fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/kthread.h>
  25 #include <linux/slab.h>
  26 #include <linux/ratelimit.h>
  27 #include "compat.h"
  28 #include "hash.h"
  29 #include "ctree.h"
  30 #include "disk-io.h"
  31 #include "print-tree.h"
  32 #include "transaction.h"
  33 #include "volumes.h"
  34 #include "locking.h"
  35 #include "free-space-cache.h"
  36
  37 /* control flags for do_chunk_alloc's force field
  38  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
  39  * if we really need one.
  40  *
  41  * CHUNK_ALLOC_FORCE means it must try to allocate one
  42  *
  43  * CHUNK_ALLOC_LIMITED means to only try and allocate one
  44  * if we have very few chunks already allocated.  This is
  45  * used as part of the clustering code to help make sure
  46  * we have a good pool of storage to cluster in, without
  47  * filling the FS with empty chunks
  48  *
  49  */
  50 enum {
  51         CHUNK_ALLOC_NO_FORCE = 0,
  52         CHUNK_ALLOC_FORCE = 1,
  53         CHUNK_ALLOC_LIMITED = 2,
  54 };
  55
  56 /*
  57  * Control how reservations are dealt with.
  58  *
  59  * RESERVE_FREE - freeing a reservation.
  60  * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
  61  *   ENOSPC accounting
  62  * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
  63  *   bytes_may_use as the ENOSPC accounting is done elsewhere
  64  */
  65 enum {
  66         RESERVE_FREE = 0,
  67         RESERVE_ALLOC = 1,
  68         RESERVE_ALLOC_NO_ACCOUNT = 2,
  69 };
  70
  71 static int update_block_group(struct btrfs_trans_handle *trans,
  72                               struct btrfs_root *root,
  73                               u64 bytenr, u64 num_bytes, int alloc);
  74 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  75                                 struct btrfs_root *root,
  76                                 u64 bytenr, u64 num_bytes, u64 parent,
  77                                 u64 root_objectid, u64 owner_objectid,
  78                                 u64 owner_offset, int refs_to_drop,
  79                                 struct btrfs_delayed_extent_op *extra_op);
  80 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  81                                     struct extent_buffer *leaf,
  82                                     struct btrfs_extent_item *ei);
  83 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  84                                       struct btrfs_root *root,
  85                                       u64 parent, u64 root_objectid,
  86                                       u64 flags, u64 owner, u64 offset,
  87                                       struct btrfs_key *ins, int ref_mod);
  88 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  89                                      struct btrfs_root *root,
  90                                      u64 parent, u64 root_objectid,
  91                                      u64 flags, struct btrfs_disk_key *key,
  92                                      int level, struct btrfs_key *ins);
  93 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  94                           struct btrfs_root *extent_root, u64 alloc_bytes,
  95                           u64 flags, int force);
  96 static int find_next_key(struct btrfs_path *path, int level,
  97                          struct btrfs_key *key);
  98 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  99                             int dump_block_groups);
 100 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
 101                                        u64 num_bytes, int reserve);
 102
 103 static noinline int
 104 block_group_cache_done(struct btrfs_block_group_cache *cache)
 105 {
 106         smp_mb();
 107         return cache->cached == BTRFS_CACHE_FINISHED;
 108 }
 109
 110 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 111 {
 112         return (cache->flags & bits) == bits;
 113 }
 114
 115 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
 116 {
 117         atomic_inc(&cache->count);
 118 }
 119
 120 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 121 {
 122         if (atomic_dec_and_test(&cache->count)) {
 123                 WARN_ON(cache->pinned > 0);
 124                 WARN_ON(cache->reserved > 0);
 125                 kfree(cache->free_space_ctl);
 126                 kfree(cache);
 127         }
 128 }
 129
 130 /*
 131  * this adds the block group to the fs_info rb tree for the block group
 132  * cache
 133  */
 134 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 135                                 struct btrfs_block_group_cache *block_group)
 136 {
 137         struct rb_node **p;
 138         struct rb_node *parent = NULL;
 139         struct btrfs_block_group_cache *cache;
 140
 141         spin_lock(&info->block_group_cache_lock);
 142         p = &info->block_group_cache_tree.rb_node;
 143
 144         while (*p) {
 145                 parent = *p;
 146                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 147                                  cache_node);
 148                 if (block_group->key.objectid < cache->key.objectid) {
 149                         p = &(*p)->rb_left;
 150                 } else if (block_group->key.objectid > cache->key.objectid) {
 151                         p = &(*p)->rb_right;
 152                 } else {
 153                         spin_unlock(&info->block_group_cache_lock);
 154                         return -EEXIST;
 155                 }
 156         }
 157
 158         rb_link_node(&block_group->cache_node, parent, p);
 159         rb_insert_color(&block_group->cache_node,
 160                         &info->block_group_cache_tree);
 161         spin_unlock(&info->block_group_cache_lock);
 162
 163         return 0;
 164 }
 165
 166 /*
 167  * This will return the block group at or after bytenr if contains is 0, else
 168  * it will return the block group that contains the bytenr
 169  */
 170 static struct btrfs_block_group_cache *
 171 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 172                               int contains)
 173 {
 174         struct btrfs_block_group_cache *cache, *ret = NULL;
 175         struct rb_node *n;
 176         u64 end, start;
 177
 178         spin_lock(&info->block_group_cache_lock);
 179         n = info->block_group_cache_tree.rb_node;
 180
 181         while (n) {
 182                 cache = rb_entry(n, struct btrfs_block_group_cache,
 183                                  cache_node);
 184                 end = cache->key.objectid + cache->key.offset - 1;
 185                 start = cache->key.objectid;
 186
 187                 if (bytenr < start) {
 188                         if (!contains && (!ret || start < ret->key.objectid))
 189                                 ret = cache;
 190                         n = n->rb_left;
 191                 } else if (bytenr > start) {
 192                         if (contains && bytenr <= end) {
 193                                 ret = cache;
 194                                 break;
 195                         }
 196                         n = n->rb_right;
 197                 } else {
 198                         ret = cache;
 199                         break;
 200                 }
 201         }
 202         if (ret)
 203                 btrfs_get_block_group(ret);
 204         spin_unlock(&info->block_group_cache_lock);
 205
 206         return ret;
 207 }
 208
 209 static int add_excluded_extent(struct btrfs_root *root,
 210                                u64 start, u64 num_bytes)
 211 {
 212         u64 end = start + num_bytes - 1;
 213         set_extent_bits(&root->fs_info->freed_extents[0],
 214                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 215         set_extent_bits(&root->fs_info->freed_extents[1],
 216                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 217         return 0;
 218 }
 219
 220 static void free_excluded_extents(struct btrfs_root *root,
 221                                   struct btrfs_block_group_cache *cache)
 222 {
 223         u64 start, end;
 224
 225         start = cache->key.objectid;
 226         end = start + cache->key.offset - 1;
 227
 228         clear_extent_bits(&root->fs_info->freed_extents[0],
 229                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 230         clear_extent_bits(&root->fs_info->freed_extents[1],
 231                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 232 }
 233
 234 static int exclude_super_stripes(struct btrfs_root *root,
 235                                  struct btrfs_block_group_cache *cache)
 236 {
 237         u64 bytenr;
 238         u64 *logical;
 239         int stripe_len;
 240         int i, nr, ret;
 241
 242         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 243                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 244                 cache->bytes_super += stripe_len;
 245                 ret = add_excluded_extent(root, cache->key.objectid,
 246                                           stripe_len);
 247                 BUG_ON(ret);
 248         }
 249
 250         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 251                 bytenr = btrfs_sb_offset(i);
 252                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 253                                        cache->key.objectid, bytenr,
 254                                        0, &logical, &nr, &stripe_len);
 255                 BUG_ON(ret);
 256
 257                 while (nr--) {
 258                         cache->bytes_super += stripe_len;
 259                         ret = add_excluded_extent(root, logical[nr],
 260                                                   stripe_len);
 261                         BUG_ON(ret);
 262                 }
 263
 264                 kfree(logical);
 265         }
 266         return 0;
 267 }
 268
 269 static struct btrfs_caching_control *
 270 get_caching_control(struct btrfs_block_group_cache *cache)
 271 {
 272         struct btrfs_caching_control *ctl;
 273
 274         spin_lock(&cache->lock);
 275         if (cache->cached != BTRFS_CACHE_STARTED) {
 276                 spin_unlock(&cache->lock);
 277                 return NULL;
 278         }
 279
 280         /* We're loading it the fast way, so we don't have a caching_ctl. */
 281         if (!cache->caching_ctl) {
 282                 spin_unlock(&cache->lock);
 283                 return NULL;
 284         }
 285
 286         ctl = cache->caching_ctl;
 287         atomic_inc(&ctl->count);
 288         spin_unlock(&cache->lock);
 289         return ctl;
 290 }
 291
 292 static void put_caching_control(struct btrfs_caching_control *ctl)
 293 {
 294         if (atomic_dec_and_test(&ctl->count))
 295                 kfree(ctl);
 296 }
 297
 298 /*
 299  * this is only called by cache_block_group, since we could have freed extents
 300  * we need to check the pinned_extents for any extents that can't be used yet
 301  * since their free space will be released as soon as the transaction commits.
