 DEFAULT_RATELIMIT_INTERVAL,
5979                                 /*DEFAULT_RATELIMIT_BURST*/ 2);
5980                 if (__ratelimit(&_rs)) {
5981                         printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
5982                         WARN_ON(1);
5983                 }
5984                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5985                 if (!ret) {
5986                         return block_rsv;
5987                 } else if (ret && block_rsv != global_rsv) {
5988                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5989                         if (!ret)
5990                                 return global_rsv;
5991                 }
5992         }
5993
5994         return ERR_PTR(-ENOSPC);
5995 }
5996
5997 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5998 {
5999         block_rsv_add_bytes(block_rsv, blocksize, 0);
6000         block_rsv_release_bytes(block_rsv, NULL, 0);
6001 }
6002
6003 /*
6004  * finds a free extent and does all the dirty work required for allocation
6005  * returns the key for the extent through ins, and a tree buffer for
6006  * the first block of the extent through buf.
6007  *
6008  * returns the tree buffer or NULL.
6009  */
6010 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6011                                         struct btrfs_root *root, u32 blocksize,
6012                                         u64 parent, u64 root_objectid,
6013                                         struct btrfs_disk_key *key, int level,
6014                                         u64 hint, u64 empty_size)
6015 {
6016         struct btrfs_key ins;
6017         struct btrfs_block_rsv *block_rsv;
6018         struct extent_buffer *buf;
6019         u64 flags = 0;
6020         int ret;
6021
6022
6023         block_rsv = use_block_rsv(trans, root, blocksize);
6024         if (IS_ERR(block_rsv))
6025                 return ERR_CAST(block_rsv);
6026
6027         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6028                                    empty_size, hint, (u64)-1, &ins, 0);
6029         if (ret) {
6030                 unuse_block_rsv(block_rsv, blocksize);
6031                 return ERR_PTR(ret);
6032         }
6033
6034         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6035                                     blocksize, level);
6036         BUG_ON(IS_ERR(buf));
6037
6038         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6039                 if (parent == 0)
6040                         parent = ins.objectid;
6041                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6042         } else
6043                 BUG_ON(parent > 0);
6044
6045         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6046                 struct btrfs_delayed_extent_op *extent_op;
6047                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6048                 BUG_ON(!extent_op);
6049                 if (key)
6050                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
6051                 else
6052                         memset(&extent_op->key, 0, sizeof(extent_op->key));
6053                 extent_op->flags_to_set = flags;
6054                 extent_op->update_key = 1;
6055                 extent_op->update_flags = 1;
6056                 extent_op->is_data = 0;
6057
6058                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
6059                                         ins.offset, parent, root_objectid,
6060                                         level, BTRFS_ADD_DELAYED_EXTENT,
6061                                         extent_op);
6062                 BUG_ON(ret);
6063         }
6064         return buf;
6065 }
6066
6067 struct walk_control {
6068         u64 refs[BTRFS_MAX_LEVEL];
6069         u64 flags[BTRFS_MAX_LEVEL];
6070         struct btrfs_key update_progress;
6071         int stage;
6072         int level;
6073         int shared_level;
6074         int update_ref;
6075         int keep_locks;
6076         int reada_slot;
6077         int reada_count;
6078 };
6079
6080 #define DROP_REFERENCE  1
6081 #define UPDATE_BACKREF  2
6082
6083 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6084                                      struct btrfs_root *root,
6085                                      struct walk_control *wc,
6086                                      struct btrfs_path *path)
6087 {
6088         u64 bytenr;
6089         u64 generation;
6090         u64 refs;
6091         u64 flags;
6092         u32 nritems;
6093         u32 blocksize;
6094         struct btrfs_key key;
6095         struct extent_buffer *eb;
6096         int ret;
6097         int slot;
6098         int nread = 0;
6099
6100         if (path->slots[wc->level] < wc->reada_slot) {
6101                 wc->reada_count = wc->reada_count * 2 / 3;
6102                 wc->reada_count = max(wc->reada_count, 2);
6103         } else {
6104                 wc->reada_count = wc->reada_count * 3 / 2;
6105                 wc->reada_count = min_t(int, wc->reada_count,
6106                                         BTRFS_NODEPTRS_PER_BLOCK(root));
6107         }
6108
6109         eb = path->nodes[wc->level];
6110         nritems = btrfs_header_nritems(eb);
6111         blocksize = btrfs_level_size(root, wc->level - 1);
6112
6113         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6114                 if (nread >= wc->reada_count)
6115                         break;
6116
6117                 cond_resched();
6118                 bytenr = btrfs_node_blockptr(eb, slot);
6119                 generation = btrfs_node_ptr_generation(eb, slot);
6120
6121                 if (slot == path->slots[wc->level])
6122                         goto reada;
6123
6124                 if (wc->stage == UPDATE_BACKREF &&
6125                     generation <= root->root_key.offset)
6126                         continue;
6127
6128                 /* We don't lock the tree block, it's OK to be racy here */
6129                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6130                                                &refs, &flags);
6131                 BUG_ON(ret);
6132                 BUG_ON(refs == 0);
6133
6134                 if (wc->stage == DROP_REFERENCE) {
6135                         if (refs == 1)
6136                                 goto reada;
6137
6138                         if (wc->level == 1 &&
6139                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6140                                 continue;
6141                         if (!wc->update_ref ||
6142                             generation <= root->root_key.offset)
6143                                 continue;
6144                         btrfs_node_key_to_cpu(eb, &key, slot);
6145                         ret = btrfs_comp_cpu_keys(&key,
6146                                                   &wc->update_progress);
6147                         if (ret < 0)
6148                                 continue;
6149                 } else {
6150                         if (wc->level == 1 &&
6151                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6152                                 continue;
6153                 }
6154 reada:
6155                 ret = readahead_tree_block(root, bytenr, blocksize,
6156                                            generation);
6157                 if (ret)
6158                         break;
6159                 nread++;
6160         }
6161         wc->reada_slot = slot;
6162 }
6163
6164 /*
6165  * hepler to process tree block while walking down the tree.
6166  *
6167  * when wc->stage == UPDATE_BACKREF, this function updates
6168  * back refs for pointers in the block.
6169  *
6170  * NOTE: return value 1 means we should stop walking down.
6171  */
6172 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6173                                    struct btrfs_root *root,
6174                                    struct btrfs_path *path,
6175                                    struct walk_control *wc, int lookup_info)
6176 {
6177         int level = wc->level;
6178         struct extent_buffer *eb = path->nodes[level];
6179         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6180         int ret;
6181
6182         if (wc->stage == UPDATE_BACKREF &&
6183             btrfs_header_owner(eb) != root->root_key.objectid)
6184                 return 1;
6185
6186         /*
6187          * when reference count of tree block is 1, it won't increase
6188          * again. once full backref flag is set, we never clear it.