 302  */
 303 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 304                               struct btrfs_fs_info *info, u64 start, u64 end)
 305 {
 306         u64 extent_start, extent_end, size, total_added = 0;
 307         int ret;
 308
 309         while (start < end) {
 310                 ret = find_first_extent_bit(info->pinned_extents, start,
 311                                             &extent_start, &extent_end,
 312                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 313                 if (ret)
 314                         break;
 315
 316                 if (extent_start <= start) {
 317                         start = extent_end + 1;
 318                 } else if (extent_start > start && extent_start < end) {
 319                         size = extent_start - start;
 320                         total_added += size;
 321                         ret = btrfs_add_free_space(block_group, start,
 322                                                    size);
 323                         BUG_ON(ret);
 324                         start = extent_end + 1;
 325                 } else {
 326                         break;
 327                 }
 328         }
 329
 330         if (start < end) {
 331                 size = end - start;
 332                 total_added += size;
 333                 ret = btrfs_add_free_space(block_group, start, size);
 334                 BUG_ON(ret);
 335         }
 336
 337         return total_added;
 338 }
 339
 340 static noinline void caching_thread(struct btrfs_work *work)
 341 {
 342         struct btrfs_block_group_cache *block_group;
 343         struct btrfs_fs_info *fs_info;
 344         struct btrfs_caching_control *caching_ctl;
 345         struct btrfs_root *extent_root;
 346         struct btrfs_path *path;
 347         struct extent_buffer *leaf;
 348         struct btrfs_key key;
 349         u64 total_found = 0;
 350         u64 last = 0;
 351         u32 nritems;
 352         int ret = 0;
 353
 354         caching_ctl = container_of(work, struct btrfs_caching_control, work);
 355         block_group = caching_ctl->block_group;
 356         fs_info = block_group->fs_info;
 357         extent_root = fs_info->extent_root;
 358
 359         path = btrfs_alloc_path();
 360         if (!path)
 361                 goto out;
 362
 363         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 364
 365         /*
 366          * We don't want to deadlock with somebody trying to allocate a new
 367          * extent for the extent root while also trying to search the extent
 368          * root to add free space.  So we skip locking and search the commit
 369          * root, since its read-only
 370          */
 371         path->skip_locking = 1;
 372         path->search_commit_root = 1;
 373         path->reada = 1;
 374
 375         key.objectid = last;
 376         key.offset = 0;
 377         key.type = BTRFS_EXTENT_ITEM_KEY;
 378 again:
 379         mutex_lock(&caching_ctl->mutex);
 380         /* need to make sure the commit_root doesn't disappear */
 381         down_read(&fs_info->extent_commit_sem);
 382
 383         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 384         if (ret < 0)
 385                 goto err;
 386
 387         leaf = path->nodes[0];
 388         nritems = btrfs_header_nritems(leaf);
 389
 390         while (1) {
 391                 if (btrfs_fs_closing(fs_info) > 1) {
 392                         last = (u64)-1;
 393                         break;
 394                 }
 395
 396                 if (path->slots[0] < nritems) {
 397                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 398                 } else {
 399                         ret = find_next_key(path, 0, &key);
 400                         if (ret)
 401                                 break;
 402
 403                         if (need_resched() ||
 404                             btrfs_next_leaf(extent_root, path)) {
 405                                 caching_ctl->progress = last;
 406                                 btrfs_release_path(path);
 407                                 up_read(&fs_info->extent_commit_sem);
 408                                 mutex_unlock(&caching_ctl->mutex);
 409                                 cond_resched();
 410                                 goto again;
 411                         }
 412                         leaf = path->nodes[0];
 413                         nritems = btrfs_header_nritems(leaf);
 414                         continue;
 415                 }
 416
 417                 if (key.objectid < block_group->key.objectid) {
 418                         path->slots[0]++;
 419                         continue;
 420                 }
 421
 422                 if (key.objectid >= block_group->key.objectid +
 423                     block_group->key.offset)
 424                         break;
 425
 426                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 427                         total_found += add_new_free_space(block_group,
 428                                                           fs_info, last,
 429                                                           key.objectid);
 430                         last = key.objectid + key.offset;
 431
 432                         if (total_found > (1024 * 1024 * 2)) {
 433                                 total_found = 0;
 434                                 wake_up(&caching_ctl->wait);
 435                         }
 436                 }
 437                 path->slots[0]++;
 438         }
 439         ret = 0;
 440
 441         total_found += add_new_free_space(block_group, fs_info, last,
 442                                           block_group->key.objectid +
 443                                           block_group->key.offset);
 444         caching_ctl->progress = (u64)-1;
 445
 446         spin_lock(&block_group->lock);
 447         block_group->caching_ctl = NULL;
 448         block_group->cached = BTRFS_CACHE_FINISHED;
 449         spin_unlock(&block_group->lock);
 450
 451 err:
 452         btrfs_free_path(path);
 453         up_read(&fs_info->extent_commit_sem);
 454
 455         free_excluded_extents(extent_root, block_group);
 456
 457         mutex_unlock(&caching_ctl->mutex);
 458 out:
 459         wake_up(&caching_ctl->wait);
 460
 461         put_caching_control(caching_ctl);
 462         btrfs_put_block_group(block_group);
 463 }
 464
 465 static int cache_block_group(struct btrfs_block_group_cache *cache,
 466                              struct btrfs_trans_handle *trans,
 467                              struct btrfs_root *root,
 468                              int load_cache_only)
 469 {
 470         DEFINE_WAIT(wait);
 471         struct btrfs_fs_info *fs_info = cache->fs_info;
 472         struct btrfs_caching_control *caching_ctl;
 473         int ret = 0;
 474
 475         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 476         BUG_ON(!caching_ctl);
 477
 478         INIT_LIST_HEAD(&caching_ctl->list);
 479         mutex_init(&caching_ctl->mutex);
 480         init_waitqueue_head(&caching_ctl->wait);
 481         caching_ctl->block_group = cache;
 482         caching_ctl->progress = cache->key.objectid;
 483         atomic_set(&caching_ctl->count, 1);
 484         caching_ctl->work.func = caching_thread;
 485
 486         spin_lock(&cache->lock);
 487         /*
 488          * This should be a rare occasion, but this could happen I think in the
 489          * case where one thread starts to load the space cache info, and then
 490          * some other thread starts a transaction commit which tries to do an
 491          * allocation while the other thread is still loading the space cache
 492          * info.  The previous loop should have kept us from choosing this block
 493          * group, but if we've moved to the state where we will wait on caching
 494          * block groups we need to first check if we're doing a fast load here,
 495          * so we can wait for it to finish, otherwise we could end up allocating
 496          * from a block group who's cache gets evicted for one reason or
 497          * another.