6189          */
6190         if (lookup_info &&
6191             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6192              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6193                 BUG_ON(!path->locks[level]);
6194                 ret = btrfs_lookup_extent_info(trans, root,
6195                                                eb->start, eb->len,
6196                                                &wc->refs[level],
6197                                                &wc->flags[level]);
6198                 BUG_ON(ret);
6199                 BUG_ON(wc->refs[level] == 0);
6200         }
6201
6202         if (wc->stage == DROP_REFERENCE) {
6203                 if (wc->refs[level] > 1)
6204                         return 1;
6205
6206                 if (path->locks[level] && !wc->keep_locks) {
6207                         btrfs_tree_unlock_rw(eb, path->locks[level]);
6208                         path->locks[level] = 0;
6209                 }
6210                 return 0;
6211         }
6212
6213         /* wc->stage == UPDATE_BACKREF */
6214         if (!(wc->flags[level] & flag)) {
6215                 BUG_ON(!path->locks[level]);
6216                 ret = btrfs_inc_ref(trans, root, eb, 1);
6217                 BUG_ON(ret);
6218                 ret = btrfs_dec_ref(trans, root, eb, 0);
6219                 BUG_ON(ret);
6220                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6221                                                   eb->len, flag, 0);
6222                 BUG_ON(ret);
6223                 wc->flags[level] |= flag;
6224         }
6225
6226         /*
6227          * the block is shared by multiple trees, so it's not good to
6228          * keep the tree lock
6229          */
6230         if (path->locks[level] && level > 0) {
6231                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6232                 path->locks[level] = 0;
6233         }
6234         return 0;
6235 }
6236
6237 /*
6238  * hepler to process tree block pointer.
6239  *
6240  * when wc->stage == DROP_REFERENCE, this function checks
6241  * reference count of the block pointed to. if the block
6242  * is shared and we need update back refs for the subtree
6243  * rooted at the block, this function changes wc->stage to
6244  * UPDATE_BACKREF. if the block is shared and there is no
6245  * need to update back, this function drops the reference
6246  * to the block.
6247  *
6248  * NOTE: return value 1 means we should stop walking down.
6249  */
6250 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6251                                  struct btrfs_root *root,
6252                                  struct btrfs_path *path,
6253                                  struct walk_control *wc, int *lookup_info)
6254 {
6255         u64 bytenr;
6256         u64 generation;
6257         u64 parent;
6258         u32 blocksize;
6259         struct btrfs_key key;
6260         struct extent_buffer *next;
6261         int level = wc->level;
6262         int reada = 0;
6263         int ret = 0;
6264
6265         generation = btrfs_node_ptr_generation(path->nodes[level],
6266                                                path->slots[level]);
6267         /*
6268          * if the lower level block was created before the snapshot
6269          * was created, we know there is no need to update back refs
6270          * for the subtree
6271          */
6272         if (wc->stage == UPDATE_BACKREF &&
6273             generation <= root->root_key.offset) {
6274                 *lookup_info = 1;
6275                 return 1;
6276         }
6277
6278         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6279         blocksize = btrfs_level_size(root, level - 1);
6280
6281         next = btrfs_find_tree_block(root, bytenr, blocksize);
6282         if (!next) {
6283                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6284                 if (!next)
6285                         return -ENOMEM;
6286                 reada = 1;
6287         }
6288         btrfs_tree_lock(next);
6289         btrfs_set_lock_blocking(next);
6290
6291         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6292                                        &wc->refs[level - 1],
6293                                        &wc->flags[level - 1]);
6294         BUG_ON(ret);
6295         BUG_ON(wc->refs[level - 1] == 0);
6296         *lookup_info = 0;
6297
6298         if (wc->stage == DROP_REFERENCE) {
6299                 if (wc->refs[level - 1] > 1) {
6300                         if (level == 1 &&
6301                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6302                                 goto skip;
6303
6304                         if (!wc->update_ref ||
6305                             generation <= root->root_key.offset)
6306                                 goto skip;
6307
6308                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6309                                               path->slots[level]);
6310                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6311                         if (ret < 0)
6312                                 goto skip;
6313
6314                         wc->stage = UPDATE_BACKREF;
6315                         wc->shared_level = level - 1;
6316                 }
6317         } else {
6318                 if (level == 1 &&
6319                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6320                         goto skip;
6321         }
6322
6323         if (!btrfs_buffer_uptodate(next, generation)) {
6324                 btrfs_tree_unlock(next);
6325                 free_extent_buffer(next);
6326                 next = NULL;
6327                 *lookup_info = 1;
6328         }
6329
6330         if (!next) {
6331                 if (reada && level == 1)
6332                         reada_walk_down(trans, root, wc, path);
6333                 next = read_tree_block(root, bytenr, blocksize, generation);
6334                 if (!next)
6335                         return -EIO;
6336                 btrfs_tree_lock(next);
6337                 btrfs_set_lock_blocking(next);
6338         }
6339
6340         level--;
6341         BUG_ON(level != btrfs_header_level(next));
6342         path->nodes[level] = next;
6343         path->slots[level] = 0;
6344         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6345         wc->level = level;
6346         if (wc->level == 1)
6347                 wc->reada_slot = 0;
6348         return 0;
6349 skip:
6350         wc->refs[level - 1] = 0;
6351         wc->flags[level - 1] = 0;
6352         if (wc->stage == DROP_REFERENCE) {
6353                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6354                         parent = path->nodes[level]->start;
6355                 } else {
6356                         BUG_ON(root->root_key.objectid !=
6357                                btrfs_header_owner(path->nodes[level]));
6358                         parent = 0;
6359                 }
6360
6361                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6362                                         root->root_key.objectid, level - 1, 0);
6363                 BUG_ON(ret);
6364         }
6365         btrfs_tree_unlock(next);
6366         free_extent_buffer(next);
6367         *lookup_info = 1;
6368         return 1;
6369 }
6370
6371 /*
6372  * hepler to process tree block while walking up the tree.
6373  *
6374  * when wc->stage == DROP_REFERENCE, this function drops
6375  * reference count on the block.
6376  *
6377  * when wc->stage == UPDATE_BACKREF, this function changes
6378  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6379  * to UPDATE_BACKREF previously while processing the block.
6380  *
6381  * NOTE: return value 1 means we should stop walking up.
6382  */
6383 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6384                                  struct btrfs_root *root,
6385                                  struct btrfs_path *path,
6386                                  struct walk_control *wc)
6387 {
6388         int ret;
6389         int level = wc->level;
6390         struct extent_buffer *eb = path->nodes[level];
6391         u64 parent = 0;
6392
6393         if (wc->stage == UPDATE_BACKREF) {
6394                 BUG_ON(wc->shared_level < level);
6395                 if (level < wc->shared_level)
6396                         goto out;
6397
6398                 ret = find_next_key(path, level + 1, &wc->update_progress);
6399                 if (ret > 0)
6400                         wc->update_ref = 0;
6401
6402                 wc->stage = DROP_REFERENCE;
6403                 wc->shared_level = -1;
6404                 path->slots[level] = 0;
6405
6406                 /*
6407                  * check reference count again if the block isn't locked.