 498          */
 499         while (cache->cached == BTRFS_CACHE_FAST) {
 500                 struct btrfs_caching_control *ctl;
 501
 502                 ctl = cache->caching_ctl;
 503                 atomic_inc(&ctl->count);
 504                 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
 505                 spin_unlock(&cache->lock);
 506
 507                 schedule();
 508
 509                 finish_wait(&ctl->wait, &wait);
 510                 put_caching_control(ctl);
 511                 spin_lock(&cache->lock);
 512         }
 513
 514         if (cache->cached != BTRFS_CACHE_NO) {
 515                 spin_unlock(&cache->lock);
 516                 kfree(caching_ctl);
 517                 return 0;
 518         }
 519         WARN_ON(cache->caching_ctl);
 520         cache->caching_ctl = caching_ctl;
 521         cache->cached = BTRFS_CACHE_FAST;
 522         spin_unlock(&cache->lock);
 523
 524         /*
 525          * We can't do the read from on-disk cache during a commit since we need
 526          * to have the normal tree locking.  Also if we are currently trying to
 527          * allocate blocks for the tree root we can't do the fast caching since
 528          * we likely hold important locks.
 529          */
 530         if (trans && (!trans->transaction->in_commit) &&
 531             (root && root != root->fs_info->tree_root) &&
 532             btrfs_test_opt(root, SPACE_CACHE)) {
 533                 ret = load_free_space_cache(fs_info, cache);
 534
 535                 spin_lock(&cache->lock);
 536                 if (ret == 1) {
 537                         cache->caching_ctl = NULL;
 538                         cache->cached = BTRFS_CACHE_FINISHED;
 539                         cache->last_byte_to_unpin = (u64)-1;
 540                 } else {
 541                         if (load_cache_only) {
 542                                 cache->caching_ctl = NULL;
 543                                 cache->cached = BTRFS_CACHE_NO;
 544                         } else {
 545                                 cache->cached = BTRFS_CACHE_STARTED;
 546                         }
 547                 }
 548                 spin_unlock(&cache->lock);
 549                 wake_up(&caching_ctl->wait);
 550                 if (ret == 1) {
 551                         put_caching_control(caching_ctl);
 552                         free_excluded_extents(fs_info->extent_root, cache);
 553                         return 0;
 554                 }
 555         } else {
 556                 /*
 557                  * We are not going to do the fast caching, set cached to the
 558                  * appropriate value and wakeup any waiters.
 559                  */
 560                 spin_lock(&cache->lock);
 561                 if (load_cache_only) {
 562                         cache->caching_ctl = NULL;
 563                         cache->cached = BTRFS_CACHE_NO;
 564                 } else {
 565                         cache->cached = BTRFS_CACHE_STARTED;
 566                 }
 567                 spin_unlock(&cache->lock);
 568                 wake_up(&caching_ctl->wait);
 569         }
 570
 571         if (load_cache_only) {
 572                 put_caching_control(caching_ctl);
 573                 return 0;
 574         }
 575
 576         down_write(&fs_info->extent_commit_sem);
 577         atomic_inc(&caching_ctl->count);
 578         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 579         up_write(&fs_info->extent_commit_sem);
 580
 581         btrfs_get_block_group(cache);
 582
 583         btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
 584
 585         return ret;
 586 }
 587
 588 /*
 589  * return the block group that starts at or after bytenr
 590  */
 591 static struct btrfs_block_group_cache *
 592 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 593 {
 594         struct btrfs_block_group_cache *cache;
 595
 596         cache = block_group_cache_tree_search(info, bytenr, 0);
 597
 598         return cache;
 599 }
 600
 601 /*
 602  * return the block group that contains the given bytenr
 603  */
 604 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 605                                                  struct btrfs_fs_info *info,
 606                                                  u64 bytenr)
 607 {
 608         struct btrfs_block_group_cache *cache;
 609
 610         cache = block_group_cache_tree_search(info, bytenr, 1);
 611
 612         return cache;
 613 }
 614
 615 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 616                                                   u64 flags)
 617 {
 618         struct list_head *head = &info->space_info;
 619         struct btrfs_space_info *found;
 620
 621         flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
 622
 623         rcu_read_lock();
 624         list_for_each_entry_rcu(found, head, list) {
 625                 if (found->flags & flags) {
 626                         rcu_read_unlock();
 627                         return found;
 628                 }
 629         }
 630         rcu_read_unlock();
 631         return NULL;
 632 }
 633
 634 /*
 635  * after adding space to the filesystem, we need to clear the full flags
 636  * on all the space infos.
 637  */
 638 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 639 {
 640         struct list_head *head = &info->space_info;
 641         struct btrfs_space_info *found;
 642
 643         rcu_read_lock();
 644         list_for_each_entry_rcu(found, head, list)
 645                 found->full = 0;
 646         rcu_read_unlock();
 647 }
 648
 649 static u64 div_factor(u64 num, int factor)
 650 {
 651         if (factor == 10)
 652                 return num;
 653         num *= factor;
 654         do_div(num, 10);
 655         return num;
 656 }
 657
 658 static u64 div_factor_fine(u64 num, int factor)
 659 {
 660         if (factor == 100)
 661                 return num;
 662         num *= factor;
 663         do_div(num, 100);
 664         return num;
 665 }
 666
 667 u64 btrfs_find_block_group(struct btrfs_root *root,
 668                            u64 search_start, u64 search_hint, int owner)
 669 {
 670         struct btrfs_block_group_cache *cache;
 671         u64 used;
 672         u64 last = max(search_hint, search_start);
 673         u64 group_start = 0;
 674         int full_search = 0;
 675         int factor = 9;
 676         int wrapped = 0;
 677 again:
 678         while (1) {
 679                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 680                 if (!cache)
 681                         break;
 682
 683                 spin_lock(&cache->lock);
 684                 last = cache->key.objectid + cache->key.offset;
 685                 used = btrfs_block_group_used(&cache->item);
 686
 687                 if ((full_search || !cache->ro) &&
 688                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 689                         if (used + cache->pinned + cache->reserved <
 690                             div_factor(cache->key.offset, factor)) {
 691                                 group_start = cache->key.objectid;
 692                                 spin_unlock(&cache->lock);
 693                                 btrfs_put_block_group(cache);
 694                                 goto found;
 695                         }
 696                 }
 697                 spin_unlock(&cache->lock);
 698                 btrfs_put_block_group(cache);
 699                 cond_resched();
 700         }
 701         if (!wrapped) {
 702                 last = search_start;
 703                 wrapped = 1;
 704                 goto again;
 705         }
 706         if (!full_search && factor < 10) {
 707                 last = search_start;
 708                 full_search = 1;
 709                 factor = 10;
 710                 goto again;
 711         }
 712 found:
 713         return group_start;
 714 }
 715
 716 /* simple helper to search for an existing extent at a given offset */
 717 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 718 {
 719         int ret;
 720         struct btrfs_key key;
 721         struct btrfs_path *path;
 722
 723         path = btrfs_alloc_path();
 724         if (!path)
 725                 return -ENOMEM;
 726
 727         key.objectid = start;
 728         key.offset = len;
 729         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 730         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 731                                 0, 0);
 732         btrfs_free_path(path);
 733         return ret;
 734 }
 735
 736 /*
 737  * helper function to lookup reference count and flags of extent.
 738  *
 739  * the head node for delayed ref is used to store the sum of all the
 740  * reference count modifications queued up in the rbtree. the head
 741  * node may also store the extent flags to set. This way you can check
 742  * to see what the reference count and extent flags would be if all of
 743  * the delayed refs are not processed.