6408                  * we should start walking down the tree again if reference
6409                  * count is one.
6410                  */
6411                 if (!path->locks[level]) {
6412                         BUG_ON(level == 0);
6413                         btrfs_tree_lock(eb);
6414                         btrfs_set_lock_blocking(eb);
6415                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6416
6417                         ret = btrfs_lookup_extent_info(trans, root,
6418                                                        eb->start, eb->len,
6419                                                        &wc->refs[level],
6420                                                        &wc->flags[level]);
6421                         BUG_ON(ret);
6422                         BUG_ON(wc->refs[level] == 0);
6423                         if (wc->refs[level] == 1) {
6424                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6425                                 return 1;
6426                         }
6427                 }
6428         }
6429
6430         /* wc->stage == DROP_REFERENCE */
6431         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6432
6433         if (wc->refs[level] == 1) {
6434                 if (level == 0) {
6435                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6436                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6437                         else
6438                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6439                         BUG_ON(ret);
6440                 }
6441                 /* make block locked assertion in clean_tree_block happy */
6442                 if (!path->locks[level] &&
6443                     btrfs_header_generation(eb) == trans->transid) {
6444                         btrfs_tree_lock(eb);
6445                         btrfs_set_lock_blocking(eb);
6446                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6447                 }
6448                 clean_tree_block(trans, root, eb);
6449         }
6450
6451         if (eb == root->node) {
6452                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6453                         parent = eb->start;
6454                 else
6455                         BUG_ON(root->root_key.objectid !=
6456                                btrfs_header_owner(eb));
6457         } else {
6458                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6459                         parent = path->nodes[level + 1]->start;
6460                 else
6461                         BUG_ON(root->root_key.objectid !=
6462                                btrfs_header_owner(path->nodes[level + 1]));
6463         }
6464
6465         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6466 out:
6467         wc->refs[level] = 0;
6468         wc->flags[level] = 0;
6469         return 0;
6470 }
6471
6472 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6473                                    struct btrfs_root *root,
6474                                    struct btrfs_path *path,
6475                                    struct walk_control *wc)
6476 {
6477         int level = wc->level;
6478         int lookup_info = 1;
6479         int ret;
6480
6481         while (level >= 0) {
6482                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6483                 if (ret > 0)
6484                         break;
6485
6486                 if (level == 0)
6487                         break;
6488
6489                 if (path->slots[level] >=
6490                     btrfs_header_nritems(path->nodes[level]))
6491                         break;
6492
6493                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6494                 if (ret > 0) {
6495                         path->slots[level]++;
6496                         continue;
6497                 } else if (ret < 0)
6498                         return ret;
6499                 level = wc->level;
6500         }
6501         return 0;
6502 }
6503
6504 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6505                                  struct btrfs_root *root,
6506                                  struct btrfs_path *path,
6507                                  struct walk_control *wc, int max_level)
6508 {
6509         int level = wc->level;
6510         int ret;
6511
6512         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6513         while (level < max_level && path->nodes[level]) {
6514                 wc->level = level;
6515                 if (path->slots[level] + 1 <
6516                     btrfs_header_nritems(path->nodes[level])) {
6517                         path->slots[level]++;
6518                         return 0;
6519                 } else {
6520                         ret = walk_up_proc(trans, root, path, wc);
6521                         if (ret > 0)
6522                                 return 0;
6523
6524                         if (path->locks[level]) {
6525                                 btrfs_tree_unlock_rw(path->nodes[level],
6526                                                      path->locks[level]);
6527                                 path->locks[level] = 0;
6528                         }
6529                         free_extent_buffer(path->nodes[level]);
6530                         path->nodes[level] = NULL;
6531                         level++;
6532                 }
6533         }
6534         return 1;
6535 }
6536
6537 /*
6538  * drop a subvolume tree.
6539  *
6540  * this function traverses the tree freeing any blocks that only
6541  * referenced by the tree.
6542  *
6543  * when a shared tree block is found. this function decreases its
6544  * reference count by one. if update_ref is true, this function
6545  * also make sure backrefs for the shared block and all lower level
6546  * blocks are properly updated.
6547  */
6548 void btrfs_drop_snapshot(struct btrfs_root *root,
6549                          struct btrfs_block_rsv *block_rsv, int update_ref)
6550 {
6551         struct btrfs_path *path;
6552         struct btrfs_trans_handle *trans;
6553         struct btrfs_root *tree_root = root->fs_info->tree_root;
6554         struct btrfs_root_item *root_item = &root->root_item;
6555         struct walk_control *wc;
6556         struct btrfs_key key;
6557         int err = 0;
6558         int ret;
6559         int level;
6560
6561         path = btrfs_alloc_path();
6562         if (!path) {
6563                 err = -ENOMEM;
6564                 goto out;
6565         }
6566
6567         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6568         if (!wc) {
6569                 btrfs_free_path(path);
6570                 err = -ENOMEM;
6571                 goto out;
6572         }
6573
6574         trans = btrfs_start_transaction(tree_root, 0);
6575         BUG_ON(IS_ERR(trans));
6576
6577         if (block_rsv)
6578                 trans->block_rsv = block_rsv;
6579
6580         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6581                 level = btrfs_header_level(root->node);
6582                 path->nodes[level] = btrfs_lock_root_node(root);
6583                 btrfs_set_lock_blocking(path->nodes[level]);
6584                 path->slots[level] = 0;
6585                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6586                 memset(&wc->update_progress, 0,
6587                        sizeof(wc->update_progress));
6588         } else {
6589                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6590                 memcpy(&wc->update_progress, &key,
6591                        sizeof(wc->update_progress));
6592
6593                 level = root_item->drop_level;
6594                 BUG_ON(level == 0);
6595                 path->lowest_level = level;
6596                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6597                 path->lowest_level = 0;
6598                 if (ret < 0) {
6599                         err = ret;
6600                         goto out_free;
6601                 }
6602                 WARN_ON(ret > 0);
6603
6604                 /*
6605                  * unlock our path, this is safe because only this
6606                  * function is allowed to delete this snapshot
6607                  */
6608                 btrfs_unlock_up_safe(path, 0);
6609
6610                 level = btrfs_header_level(root->node);
6611                 while (1) {
6612                         btrfs_tree_lock(path->nodes[level]);
6613                         btrfs_set_lock_blocking(path->nodes[level]);
6614
6615                         ret = btrfs_lookup_extent_info(trans, root,
6616                                                 path->nodes[level]->start,
6617                                                 path->nodes[level]->len,
6618                                                 &wc->refs[level],
6619                                                 &wc->flags[level]);
6620                         BUG_ON(ret);
6621                         BUG_ON(wc->refs[level] == 0);
6622
6623                         if (level == root_item->drop_level)
6624                                 break;
6625
6626                         btrfs_tree_unlock(path->nodes[level]);
6627                         WARN_ON(wc->refs[level] != 1);
6628                         level--;
6629                 }
6630         }
6631
6632         wc->level = level;
6633         wc->shared_level = -1;
6634         wc->stage = DROP_REFERENCE;
6635         wc->update_ref = update_ref;
6636         wc->keep_locks = 0;
6637         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6638
6639         while (1) {
6640                 ret = walk_down_tree(trans, root, path, wc);
6641                 if (ret < 0) {
6642                         err = ret;
6643                         break;
6644                 }
6645
6646                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6647                 if (ret < 0) {
6648                         err = ret;
6649                         break;
6650                 }
6651
6652                 if (ret > 0) {
6653                         BUG_ON(wc->stage != DROP_REFERENCE);
6654                         break;
6655                 }
6656
6657                 if (wc->stage == DROP_REFERENCE) {
6658                         level = wc->level;
6659                         btrfs_node_key(path->nodes[level],
6660                                        &root_item->drop_progress,
6661                                        path->slots[level]);
6662                         root_item->drop_level = level;
6663                 }
6664
6665                 BUG_ON(wc->level == 0);
6666                 if (btrfs_should_end_transaction(trans, tree_root)) {
6667                         ret = btrfs_update_root(trans, tree_root,
6668                                                 &root->root_key,
6669                                                 root_item);
6670                         BUG_ON(ret);
6671
6672                         btrfs_end_transaction_throttle(trans, tree_root);
6673                         trans = btrfs_start_transaction(tree_root, 0);
6674                         BUG_ON(IS_ERR(trans));
6675                         if (block_rsv)
6676                                 trans->block_rsv = block_rsv;
6677                 }
6678         }
6679         btrfs_release_path(path);
6680         BUG_ON(err);
6681
6682         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6683         BUG_ON(ret);
6684
6685         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6686                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6687                                            NULL, NULL);
6688                 BUG_ON(ret < 0);
6689                 if (ret > 0) {
6690                         /* if we fail to delete the orphan item this time
6691                          * around, it'll get picked up the next time.
6692                          *
6693                          * The most common failure here is just -ENOENT.
6694                          */
6695                         btrfs_del_orphan_item(trans, tree_root,
6696                                               root->root_key.objectid);
6697                 }
6698         }
6699
6700         if (root->in_radix) {
6701                 btrfs_free_fs_root(tree_root->fs_info, root);
6702         } else {
6703                 free_extent_buffer(root->node);
6704                 free_extent_buffer(root->commit_root);
6705                 kfree(root);
6706         }
6707 out_free:
6708         btrfs_end_transaction_throttle(trans, tree_root);
6709         kfree(wc);
6710         btrfs_free_path(path);
6711 out:
6712         if (err)
6713                 btrfs_std_error(root->fs_info, err);
6714         return;
6715 }
6716
6717 /*
6718  * drop subtree rooted at tree block 'node'.
6719  *
6720  * NOTE: this function will unlock and release tree block 'node'
6721  */
6722 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6723                         struct btrfs_root *root,
6724                         struct extent_buffer *node,
6725                         struct extent_buffer *parent)
6726 {
6727         struct btrfs_path *path;
6728         struct walk_control *wc;
6729         int level;
6730         int parent_level;
6731         int ret = 0;
6732         int wret;
6733
6734         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6735
6736         path = btrfs_alloc_path();
6737         if (!path)
6738                 return -ENOMEM;
6739
6740         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6741         if (!wc) {
6742                 btrfs_free_path(path);
6743                 return -ENOMEM;
6744         }
6745
6746         btrfs_assert_tree_locked(parent);
6747         parent_level = btrfs_header_level(parent);
6748         extent_buffer_get(parent);
6749         path->nodes[parent_level] = parent;
6750         path->slots[parent_level] = btrfs_header_nritems(parent);
6751
6752         btrfs_assert_tree_locked(node);
6753         level = btrfs_header_level(node);
6754         path->nodes[level] = node;
6755         path->slots[level] = 0;
6756         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6757
6758         wc->refs[parent_level] = 1;
6759         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6760         wc->level = level;
6761         wc->shared_level = -1;
6762         wc->stage = DROP_REFERENCE;
6763         wc->update_ref = 0;
6764         wc->keep_locks = 1;
6765         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6766
6767         while (1) {
6768                 wret = walk_down_tree(trans, root, path, wc);
6769                 if (wret < 0) {
6770                         ret = wret;
6771                         break;
6772                 }
6773
6774                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6775                 if (wret < 0)
6776                         ret = wret;
6777                 if (wret != 0)
6778                         break;
6779         }
6780
6781         kfree(wc);
6782         btrfs_free_path(path);
6783         return ret;
6784 }
6785
6786 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6787 {
6788         u64 num_devices;
6789         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6790                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6791
6792         /*
6793          * we add in the count of missing devices because we want
6794          * to make sure that any RAID levels on a degraded FS
6795          * continue to be honored.
6796          */
6797         num_devices = root->fs_info->fs_devices->rw_devices +
6798                 root->fs_info->fs_devices->missing_devices;
6799
6800         if (num_devices == 1) {
6801                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6802                 stripped = flags & ~stripped;
6803
6804                 /* turn raid0 into single device chunks */
6805                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6806                         return stripped;
6807
6808                 /* turn mirroring into duplication */
6809                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6810                              BTRFS_BLOCK_GROUP_RAID10))
6811                         return stripped | BTRFS_BLOCK_GROUP_DUP;
6812                 return flags;
6813         } else {
6814                 /* they already had raid on here, just return */
6815                 if (flags & stripped)
6816                         return flags;
6817
6818                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6819                 stripped = flags & ~stripped;
6820
6821                 /* switch duplicated blocks with raid1 */
6822                 if (flags & BTRFS_BLOCK_GROUP_DUP)
6823                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
6824
6825                 /* turn single device chunks into raid0 */
6826                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6827         }
6828         return flags;
6829 }
6830
6831 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6832 {
6833         struct btrfs_space_info *sinfo = cache->space_info;
6834         u64 num_bytes;
6835         u64 min_allocable_bytes;
6836         int ret = -ENOSPC;
6837
6838
6839         /*
6840          * We need some metadata space and system metadata space for
6841          * allocating chunks in some corner cases until we force to set
6842          * it to be readonly.