 744  */
 745 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 746                              struct btrfs_root *root, u64 bytenr,
 747                              u64 num_bytes, u64 *refs, u64 *flags)
 748 {
 749         struct btrfs_delayed_ref_head *head;
 750         struct btrfs_delayed_ref_root *delayed_refs;
 751         struct btrfs_path *path;
 752         struct btrfs_extent_item *ei;
 753         struct extent_buffer *leaf;
 754         struct btrfs_key key;
 755         u32 item_size;
 756         u64 num_refs;
 757         u64 extent_flags;
 758         int ret;
 759
 760         path = btrfs_alloc_path();
 761         if (!path)
 762                 return -ENOMEM;
 763
 764         key.objectid = bytenr;
 765         key.type = BTRFS_EXTENT_ITEM_KEY;
 766         key.offset = num_bytes;
 767         if (!trans) {
 768                 path->skip_locking = 1;
 769                 path->search_commit_root = 1;
 770         }
 771 again:
 772         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 773                                 &key, path, 0, 0);
 774         if (ret < 0)
 775                 goto out_free;
 776
 777         if (ret == 0) {
 778                 leaf = path->nodes[0];
 779                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 780                 if (item_size >= sizeof(*ei)) {
 781                         ei = btrfs_item_ptr(leaf, path->slots[0],
 782                                             struct btrfs_extent_item);
 783                         num_refs = btrfs_extent_refs(leaf, ei);
 784                         extent_flags = btrfs_extent_flags(leaf, ei);
 785                 } else {
 786 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 787                         struct btrfs_extent_item_v0 *ei0;
 788                         BUG_ON(item_size != sizeof(*ei0));
 789                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 790                                              struct btrfs_extent_item_v0);
 791                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 792                         /* FIXME: this isn't correct for data */
 793                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 794 #else
 795                         BUG();
 796 #endif
 797                 }
 798                 BUG_ON(num_refs == 0);
 799         } else {
 800                 num_refs = 0;
 801                 extent_flags = 0;
 802                 ret = 0;
 803         }
 804
 805         if (!trans)
 806                 goto out;
 807
 808         delayed_refs = &trans->transaction->delayed_refs;
 809         spin_lock(&delayed_refs->lock);
 810         head = btrfs_find_delayed_ref_head(trans, bytenr);
 811         if (head) {
 812                 if (!mutex_trylock(&head->mutex)) {
 813                         atomic_inc(&head->node.refs);
 814                         spin_unlock(&delayed_refs->lock);
 815
 816                         btrfs_release_path(path);
 817
 818                         /*
 819                          * Mutex was contended, block until it's released and try
 820                          * again
 821                          */
 822                         mutex_lock(&head->mutex);
 823                         mutex_unlock(&head->mutex);
 824                         btrfs_put_delayed_ref(&head->node);
 825                         goto again;
 826                 }
 827                 if (head->extent_op && head->extent_op->update_flags)
 828                         extent_flags |= head->extent_op->flags_to_set;
 829                 else
 830                         BUG_ON(num_refs == 0);
 831
 832                 num_refs += head->node.ref_mod;
 833                 mutex_unlock(&head->mutex);
 834         }
 835         spin_unlock(&delayed_refs->lock);
 836 out:
 837         WARN_ON(num_refs == 0);
 838         if (refs)
 839                 *refs = num_refs;
 840         if (flags)
 841                 *flags = extent_flags;
 842 out_free:
 843         btrfs_free_path(path);
 844         return ret;
 845 }
 846
 847 /*
 848  * Back reference rules.  Back refs have three main goals:
 849  *
 850  * 1) differentiate between all holders of references to an extent so that
 851  *    when a reference is dropped we can make sure it was a valid reference
 852  *    before freeing the extent.
 853  *
 854  * 2) Provide enough information to quickly find the holders of an extent
 855  *    if we notice a given block is corrupted or bad.
 856  *
 857  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 858  *    maintenance.  This is actually the same as #2, but with a slightly
 859  *    different use case.
 860  *
 861  * There are two kinds of back refs. The implicit back refs is optimized
 862  * for pointers in non-shared tree blocks. For a given pointer in a block,
 863  * back refs of this kind provide information about the block's owner tree
 864  * and the pointer's key. These information allow us to find the block by
 865  * b-tree searching. The full back refs is for pointers in tree blocks not
 866  * referenced by their owner trees. The location of tree block is recorded
 867  * in the back refs. Actually the full back refs is generic, and can be
 868  * used in all cases the implicit back refs is used. The major shortcoming
 869  * of the full back refs is its overhead. Every time a tree block gets
 870  * COWed, we have to update back refs entry for all pointers in it.
 871  *
 872  * For a newly allocated tree block, we use implicit back refs for
 873  * pointers in it. This means most tree related operations only involve
 874  * implicit back refs. For a tree block created in old transaction, the
 875  * only way to drop a reference to it is COW it. So we can detect the
 876  * event that tree block loses its owner tree's reference and do the
 877  * back refs conversion.
 878  *
 879  * When a tree block is COW'd through a tree, there are four cases:
 880  *
 881  * The reference count of the block is one and the tree is the block's
 882  * owner tree. Nothing to do in this case.
 883  *
 884  * The reference count of the block is one and the tree is not the
 885  * block's owner tree. In this case, full back refs is used for pointers
 886  * in the block. Remove these full back refs, add implicit back refs for
 887  * every pointers in the new block.
 888  *
 889  * The reference count of the block is greater than one and the tree is
 890  * the block's owner tree. In this case, implicit back refs is used for
 891  * pointers in the block. Add full back refs for every pointers in the
 892  * block, increase lower level extents' reference counts. The original
 893  * implicit back refs are entailed to the new block.
 894  *
 895  * The reference count of the block is greater than one and the tree is
 896  * not the block's owner tree. Add implicit back refs for every pointer in
 897  * the new block, increase lower level extents' reference count.
 898  *
 899  * Back Reference Key composing:
 900  *
 901  * The key objectid corresponds to the first byte in the extent,
 902  * The key type is used to differentiate between types of back refs.
 903  * There are different meanings of the key offset for different types
 904  * of back refs.
 905  *
 906  * File extents can be referenced by:
 907  *
 908  * - multiple snapshots, subvolumes, or different generations in one subvol
 909  * - different files inside a single subvolume
 910  * - different offsets inside a file (bookend extents in file.c)
 911  *
 912  * The extent ref structure for the implicit back refs has fields for:
 913  *
 914  * - Objectid of the subvolume root
 915  * - objectid of the file holding the reference
 916  * - original offset in the file
 917  * - how many bookend extents
 918  *
 919  * The key offset for the implicit back refs is hash of the first
 920  * three fields.
 921  *
 922  * The extent ref structure for the full back refs has field for:
 923  *
 924  * - number of pointers in the tree leaf
 925  *
 926  * The key offset for the implicit back refs is the first byte of
 927  * the tree leaf
 928  *
 929  * When a file extent is allocated, The implicit back refs is used.
 930  * the fields are filled in:
 931  *
 932  *     (root_key.objectid, inode objectid, offset in file, 1)
 933  *
 934  * When a file extent is removed file truncation, we find the
 935  * corresponding implicit back refs and check the following fields:
 936  *
 937  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 938  *
 939  * Btree extents can be referenced by:
 940  *
 941  * - Different subvolumes
 942  *
 943  * Both the implicit back refs and the full back refs for tree blocks
 944  * only consist of key. The key offset for the implicit back refs is
 945  * objectid of block's owner tree. The key offset for the full back refs
 946  * is the first byte of parent block.
 947  *
 948  * When implicit back refs is used, information about the lowest key and
 949  * level of the tree block are required. These information are stored in
 950  * tree block info structure.