6843          */
6844         if ((sinfo->flags &
6845              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6846             !force)
6847                 min_allocable_bytes = 1 * 1024 * 1024;
6848         else
6849                 min_allocable_bytes = 0;
6850
6851         spin_lock(&sinfo->lock);
6852         spin_lock(&cache->lock);
6853
6854         if (cache->ro) {
6855                 ret = 0;
6856                 goto out;
6857         }
6858
6859         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6860                     cache->bytes_super - btrfs_block_group_used(&cache->item);
6861
6862         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6863             sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6864             min_allocable_bytes <= sinfo->total_bytes) {
6865                 sinfo->bytes_readonly += num_bytes;
6866                 cache->ro = 1;
6867                 ret = 0;
6868         }
6869 out:
6870         spin_unlock(&cache->lock);
6871         spin_unlock(&sinfo->lock);
6872         return ret;
6873 }
6874
6875 int btrfs_set_block_group_ro(struct btrfs_root *root,
6876                              struct btrfs_block_group_cache *cache)
6877
6878 {
6879         struct btrfs_trans_handle *trans;
6880         u64 alloc_flags;
6881         int ret;
6882
6883         BUG_ON(cache->ro);
6884
6885         trans = btrfs_join_transaction(root);
6886         BUG_ON(IS_ERR(trans));
6887
6888         alloc_flags = update_block_group_flags(root, cache->flags);
6889         if (alloc_flags != cache->flags)
6890                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6891                                CHUNK_ALLOC_FORCE);
6892
6893         ret = set_block_group_ro(cache, 0);
6894         if (!ret)
6895                 goto out;
6896         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6897         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6898                              CHUNK_ALLOC_FORCE);
6899         if (ret < 0)
6900                 goto out;
6901         ret = set_block_group_ro(cache, 0);
6902 out:
6903         btrfs_end_transaction(trans, root);
6904         return ret;
6905 }
6906
6907 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6908                             struct btrfs_root *root, u64 type)
6909 {
6910         u64 alloc_flags = get_alloc_profile(root, type);
6911         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6912                               CHUNK_ALLOC_FORCE);
6913 }
6914
6915 /*
6916  * helper to account the unused space of all the readonly block group in the
6917  * list. takes mirrors into account.
6918  */
6919 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6920 {
6921         struct btrfs_block_group_cache *block_group;
6922         u64 free_bytes = 0;
6923         int factor;
6924
6925         list_for_each_entry(block_group, groups_list, list) {
6926                 spin_lock(&block_group->lock);
6927
6928                 if (!block_group->ro) {
6929                         spin_unlock(&block_group->lock);
6930                         continue;
6931                 }
6932
6933                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6934                                           BTRFS_BLOCK_GROUP_RAID10 |
6935                                           BTRFS_BLOCK_GROUP_DUP))
6936                         factor = 2;
6937                 else
6938                         factor = 1;
6939
6940                 free_bytes += (block_group->key.offset -
6941                                btrfs_block_group_used(&block_group->item)) *
6942                                factor;
6943
6944                 spin_unlock(&block_group->lock);
6945         }
6946
6947         return free_bytes;
6948 }
6949
6950 /*
6951  * helper to account the unused space of all the readonly block group in the
6952  * space_info. takes mirrors into account.
6953  */
6954 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6955 {
6956         int i;
6957         u64 free_bytes = 0;
6958
6959         spin_lock(&sinfo->lock);
6960
6961         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6962                 if (!list_empty(&sinfo->block_groups[i]))
6963                         free_bytes += __btrfs_get_ro_block_group_free_space(
6964                                                 &sinfo->block_groups[i]);
6965
6966         spin_unlock(&sinfo->lock);
6967
6968         return free_bytes;
6969 }
6970
6971 int btrfs_set_block_group_rw(struct btrfs_root *root,
6972                               struct btrfs_block_group_cache *cache)
6973 {
6974         struct btrfs_space_info *sinfo = cache->space_info;
6975         u64 num_bytes;
6976
6977         BUG_ON(!cache->ro);
6978
6979         spin_lock(&sinfo->lock);
6980         spin_lock(&cache->lock);
6981         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6982                     cache->bytes_super - btrfs_block_group_used(&cache->item);
6983         sinfo->bytes_readonly -= num_bytes;
6984         cache->ro = 0;
6985         spin_unlock(&cache->lock);
6986         spin_unlock(&sinfo->lock);
6987         return 0;
6988 }
6989
6990 /*
6991  * checks to see if its even possible to relocate this block group.
6992  *
6993  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6994  * ok to go ahead and try.
6995  */
6996 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6997 {
6998         struct btrfs_block_group_cache *block_group;
6999         struct btrfs_space_info *space_info;
7000         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7001         struct btrfs_device *device;
7002         u64 min_free;
7003         u64 dev_min = 1;
7004         u64 dev_nr = 0;
7005         int index;
7006         int full = 0;
7007         int ret = 0;
7008
7009         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7010
7011         /* odd, couldn't find the block group, leave it alone */
7012         if (!block_group)
7013                 return -1;
7014
7015         min_free = btrfs_block_group_used(&block_group->item);
7016
7017         /* no bytes used, we're good */
7018         if (!min_free)
7019                 goto out;
7020
7021         space_info = block_group->space_info;
7022         spin_lock(&space_info->lock);
7023
7024         full = space_info->full;
7025
7026         /*
7027          * if this is the last block group we have in this space, we can't
7028          * relocate it unless we're able to allocate a new chunk below.
7029          *
7030          * Otherwise, we need to make sure we have room in the space to handle
7031          * all of the extents from this block group.  If we can, we're good
7032          */
7033         if ((space_info->total_bytes != block_group->key.offset) &&
7034             (space_info->bytes_used + space_info->bytes_reserved +
7035              space_info->bytes_pinned + space_info->bytes_readonly +
7036              min_free < space_info->total_bytes)) {
7037                 spin_unlock(&space_info->lock);
7038                 goto out;
7039         }
7040         spin_unlock(&space_info->lock);
7041
7042         /*
7043          * ok we don't have enough space, but maybe we have free space on our
7044          * devices to allocate new chunks for relocation, so loop through our
7045          * alloc devices and guess if we have enough space.  However, if we
7046          * were marked as full, then we know there aren't enough chunks, and we
7047          * can just return.
7048          */
7049         ret = -1;
7050         if (full)
7051                 goto out;
7052
7053         /*
7054          * index:
7055          *      0: raid10
7056          *      1: raid1
7057          *      2: dup
7058          *      3: raid0
7059          *      4: single
7060          */
7061         index = get_block_group_index(block_group);
7062         if (index == 0) {
7063                 dev_min = 4;
7064                 /* Divide by 2 */
7065                 min_free >>= 1;
7066         } else if (index == 1) {
7067                 dev_min = 2;
7068         } else if (index == 2) {
7069                 /* Multiply by 2 */
7070                 min_free <<= 1;
7071         } else if (index == 3) {
7072                 dev_min = fs_devices->rw_devices;
7073                 do_div(min_free, dev_min);
7074         }
7075
7076         mutex_lock(&root->fs_info->chunk_mutex);
7077         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7078                 u64 dev_offset;
7079
7080                 /*
7081                  * check to make sure we can actually find a chunk with enough
7082                  * space to fit our block group in.