 951  */
 952
 953 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 954 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 955                                   struct btrfs_root *root,
 956                                   struct btrfs_path *path,
 957                                   u64 owner, u32 extra_size)
 958 {
 959         struct btrfs_extent_item *item;
 960         struct btrfs_extent_item_v0 *ei0;
 961         struct btrfs_extent_ref_v0 *ref0;
 962         struct btrfs_tree_block_info *bi;
 963         struct extent_buffer *leaf;
 964         struct btrfs_key key;
 965         struct btrfs_key found_key;
 966         u32 new_size = sizeof(*item);
 967         u64 refs;
 968         int ret;
 969
 970         leaf = path->nodes[0];
 971         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 972
 973         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 974         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 975                              struct btrfs_extent_item_v0);
 976         refs = btrfs_extent_refs_v0(leaf, ei0);
 977
 978         if (owner == (u64)-1) {
 979                 while (1) {
 980                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 981                                 ret = btrfs_next_leaf(root, path);
 982                                 if (ret < 0)
 983                                         return ret;
 984                                 BUG_ON(ret > 0);
 985                                 leaf = path->nodes[0];
 986                         }
 987                         btrfs_item_key_to_cpu(leaf, &found_key,
 988                                               path->slots[0]);
 989                         BUG_ON(key.objectid != found_key.objectid);
 990                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 991                                 path->slots[0]++;
 992                                 continue;
 993                         }
 994                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 995                                               struct btrfs_extent_ref_v0);
 996                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 997                         break;
 998                 }
 999         }
1000         btrfs_release_path(path);
1001
1002         if (owner < BTRFS_FIRST_FREE_OBJECTID)
1003                 new_size += sizeof(*bi);
1004
1005         new_size -= sizeof(*ei0);
1006         ret = btrfs_search_slot(trans, root, &key, path,
1007                                 new_size + extra_size, 1);
1008         if (ret < 0)
1009                 return ret;
1010         BUG_ON(ret);
1011
1012         ret = btrfs_extend_item(trans, root, path, new_size);
1013
1014         leaf = path->nodes[0];
1015         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1016         btrfs_set_extent_refs(leaf, item, refs);
1017         /* FIXME: get real generation */
1018         btrfs_set_extent_generation(leaf, item, 0);
1019         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1020                 btrfs_set_extent_flags(leaf, item,
1021                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
1022                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
1023                 bi = (struct btrfs_tree_block_info *)(item + 1);
1024                 /* FIXME: get first key of the block */
1025                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1026                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1027         } else {
1028                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1029         }
1030         btrfs_mark_buffer_dirty(leaf);
1031         return 0;
1032 }
1033 #endif
1034
1035 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1036 {
1037         u32 high_crc = ~(u32)0;
1038         u32 low_crc = ~(u32)0;
1039         __le64 lenum;
1040
1041         lenum = cpu_to_le64(root_objectid);
1042         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1043         lenum = cpu_to_le64(owner);
1044         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1045         lenum = cpu_to_le64(offset);
1046         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1047
1048         return ((u64)high_crc << 31) ^ (u64)low_crc;
1049 }
1050
1051 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1052                                      struct btrfs_extent_data_ref *ref)
1053 {
1054         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1055                                     btrfs_extent_data_ref_objectid(leaf, ref),
1056                                     btrfs_extent_data_ref_offset(leaf, ref));
1057 }
1058
1059 static int match_extent_data_ref(struct extent_buffer *leaf,
1060                                  struct btrfs_extent_data_ref *ref,
1061                                  u64 root_objectid, u64 owner, u64 offset)
1062 {
1063         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1064             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1065             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1066                 return 0;
1067         return 1;
1068 }
1069
1070 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1071                                            struct btrfs_root *root,
1072                                            struct btrfs_path *path,
1073                                            u64 bytenr, u64 parent,
1074                                            u64 root_objectid,
1075                                            u64 owner, u64 offset)
1076 {
1077         struct btrfs_key key;
1078         struct btrfs_extent_data_ref *ref;
1079         struct extent_buffer *leaf;
1080         u32 nritems;
1081         int ret;
1082         int recow;
1083         int err = -ENOENT;
1084
1085         key.objectid = bytenr;
1086         if (parent) {
1087                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1088                 key.offset = parent;
1089         } else {
1090                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1091                 key.offset = hash_extent_data_ref(root_objectid,
1092                                                   owner, offset);
1093         }
1094 again:
1095         recow = 0;
1096         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1097         if (ret < 0) {
1098                 err = ret;
1099                 goto fail;
1100         }
1101
1102         if (parent) {
1103                 if (!ret)
1104                         return 0;
1105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1106                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1107                 btrfs_release_path(path);
1108                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1109                 if (ret < 0) {
1110                         err = ret;
1111                         goto fail;
1112                 }
1113                 if (!ret)
1114                         return 0;
1115 #endif
1116                 goto fail;
1117         }
1118
1119         leaf = path->nodes[0];
1120         nritems = btrfs_header_nritems(leaf);
1121         while (1) {
1122                 if (path->slots[0] >= nritems) {
1123                         ret = btrfs_next_leaf(root, path);
1124                         if (ret < 0)
1125                                 err = ret;
1126                         if (ret)
1127                                 goto fail;
1128
1129                         leaf = path->nodes[0];
1130                         nritems = btrfs_header_nritems(leaf);
1131                         recow = 1;
1132                 }
1133
1134                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1135                 if (key.objectid != bytenr ||
1136                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1137                         goto fail;
1138
1139                 ref = btrfs_item_ptr(leaf, path->slots[0],
1140                                      struct btrfs_extent_data_ref);
1141
1142                 if (match_extent_data_ref(leaf, ref, root_objectid,
1143                                           owner, offset)) {
1144                         if (recow) {
1145                                 btrfs_release_path(path);
1146                                 goto again;
1147                         }
1148                         err = 0;
1149                         break;
1150                 }
1151                 path->slots[0]++;
1152         }
1153 fail:
1154         return err;
1155 }
1156
1157 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1158                                            struct btrfs_root *root,
1159                                            struct btrfs_path *path,
1160                                            u64 bytenr, u64 parent,
1161                                            u64 root_objectid, u64 owner,
1162                                            u64 offset, int refs_to_add)
1163 {
1164         struct btrfs_key key;
1165         struct extent_buffer *leaf;
1166         u32 size;
1167         u32 num_refs;
1168         int ret;
1169
1170         key.objectid = bytenr;
1171         if (parent) {
1172                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1173                 key.offset = parent;
1174                 size = sizeof(struct btrfs_shared_data_ref);
1175         } else {
1176                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1177                 key.offset = hash_extent_data_ref(root_objectid,
1178                                                   owner, offset);
1179                 size = sizeof(struct btrfs_extent_data_ref);
1180         }
1181
1182         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1183         if (ret && ret != -EEXIST)
1184                 goto fail;
1185
1186         leaf = path->nodes[0];
1187         if (parent) {
1188                 struct btrfs_shared_data_ref *ref;
1189                 ref = btrfs_item_ptr(leaf, path->slots[0],
1190                                      struct btrfs_shared_data_ref);
1191                 if (ret == 0) {
1192                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1193                 } else {
1194                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1195                         num_refs += refs_to_add;
1196                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1197                 }
1198         } else {
1199                 struct btrfs_extent_data_ref *ref;
1200                 while (ret == -EEXIST) {
1201                         ref = btrfs_item_ptr(leaf, path->slots[0],
1202                                              struct btrfs_extent_data_ref);
1203                         if (match_extent_data_ref(leaf, ref, root_objectid,
1204                                                   owner, offset))
1205                                 break;
1206                         btrfs_release_path(path);
1207                         key.offset++;
1208                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1209                                                       size);
1210                         if (ret && ret != -EEXIST)
1211                                 goto fail;
1212
1213                         leaf = path->nodes[0];
1214                 }
1215                 ref = btrfs_item_ptr(leaf, path->slots[0],
1216                                      struct btrfs_extent_data_ref);
1217                 if (ret == 0) {
1218                         btrfs_set_extent_data_ref_root(leaf, ref,
1219                                                        root_objectid);
1220                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1221                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1222                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1223                 } else {
1224                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1225                         num_refs += refs_to_add;
1226                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1227                 }
1228         }
1229         btrfs_mark_buffer_dirty(leaf);
1230         ret = 0;
1231 fail:
1232         btrfs_release_path(path);
1233         return ret;
1234 }
1235
1236 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1237                                            struct btrfs_root *root,
1238                                            struct btrfs_path *path,
1239                                            int refs_to_drop)
1240 {
1241         struct btrfs_key key;
1242         struct btrfs_extent_data_ref *ref1 = NULL;
1243         struct btrfs_shared_data_ref *ref2 = NULL;
1244         struct extent_buffer *leaf;
1245         u32 num_refs = 0;
1246         int ret = 0;
1247
1248         leaf = path->nodes[0];
1249         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1250
1251         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1252                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1253                                       struct btrfs_extent_data_ref);
1254                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1255         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1256                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1257                                       struct btrfs_shared_data_ref);
1258                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1259 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1260         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1261                 struct btrfs_extent_ref_v0 *ref0;
1262                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1263                                       struct btrfs_extent_ref_v0);
1264                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1265 #endif
1266         } else {
1267                 BUG();
1268         }
1269
1270         BUG_ON(num_refs < refs_to_drop);
1271         num_refs -= refs_to_drop;
1272
1273         if (num_refs == 0) {
1274                 ret = btrfs_del_item(trans, root, path);
1275         } else {
1276                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1277                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1278                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1279                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1281                 else {
1282                         struct btrfs_extent_ref_v0 *ref0;
1283                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1284                                         struct btrfs_extent_ref_v0);
1285                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1286                 }
1287 #endif
1288                 btrfs_mark_buffer_dirty(leaf);
1289         }
1290         return ret;
1291 }
1292
1293 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1294                                           struct btrfs_path *path,
1295                                           struct btrfs_extent_inline_ref *iref)
1296 {
1297         struct btrfs_key key;
1298         struct extent_buffer *leaf;
1299         struct btrfs_extent_data_ref *ref1;
1300         struct btrfs_shared_data_ref *ref2;
1301         u32 num_refs = 0;
1302
1303         leaf = path->nodes[0];
1304         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1305         if (iref) {
1306                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1307                     BTRFS_EXTENT_DATA_REF_KEY) {
1308                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1309                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1310                 } else {
1311                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1312                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1313                 }
1314         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1315                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1316                                       struct btrfs_extent_data_ref);
1317                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1318         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1319                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1320                                       struct btrfs_shared_data_ref);
1321                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1322 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1323         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1324                 struct btrfs_extent_ref_v0 *ref0;
1325                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1326                                       struct btrfs_extent_ref_v0);
1327                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1328 #endif
1329         } else {
1330                 WARN_ON(1);
1331         }
1332         return num_refs;
1333 }
1334
1335 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1336                                           struct btrfs_root *root,
1337                                           struct btrfs_path *path,
1338                                           u64 bytenr, u64 parent,
1339                                           u64 root_objectid)
1340 {
1341         struct btrfs_key key;
1342         int ret;
1343
1344         key.objectid = bytenr;
1345         if (parent) {
1346                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1347                 key.offset = parent;
1348         } else {
1349                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1350                 key.offset = root_objectid;
1351         }
1352
1353         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1354         if (ret > 0)
1355                 ret = -ENOENT;
1356 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1357         if (ret == -ENOENT && parent) {
1358                 btrfs_release_path(path);
1359                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1360                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1361                 if (ret > 0)
1362                         ret = -ENOENT;
1363         }
1364 #endif
1365         return ret;
1366 }
1367
1368 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1369                                           struct btrfs_root *root,
1370                                           struct btrfs_path *path,
1371                                           u64 bytenr, u64 parent,
1372                                           u64 root_objectid)
1373 {
1374         struct btrfs_key key;
1375         int ret;
1376
1377         key.objectid = bytenr;
1378         if (parent) {
1379                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1380                 key.offset = parent;
1381         } else {
1382                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1383                 key.offset = root_objectid;
1384         }
1385
1386         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1387         btrfs_release_path(path);
1388         return ret;
1389 }
1390
1391 static inline int extent_ref_type(u64 parent, u64 owner)
1392 {
1393         int type;
1394         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1395                 if (parent > 0)
1396                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1397                 else
1398                         type = BTRFS_TREE_BLOCK_REF_KEY;
1399         } else {
1400                 if (parent > 0)
1401                         type = BTRFS_SHARED_DATA_REF_KEY;
1402                 else
1403                         type = BTRFS_EXTENT_DATA_REF_KEY;
1404         }
1405         return type;
1406 }
1407
1408 static int find_next_key(struct btrfs_path *path, int level,
1409                          struct btrfs_key *key)
1410
1411 {
1412         for (; level < BTRFS_MAX_LEVEL; level++) {
1413                 if (!path->nodes[level])
1414                         break;
1415                 if (path->slots[level] + 1 >=
1416                     btrfs_header_nritems(path->nodes[level]))
1417                         continue;
1418                 if (level == 0)
1419                         btrfs_item_key_to_cpu(path->nodes[level], key,
1420                                               path->slots[level] + 1);
1421                 else
1422                         btrfs_node_key_to_cpu(path->nodes[level], key,
1423                                               path->slots[level] + 1);
1424                 return 0;
1425         }
1426         return 1;
1427 }
1428
1429 /*
1430  * look for inline back ref. if back ref is found, *ref_ret is set
1431  * to the address of inline back ref, and 0 is returned.
1432  *
1433  * if back ref isn't found, *ref_ret is set to the address where it
1434  * should be inserted, and -ENOENT is returned.
1435  *
1436  * if insert is true and there are too many inline back refs, the path
1437  * points to the extent item, and -EAGAIN is returned.
1438  *
1439  * NOTE: inline back refs are ordered in the same way that back ref
1440  *       items in the tree are ordered.
1441  */
1442 static noinline_for_stack
1443 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1444                                  struct btrfs_root *root,
1445                                  struct btrfs_path *path,
1446                                  struct btrfs_extent_inline_ref **ref_ret,
1447                                  u64 bytenr, u64 num_bytes,
1448                                  u64 parent, u64 root_objectid,
1449                                  u64 owner, u64 offset, int insert)
1450 {
1451         struct btrfs_key key;
1452         struct extent_buffer *leaf;
1453         struct btrfs_extent_item *ei;
1454         struct btrfs_extent_inline_ref *iref;
1455         u64 flags;
1456         u64 item_size;
1457         unsigned long ptr;
1458         unsigned long end;
1459         int extra_size;
1460         int type;
1461         int want;
1462         int ret;
1463         int err = 0;
1464
1465         key.objectid = bytenr;
1466         key.type = BTRFS_EXTENT_ITEM_KEY;
1467         key.offset = num_bytes;
1468
1469         want = extent_ref_type(parent, owner);
1470         if (insert) {
1471                 extra_size = btrfs_extent_inline_ref_size(want);
1472                 path->keep_locks = 1;
1473         } else
1474                 extra_size = -1;
1475         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1476         if (ret < 0) {
1477                 err = ret;
1478                 goto out;
1479         }
1480         BUG_ON(ret);
1481
1482         leaf = path->nodes[0];
1483         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1484 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1485         if (item_size < sizeof(*ei)) {
1486                 if (!insert) {
1487                         err = -ENOENT;
1488                         goto out;
1489                 }
1490                 ret = convert_extent_item_v0(trans, root, path, owner,
1491                                              extra_size);
1492                 if (ret < 0) {
1493                         err = ret;
1494                         goto out;
1495                 }
1496                 leaf = path->nodes[0];
1497                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1498         }
1499 #endif
1500         BUG_ON(item_size < sizeof(*ei));
1501
1502         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1503         flags = btrfs_extent_flags(leaf, ei);
1504
1505         ptr = (unsigned long)(ei + 1);
1506         end = (unsigned long)ei + item_size;
1507
1508         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1509                 ptr += sizeof(struct btrfs_tree_block_info);
1510                 BUG_ON(ptr > end);
1511         } else {
1512                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1513         }
1514
1515         err = -ENOENT;
1516         while (1) {
1517                 if (ptr >= end) {
1518                         WARN_ON(ptr > end);
1519                         break;
1520                 }
1521                 iref = (struct btrfs_extent_inline_ref *)ptr;
1522                 type = btrfs_extent_inline_ref_type(leaf, iref);
1523                 if (want < type)
1524                         break;
1525                 if (want > type) {
1526                         ptr += btrfs_extent_inline_ref_size(type);
1527                         continue;
1528                 }
1529
1530                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1531                         struct btrfs_extent_data_ref *dref;
1532                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1533                         if (match_extent_data_ref(leaf, dref, root_objectid,
1534                                                   owner, offset)) {
1535                                 err = 0;
1536                                 break;
1537                         }
1538                         if (hash_extent_data_ref_item(leaf, dref) <
1539                             hash_extent_data_ref(root_objectid, owner, offset))
1540                                 break;
1541                 } else {
1542                         u64 ref_offset;
1543                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1544                         if (parent > 0) {
1545                                 if (parent == ref_offset) {
1546                                         err = 0;
1547                                         break;
1548                                 }
1549                                 if (ref_offset < parent)
1550                                         break;
1551                         } else {
1552                                 if (root_objectid == ref_offset) {
1553                                         err = 0;
1554                                         break;
1555                                 }
1556                                 if (ref_offset < root_objectid)
1557                                         break;
1558                         }
1559                 }
1560                 ptr += btrfs_extent_inline_ref_size(type);
1561         }
1562         if (err == -ENOENT && insert) {
1563                 if (item_size + extra_size >=
1564                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1565                         err = -EAGAIN;
1566                         goto out;
1567                 }
1568                 /*
1569                  * To add new inline back ref, we have to make sure
1570                  * there is no corresponding back ref item.