7083                  */
7084                 if (device->total_bytes > device->bytes_used + min_free) {
7085                         ret = find_free_dev_extent(NULL, device, min_free,
7086                                                    &dev_offset, NULL);
7087                         if (!ret)
7088                                 dev_nr++;
7089
7090                         if (dev_nr >= dev_min)
7091                                 break;
7092
7093                         ret = -1;
7094                 }
7095         }
7096         mutex_unlock(&root->fs_info->chunk_mutex);
7097 out:
7098         btrfs_put_block_group(block_group);
7099         return ret;
7100 }
7101
7102 static int find_first_block_group(struct btrfs_root *root,
7103                 struct btrfs_path *path, struct btrfs_key *key)
7104 {
7105         int ret = 0;
7106         struct btrfs_key found_key;
7107         struct extent_buffer *leaf;
7108         int slot;
7109
7110         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7111         if (ret < 0)
7112                 goto out;
7113
7114         while (1) {
7115                 slot = path->slots[0];
7116                 leaf = path->nodes[0];
7117                 if (slot >= btrfs_header_nritems(leaf)) {
7118                         ret = btrfs_next_leaf(root, path);
7119                         if (ret == 0)
7120                                 continue;
7121                         if (ret < 0)
7122                                 goto out;
7123                         break;
7124                 }
7125                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7126
7127                 if (found_key.objectid >= key->objectid &&
7128                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7129                         ret = 0;
7130                         goto out;
7131                 }
7132                 path->slots[0]++;
7133         }
7134 out:
7135         return ret;
7136 }
7137
7138 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7139 {
7140         struct btrfs_block_group_cache *block_group;
7141         u64 last = 0;
7142
7143         while (1) {
7144                 struct inode *inode;
7145
7146                 block_group = btrfs_lookup_first_block_group(info, last);
7147                 while (block_group) {
7148                         spin_lock(&block_group->lock);
7149                         if (block_group->iref)
7150                                 break;
7151                         spin_unlock(&block_group->lock);
7152                         block_group = next_block_group(info->tree_root,
7153                                                        block_group);
7154                 }
7155                 if (!block_group) {
7156                         if (last == 0)
7157                                 break;
7158                         last = 0;
7159                         continue;
7160                 }
7161
7162                 inode = block_group->inode;
7163                 block_group->iref = 0;
7164                 block_group->inode = NULL;
7165                 spin_unlock(&block_group->lock);
7166                 iput(inode);
7167                 last = block_group->key.objectid + block_group->key.offset;
7168                 btrfs_put_block_group(block_group);
7169         }
7170 }
7171
7172 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7173 {
7174         struct btrfs_block_group_cache *block_group;
7175         struct btrfs_space_info *space_info;
7176         struct btrfs_caching_control *caching_ctl;
7177         struct rb_node *n;
7178
7179         down_write(&info->extent_commit_sem);
7180         while (!list_empty(&info->caching_block_groups)) {
7181                 caching_ctl = list_entry(info->caching_block_groups.next,
7182                                          struct btrfs_caching_control, list);
7183                 list_del(&caching_ctl->list);
7184                 put_caching_control(caching_ctl);
7185         }
7186         up_write(&info->extent_commit_sem);
7187
7188         spin_lock(&info->block_group_cache_lock);
7189         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7190                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7191                                        cache_node);
7192                 rb_erase(&block_group->cache_node,
7193                          &info->block_group_cache_tree);
7194                 spin_unlock(&info->block_group_cache_lock);
7195
7196                 down_write(&block_group->space_info->groups_sem);
7197                 list_del(&block_group->list);
7198                 up_write(&block_group->space_info->groups_sem);
7199
7200                 if (block_group->cached == BTRFS_CACHE_STARTED)
7201                         wait_block_group_cache_done(block_group);
7202
7203                 /*
7204                  * We haven't cached this block group, which means we could
7205                  * possibly have excluded extents on this block group.
7206                  */
7207                 if (block_group->cached == BTRFS_CACHE_NO)
7208                         free_excluded_extents(info->extent_root, block_group);
7209
7210                 btrfs_remove_free_space_cache(block_group);
7211                 btrfs_put_block_group(block_group);
7212
7213                 spin_lock(&info->block_group_cache_lock);
7214         }
7215         spin_unlock(&info->block_group_cache_lock);
7216
7217         /* now that all the block groups are freed, go through and
7218          * free all the space_info structs.  This is only called during
7219          * the final stages of unmount, and so we know nobody is
7220          * using them.  We call synchronize_rcu() once before we start,
7221          * just to be on the safe side.
7222          */
7223         synchronize_rcu();
7224
7225         release_global_block_rsv(info);
7226
7227         while(!list_empty(&info->space_info)) {
7228                 space_info = list_entry(info->space_info.next,
7229                                         struct btrfs_space_info,
7230                                         list);
7231                 if (space_info->bytes_pinned > 0 ||
7232                     space_info->bytes_reserved > 0 ||
7233                     space_info->bytes_may_use > 0) {
7234                         WARN_ON(1);
7235                         dump_space_info(space_info, 0, 0);
7236                 }
7237                 list_del(&space_info->list);
7238                 kfree(space_info);
7239         }
7240         return 0;
7241 }
7242
7243 static void __link_block_group(struct btrfs_space_info *space_info,
7244                                struct btrfs_block_group_cache *cache)
7245 {
7246         int index = get_block_group_index(cache);
7247
7248         down_write(&space_info->groups_sem);
7249         list_add_tail(&cache->list, &space_info->block_groups[index]);
7250         up_write(&space_info->groups_sem);
7251 }
7252
7253 int btrfs_read_block_groups(struct btrfs_root *root)
7254 {
7255         struct btrfs_path *path;
7256         int ret;
7257         struct btrfs_block_group_cache *cache;
7258         struct btrfs_fs_info *info = root->fs_info;
7259         struct btrfs_space_info *space_info;
7260         struct btrfs_key key;
7261         struct btrfs_key found_key;
7262         struct extent_buffer *leaf;
7263         int need_clear = 0;
7264         u64 cache_gen;
7265
7266         root = info->extent_root;
7267         key.objectid = 0;
7268         key.offset = 0;
7269         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7270         path = btrfs_alloc_path();
7271         if (!path)
7272                 return -ENOMEM;
7273         path->reada = 1;
7274
7275         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7276         if (btrfs_test_opt(root, SPACE_CACHE) &&
7277             btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7278                 need_clear = 1;
7279         if (btrfs_test_opt(root, CLEAR_CACHE))
7280                 need_clear = 1;
7281
7282         while (1) {
7283                 ret = find_first_block_group(root, path, &key);
7284                 if (ret > 0)
7285                         break;
7286                 if (ret != 0)
7287                         goto error;
7288                 leaf = path->nodes[0];
7289                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7290                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7291                 if (!cache) {
7292                         ret = -ENOMEM;
7293                         goto error;
7294                 }
7295                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7296                                                 GFP_NOFS);
7297                 if (!cache->free_space_ctl) {
7298                         kfree(cache);
7299                         ret = -ENOMEM;
7300                         goto error;
7301                 }
7302
7303                 atomic_set(&cache->count, 1);
7304                 spin_lock_init(&cache->lock);
7305                 cache->fs_info = info;
7306                 INIT_LIST_HEAD(&cache->list);
7307                 INIT_LIST_HEAD(&cache->cluster_list);
7308
7309                 if (need_clear)
7310                         cache->disk_cache_state = BTRFS_DC_CLEAR;
7311
7312                 read_extent_buffer(leaf, &cache->item,
7313                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7314                                    sizeof(cache->item));
7315                 memcpy(&cache->key, &found_key, sizeof(found_key));
7316
7317                 key.objectid = found_key.objectid + found_key.offset;
7318                 btrfs_release_path(path);
7319                 cache->flags = btrfs_block_group_flags(&cache->item);
7320                 cache->sectorsize = root->sectorsize;
7321
7322                 btrfs_init_free_space_ctl(cache);
7323
7324                 /*
7325                  * We need to exclude the super stripes now so that the space
7326                  * info has super bytes accounted for, otherwise we'll think
7327                  * we have more space than we actually do.
7328                  */
7329                 exclude_super_stripes(root, cache);
7330
7331                 /*
7332                  * check for two cases, either we are full, and therefore
7333                  * don't need to bother with the caching work since we won't
7334                  * find any space, or we are empty, and we can just add all
7335                  * the space in and be done with it.  This saves us _alot_ of
7336                  * time, particularly in the full case.
7337                  */
7338                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7339                         cache->last_byte_to_unpin = (u64)-1;
7340                         cache->cached = BTRFS_CACHE_FINISHED;
7341                         free_excluded_extents(root, cache);
7342                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7343                         cache->last_byte_to_unpin = (u64)-1;
7344                         cache->cached = BTRFS_CACHE_FINISHED;
7345                         add_new_free_space(cache, root->fs_info,
7346                                            found_key.objectid,
7347                                            found_key.objectid +
7348                                            found_key.offset);
7349                         free_excluded_extents(root, cache);
7350                 }
7351
7352                 ret = update_space_info(info, cache->flags, found_key.offset,
7353                                         btrfs_block_group_used(&cache->item),
7354                                         &space_info);
7355                 BUG_ON(ret);
7356                 cache->space_info = space_info;
7357                 spin_lock(&cache->space_info->lock);
7358                 cache->space_info->bytes_readonly += cache->bytes_super;
7359                 spin_unlock(&cache->space_info->lock);
7360
7361                 __link_block_group(space_info, cache);
7362
7363                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7364                 BUG_ON(ret);
7365
7366                 set_avail_alloc_bits(root->fs_info, cache->flags);
7367                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7368                         set_block_group_ro(cache, 1);
7369         }
7370
7371         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7372                 if (!(get_alloc_profile(root, space_info->flags) &
7373                       (BTRFS_BLOCK_GROUP_RAID10 |
7374                        BTRFS_BLOCK_GROUP_RAID1 |
7375                        BTRFS_BLOCK_GROUP_DUP)))
7376                         continue;
7377                 /*
7378                  * avoid allocating from un-mirrored block group if there are
7379                  * mirrored block groups.
7380                  */
7381                 list_for_each_entry(cache, &space_info->block_groups[3], list)
7382                         set_block_group_ro(cache, 1);
7383                 list_for_each_entry(cache, &space_info->block_groups[4], list)
7384                         set_block_group_ro(cache, 1);
7385         }
7386
7387         init_global_block_rsv(info);
7388         ret = 0;
7389 error:
7390         btrfs_free_path(path);
7391         return ret;
7392 }
7393
7394 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7395                            struct btrfs_root *root, u64 bytes_used,
7396                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7397                            u64 size)
7398 {
7399         int ret;
7400         struct btrfs_root *extent_root;
7401         struct btrfs_block_group_cache *cache;
7402
7403         extent_root = root->fs_info->extent_root;
7404
7405         root->fs_info->last_trans_log_full_commit = trans->transid;
7406
7407         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7408         if (!cache)
7409                 return -ENOMEM;
7410         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7411                                         GFP_NOFS);
7412         if (!cache->free_space_ctl) {
7413                 kfree(cache);
7414                 return -ENOMEM;
7415         }
7416
7417         cache->key.objectid = chunk_offset;
7418         cache->key.offset = size;
7419         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7420         cache->sectorsize = root->sectorsize;
7421         cache->fs_info = root->fs_info;
7422
7423         atomic_set(&cache->count, 1);
7424         spin_lock_init(&cache->lock);
7425         INIT_LIST_HEAD(&cache->list);
7426         INIT_LIST_HEAD(&cache->cluster_list);
7427
7428         btrfs_init_free_space_ctl(cache);
7429
7430         btrfs_set_block_group_used(&cache->item, bytes_used);
7431         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7432         cache->flags = type;
7433         btrfs_set_block_group_flags(&cache->item, type);
7434
7435         cache->last_byte_to_unpin = (u64)-1;
7436         cache->cached = BTRFS_CACHE_FINISHED;
7437         exclude_super_stripes(root, cache);
7438
7439         add_new_free_space(cache, root->fs_info, chunk_offset,
7440                            chunk_offset + size);
7441
7442         free_excluded_extents(root, cache);
7443
7444         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7445                                 &cache->space_info);
7446         BUG_ON(ret);
7447
7448         spin_lock(&cache->space_info->lock);
7449         cache->space_info->bytes_readonly += cache->bytes_super;
7450         spin_unlock(&cache->space_info->lock);
7451
7452         __link_block_group(cache->space_info, cache);
7453
7454         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7455         BUG_ON(ret);
7456
7457         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7458                                 sizeof(cache->item));
7459         BUG_ON(ret);
7460
7461         set_avail_alloc_bits(extent_root->fs_info, type);
7462
7463         return 0;
7464 }
7465
7466 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7467                              struct btrfs_root *root, u64 group_start)
7468 {
7469         struct btrfs_path *path;
7470         struct btrfs_block_group_cache *block_group;
7471         struct btrfs_free_cluster *cluster;
7472         struct btrfs_root *tree_root = root->fs_info->tree_root;
7473         struct btrfs_key key;
7474         struct inode *inode;
7475         int ret;
7476         int factor;
7477
7478         root = root->fs_info->extent_root;
7479
7480         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7481         BUG_ON(!block_group);
7482         BUG_ON(!block_group->ro);
7483
7484         /*
7485          * Free the reserved super bytes from this block group before
7486          * remove it.