1571                  * For simplicity, we just do not add new inline back
1572                  * ref if there is any kind of item for this block
1573                  */
1574                 if (find_next_key(path, 0, &key) == 0 &&
1575                     key.objectid == bytenr &&
1576                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1577                         err = -EAGAIN;
1578                         goto out;
1579                 }
1580         }
1581         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1582 out:
1583         if (insert) {
1584                 path->keep_locks = 0;
1585                 btrfs_unlock_up_safe(path, 1);
1586         }
1587         return err;
1588 }
1589
1590 /*
1591  * helper to add new inline back ref
1592  */
1593 static noinline_for_stack
1594 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1595                                 struct btrfs_root *root,
1596                                 struct btrfs_path *path,
1597                                 struct btrfs_extent_inline_ref *iref,
1598                                 u64 parent, u64 root_objectid,
1599                                 u64 owner, u64 offset, int refs_to_add,
1600                                 struct btrfs_delayed_extent_op *extent_op)
1601 {
1602         struct extent_buffer *leaf;
1603         struct btrfs_extent_item *ei;
1604         unsigned long ptr;
1605         unsigned long end;
1606         unsigned long item_offset;
1607         u64 refs;
1608         int size;
1609         int type;
1610         int ret;
1611
1612         leaf = path->nodes[0];
1613         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1614         item_offset = (unsigned long)iref - (unsigned long)ei;
1615
1616         type = extent_ref_type(parent, owner);
1617         size = btrfs_extent_inline_ref_size(type);
1618
1619         ret = btrfs_extend_item(trans, root, path, size);
1620
1621         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1622         refs = btrfs_extent_refs(leaf, ei);
1623         refs += refs_to_add;
1624         btrfs_set_extent_refs(leaf, ei, refs);
1625         if (extent_op)
1626                 __run_delayed_extent_op(extent_op, leaf, ei);
1627
1628         ptr = (unsigned long)ei + item_offset;
1629         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1630         if (ptr < end - size)
1631                 memmove_extent_buffer(leaf, ptr + size, ptr,
1632                                       end - size - ptr);
1633
1634         iref = (struct btrfs_extent_inline_ref *)ptr;
1635         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1636         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1637                 struct btrfs_extent_data_ref *dref;
1638                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1639                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1640                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1641                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1642                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1643         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1644                 struct btrfs_shared_data_ref *sref;
1645                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1646                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1647                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1648         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1649                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1650         } else {
1651                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1652         }
1653         btrfs_mark_buffer_dirty(leaf);
1654         return 0;
1655 }
1656
1657 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1658                                  struct btrfs_root *root,
1659                                  struct btrfs_path *path,
1660                                  struct btrfs_extent_inline_ref **ref_ret,
1661                                  u64 bytenr, u64 num_bytes, u64 parent,
1662                                  u64 root_objectid, u64 owner, u64 offset)
1663 {
1664         int ret;
1665
1666         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1667                                            bytenr, num_bytes, parent,
1668                                            root_objectid, owner, offset, 0);
1669         if (ret != -ENOENT)
1670                 return ret;
1671
1672         btrfs_release_path(path);
1673         *ref_ret = NULL;
1674
1675         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1676                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1677                                             root_objectid);
1678         } else {
1679                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1680                                              root_objectid, owner, offset);
1681         }
1682         return ret;
1683 }
1684
1685 /*
1686  * helper to update/remove inline back ref
1687  */
1688 static noinline_for_stack
1689 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1690                                  struct btrfs_root *root,
1691                                  struct btrfs_path *path,
1692                                  struct btrfs_extent_inline_ref *iref,
1693                                  int refs_to_mod,
1694                                  struct btrfs_delayed_extent_op *extent_op)
1695 {
1696         struct extent_buffer *leaf;
1697         struct btrfs_extent_item *ei;
1698         struct btrfs_extent_data_ref *dref = NULL;
1699         struct btrfs_shared_data_ref *sref = NULL;
1700         unsigned long ptr;
1701         unsigned long end;
1702         u32 item_size;
1703         int size;
1704         int type;
1705         int ret;
1706         u64 refs;
1707
1708         leaf = path->nodes[0];
1709         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1710         refs = btrfs_extent_refs(leaf, ei);
1711         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1712         refs += refs_to_mod;
1713         btrfs_set_extent_refs(leaf, ei, refs);
1714         if (extent_op)
1715                 __run_delayed_extent_op(extent_op, leaf, ei);
1716
1717         type = btrfs_extent_inline_ref_type(leaf, iref);
1718
1719         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1720                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1721                 refs = btrfs_extent_data_ref_count(leaf, dref);
1722         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1723                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1724                 refs = btrfs_shared_data_ref_count(leaf, sref);
1725         } else {
1726                 refs = 1;
1727                 BUG_ON(refs_to_mod != -1);
1728         }
1729
1730         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1731         refs += refs_to_mod;
1732
1733         if (refs > 0) {
1734                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1735                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1736                 else
1737                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1738         } else {
1739                 size =  btrfs_extent_inline_ref_size(type);
1740                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1741                 ptr = (unsigned long)iref;
1742                 end = (unsigned long)ei + item_size;
1743                 if (ptr + size < end)
1744                         memmove_extent_buffer(leaf, ptr, ptr + size,
1745                                               end - ptr - size);
1746                 item_size -= size;
1747                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1748         }
1749         btrfs_mark_buffer_dirty(leaf);
1750         return 0;
1751 }
1752
1753 static noinline_for_stack
1754 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1755                                  struct btrfs_root *root,
1756                                  struct btrfs_path *path,
1757                                  u64 bytenr, u64 num_bytes, u64 parent,
1758                                  u64 root_objectid, u64 owner,
1759                                  u64 offset, int refs_to_add,
1760                                  struct btrfs_delayed_extent_op *extent_op)
1761 {
1762         struct btrfs_extent_inline_ref *iref;
1763         int ret;
1764
1765         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1766                                            bytenr, num_bytes, parent,
1767                                            root_objectid, owner, offset, 1);
1768         if (ret == 0) {
1769                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1770                 ret = update_inline_extent_backref(trans, root, path, iref,
1771                                                    refs_to_add, extent_op);
1772         } else if (ret == -ENOENT) {
1773                 ret = setup_inline_extent_backref(trans, root, path, iref,
1774                                                   parent, root_objectid,
1775                                                   owner, offset, refs_to_add,
1776                                                   extent_op);
1777         }
1778         return ret;
1779 }
1780
1781 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1782                                  struct btrfs_root *root,
1783                                  struct btrfs_path *path,
1784                                  u64 bytenr, u64 parent, u64 root_objectid,
1785                                  u64 owner, u64 offset, int refs_to_add)
1786 {
1787         int ret;
1788         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1789                 BUG_ON(refs_to_add != 1);
1790                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1791                                             parent, root_objectid);
1792         } else {
1793                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1794                                              parent, root_objectid,
1795                                              owner, offset, refs_to_add);
1796         }
1797         return ret;
1798 }
1799
1800 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1801                                  struct btrfs_root *root,
1802                                  struct btrfs_path *path,
1803                                  struct btrfs_extent_inline_ref *iref,
1804                                  int refs_to_drop, int is_data)
1805 {
1806         int ret;
1807
1808         BUG_ON(!is_data && refs_to_drop != 1);
1809         if (iref) {
1810                 ret = update_inline_extent_backref(trans, root, path, iref,
1811                                                    -refs_to_drop, NULL);
1812         } else if (is_data) {
1813                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1814         } else {
1815                 ret = btrfs_del_item(trans, root, path);
1816         }
1817         return ret;
1818 }
1819
1820 static int btrfs_issue_discard(struct block_device *bdev,
1821                                 u64 start, u64 len)
1822 {
1823         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1824 }
1825
1826 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1827                                 u64 num_bytes, u64 *actual_bytes)
1828 {
1829         int ret;
1830         u64 discarded_bytes = 0;
1831         struct btrfs_bio *bbio = NULL;
1832
1833
1834         /* Tell the block device(s) that the sectors can be discarded */
1835         ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1836                               bytenr, &num_bytes, &bbio, 0);
1837         if (!ret) {
1838                 struct btrfs_bio_stripe *stripe = bbio->stripes;
1839                 int i;
1840
1841
1842                 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1843                         if (!stripe->dev->can_discard)
1844                                 continue;
1845
1846                         ret = btrfs_issue_discard(stripe->dev->bdev,
1847                                                   stripe->physical,
1848                                                   stripe->length);
1849                         if (!ret)
1850                                 discarded_bytes += stripe->length;
1851                         else if (ret != -EOPNOTSUPP)
1852                                 break;
1853
1854                         /*
1855                          * Just in case we get back EOPNOTSUPP for some reason,
1856                          * just ignore the return value so we don't screw up
1857                          * people calling discard_extent.