7487          */
7488         free_excluded_extents(root, block_group);
7489
7490         memcpy(&key, &block_group->key, sizeof(key));
7491         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7492                                   BTRFS_BLOCK_GROUP_RAID1 |
7493                                   BTRFS_BLOCK_GROUP_RAID10))
7494                 factor = 2;
7495         else
7496                 factor = 1;
7497
7498         /* make sure this block group isn't part of an allocation cluster */
7499         cluster = &root->fs_info->data_alloc_cluster;
7500         spin_lock(&cluster->refill_lock);
7501         btrfs_return_cluster_to_free_space(block_group, cluster);
7502         spin_unlock(&cluster->refill_lock);
7503
7504         /*
7505          * make sure this block group isn't part of a metadata
7506          * allocation cluster
7507          */
7508         cluster = &root->fs_info->meta_alloc_cluster;
7509         spin_lock(&cluster->refill_lock);
7510         btrfs_return_cluster_to_free_space(block_group, cluster);
7511         spin_unlock(&cluster->refill_lock);
7512
7513         path = btrfs_alloc_path();
7514         if (!path) {
7515                 ret = -ENOMEM;
7516                 goto out;
7517         }
7518
7519         inode = lookup_free_space_inode(tree_root, block_group, path);
7520         if (!IS_ERR(inode)) {
7521                 ret = btrfs_orphan_add(trans, inode);
7522                 BUG_ON(ret);
7523                 clear_nlink(inode);
7524                 /* One for the block groups ref */
7525                 spin_lock(&block_group->lock);
7526                 if (block_group->iref) {
7527                         block_group->iref = 0;
7528                         block_group->inode = NULL;
7529                         spin_unlock(&block_group->lock);
7530                         iput(inode);
7531                 } else {
7532                         spin_unlock(&block_group->lock);
7533                 }
7534                 /* One for our lookup ref */
7535                 btrfs_add_delayed_iput(inode);
7536         }
7537
7538         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7539         key.offset = block_group->key.objectid;
7540         key.type = 0;
7541
7542         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7543         if (ret < 0)
7544                 goto out;
7545         if (ret > 0)
7546                 btrfs_release_path(path);
7547         if (ret == 0) {
7548                 ret = btrfs_del_item(trans, tree_root, path);
7549                 if (ret)
7550                         goto out;
7551                 btrfs_release_path(path);
7552         }
7553
7554         spin_lock(&root->fs_info->block_group_cache_lock);
7555         rb_erase(&block_group->cache_node,
7556                  &root->fs_info->block_group_cache_tree);
7557         spin_unlock(&root->fs_info->block_group_cache_lock);
7558
7559         down_write(&block_group->space_info->groups_sem);
7560         /*
7561          * we must use list_del_init so people can check to see if they
7562          * are still on the list after taking the semaphore
7563          */
7564         list_del_init(&block_group->list);
7565         up_write(&block_group->space_info->groups_sem);
7566
7567         if (block_group->cached == BTRFS_CACHE_STARTED)
7568                 wait_block_group_cache_done(block_group);
7569
7570         btrfs_remove_free_space_cache(block_group);
7571
7572         spin_lock(&block_group->space_info->lock);
7573         block_group->space_info->total_bytes -= block_group->key.offset;
7574         block_group->space_info->bytes_readonly -= block_group->key.offset;
7575         block_group->space_info->disk_total -= block_group->key.offset * factor;
7576         spin_unlock(&block_group->space_info->lock);
7577
7578         memcpy(&key, &block_group->key, sizeof(key));
7579
7580         btrfs_clear_space_info_full(root->fs_info);
7581
7582         btrfs_put_block_group(block_group);
7583         btrfs_put_block_group(block_group);
7584
7585         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7586         if (ret > 0)
7587                 ret = -EIO;
7588         if (ret < 0)
7589                 goto out;
7590
7591         ret = btrfs_del_item(trans, root, path);
7592 out:
7593         btrfs_free_path(path);
7594         return ret;
7595 }
7596
7597 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7598 {
7599         struct btrfs_space_info *space_info;
7600         struct btrfs_super_block *disk_super;
7601         u64 features;
7602         u64 flags;
7603         int mixed = 0;
7604         int ret;
7605
7606         disk_super = fs_info->super_copy;
7607         if (!btrfs_super_root(disk_super))
7608                 return 1;
7609
7610         features = btrfs_super_incompat_flags(disk_super);
7611         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7612                 mixed = 1;
7613
7614         flags = BTRFS_BLOCK_GROUP_SYSTEM;
7615         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7616         if (ret)
7617                 goto out;
7618
7619         if (mixed) {
7620                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7621                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7622         } else {
7623                 flags = BTRFS_BLOCK_GROUP_METADATA;
7624                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7625                 if (ret)
7626                         goto out;
7627
7628                 flags = BTRFS_BLOCK_GROUP_DATA;
7629                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7630         }
7631 out:
7632         return ret;
7633 }
7634
7635 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7636 {
7637         return unpin_extent_range(root, start, end);
7638 }
7639
7640 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7641                                u64 num_bytes, u64 *actual_bytes)
7642 {
7643         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7644 }
7645
7646 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7647 {
7648         struct btrfs_fs_info *fs_info = root->fs_info;
7649         struct btrfs_block_group_cache *cache = NULL;
7650         u64 group_trimmed;
7651         u64 start;
7652         u64 end;
7653         u64 trimmed = 0;
7654         int ret = 0;
7655
7656         cache = btrfs_lookup_block_group(fs_info, range->start);
7657
7658         while (cache) {
7659                 if (cache->key.objectid >= (range->start + range->len)) {
7660                         btrfs_put_block_group(cache);
7661                         break;
7662                 }
7663
7664                 start = max(range->start, cache->key.objectid);
7665                 end = min(range->start + range->len,
7666                                 cache->key.objectid + cache->key.offset);
7667
7668                 if (end - start >= range->minlen) {
7669                         if (!block_group_cache_done(cache)) {
7670                                 ret = cache_block_group(cache, NULL, root, 0);
7671                                 if (!ret)
7672                                         wait_block_group_cache_done(cache);
7673                         }
7674                         ret = btrfs_trim_block_group(cache,
7675                                                      &group_trimmed,
7676                                                      start,
7677                                                      end,
7678                                                      range->minlen);
7679
7680                         trimmed += group_trimmed;
7681                         if (ret) {
7682                                 btrfs_put_block_group(cache);
7683                                 break;
7684                         }
7685                 }
7686
7687                 cache = next_block_group(fs_info->tree_root, cache);
7688         }
7689
7690         range->len = trimmed;
7691         return ret;
7692 }