1858                          */
1859                         ret = 0;
1860                 }
1861                 kfree(bbio);
1862         }
1863
1864         if (actual_bytes)
1865                 *actual_bytes = discarded_bytes;
1866
1867
1868         return ret;
1869 }
1870
1871 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1872                          struct btrfs_root *root,
1873                          u64 bytenr, u64 num_bytes, u64 parent,
1874                          u64 root_objectid, u64 owner, u64 offset, int for_cow)
1875 {
1876         int ret;
1877         struct btrfs_fs_info *fs_info = root->fs_info;
1878
1879         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1880                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1881
1882         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1883                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1884                                         num_bytes,
1885                                         parent, root_objectid, (int)owner,
1886                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1887         } else {
1888                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1889                                         num_bytes,
1890                                         parent, root_objectid, owner, offset,
1891                                         BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1892         }
1893         return ret;
1894 }
1895
1896 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1897                                   struct btrfs_root *root,
1898                                   u64 bytenr, u64 num_bytes,
1899                                   u64 parent, u64 root_objectid,
1900                                   u64 owner, u64 offset, int refs_to_add,
1901                                   struct btrfs_delayed_extent_op *extent_op)
1902 {
1903         struct btrfs_path *path;
1904         struct extent_buffer *leaf;
1905         struct btrfs_extent_item *item;
1906         u64 refs;
1907         int ret;
1908         int err = 0;
1909
1910         path = btrfs_alloc_path();
1911         if (!path)
1912                 return -ENOMEM;
1913
1914         path->reada = 1;
1915         path->leave_spinning = 1;
1916         /* this will setup the path even if it fails to insert the back ref */
1917         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1918                                            path, bytenr, num_bytes, parent,
1919                                            root_objectid, owner, offset,
1920                                            refs_to_add, extent_op);
1921         if (ret == 0)
1922                 goto out;
1923
1924         if (ret != -EAGAIN) {
1925                 err = ret;
1926                 goto out;
1927         }
1928
1929         leaf = path->nodes[0];
1930         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1931         refs = btrfs_extent_refs(leaf, item);
1932         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1933         if (extent_op)
1934                 __run_delayed_extent_op(extent_op, leaf, item);
1935
1936         btrfs_mark_buffer_dirty(leaf);
1937         btrfs_release_path(path);
1938
1939         path->reada = 1;
1940         path->leave_spinning = 1;
1941
1942         /* now insert the actual backref */
1943         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1944                                     path, bytenr, parent, root_objectid,
1945                                     owner, offset, refs_to_add);
1946         BUG_ON(ret);
1947 out:
1948         btrfs_free_path(path);
1949         return err;
1950 }
1951
1952 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1953                                 struct btrfs_root *root,
1954                                 struct btrfs_delayed_ref_node *node,
1955                                 struct btrfs_delayed_extent_op *extent_op,
1956                                 int insert_reserved)
1957 {
1958         int ret = 0;
1959         struct btrfs_delayed_data_ref *ref;
1960         struct btrfs_key ins;
1961         u64 parent = 0;
1962         u64 ref_root = 0;
1963         u64 flags = 0;
1964
1965         ins.objectid = node->bytenr;
1966         ins.offset = node->num_bytes;
1967         ins.type = BTRFS_EXTENT_ITEM_KEY;
1968
1969         ref = btrfs_delayed_node_to_data_ref(node);
1970         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1971                 parent = ref->parent;
1972         else
1973                 ref_root = ref->root;
1974
1975         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1976                 if (extent_op) {
1977                         BUG_ON(extent_op->update_key);
1978                         flags |= extent_op->flags_to_set;
1979                 }
1980                 ret = alloc_reserved_file_extent(trans, root,
1981                                                  parent, ref_root, flags,
1982                                                  ref->objectid, ref->offset,
1983                                                  &ins, node->ref_mod);
1984         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1985                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1986                                              node->num_bytes, parent,
1987                                              ref_root, ref->objectid,
1988                                              ref->offset, node->ref_mod,
1989                                              extent_op);
1990         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1991                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1992                                           node->num_bytes, parent,
1993                                           ref_root, ref->objectid,
1994                                           ref->offset, node->ref_mod,
1995                                           extent_op);
1996         } else {
1997                 BUG();
1998         }
1999         return ret;
2000 }
2001
2002 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2003                                     struct extent_buffer *leaf,
2004                                     struct btrfs_extent_item *ei)
2005 {
2006         u64 flags = btrfs_extent_flags(leaf, ei);
2007         if (extent_op->update_flags) {
2008                 flags |= extent_op->flags_to_set;
2009                 btrfs_set_extent_flags(leaf, ei, flags);
2010         }
2011
2012         if (extent_op->update_key) {
2013                 struct btrfs_tree_block_info *bi;
2014                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2015                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2016                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2017         }
2018 }
2019
2020 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2021                                  struct btrfs_root *root,
2022                                  struct btrfs_delayed_ref_node *node,
2023                                  struct btrfs_delayed_extent_op *extent_op)
2024 {
2025         struct btrfs_key key;
2026         struct btrfs_path *path;
2027         struct btrfs_extent_item *ei;
2028         struct extent_buffer *leaf;
2029         u32 item_size;
2030         int ret;
2031         int err = 0;
2032
2033         path = btrfs_alloc_path();
2034         if (!path)
2035                 return -ENOMEM;
2036
2037         key.objectid = node->bytenr;
2038         key.type = BTRFS_EXTENT_ITEM_KEY;
2039         key.offset = node->num_bytes;
2040
2041         path->reada = 1;
2042         path->leave_spinning = 1;
2043         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2044                                 path, 0, 1);
2045         if (ret < 0) {
2046                 err = ret;
2047                 goto out;
2048         }
2049         if (ret > 0) {
2050                 err = -EIO;
2051                 goto out;
2052         }
2053
2054         leaf = path->nodes[0];
2055         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2056 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2057         if (item_size < sizeof(*ei)) {
2058                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2059                                              path, (u64)-1, 0);
2060                 if (ret < 0) {
2061                         err = ret;
2062                         goto out;
2063                 }
2064                 leaf = path->nodes[0];
2065                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2066         }
2067 #endif
2068         BUG_ON(item_size < sizeof(*ei));
2069         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2070         __run_delayed_extent_op(extent_op, leaf, ei);
2071
2072         btrfs_mark_buffer_dirty(leaf);
2073 out:
2074         btrfs_free_path(path);
2075         return err;
2076 }
2077
2078 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2079                                 struct btrfs_root *root,
2080                                 struct btrfs_delayed_ref_node *node,
2081                                 struct btrfs_delayed_extent_op *extent_op,
2082                                 int insert_reserved)
2083 {
2084         int ret = 0;
2085         struct btrfs_delayed_tree_ref *ref;
2086         struct btrfs_key ins;
2087         u64 parent = 0;
2088         u64 ref_root = 0;
2089
2090         ins.objectid = node->bytenr;
2091         ins.offset = node->num_bytes;
2092         ins.type = BTRFS_EXTENT_ITEM_KEY;
2093
2094         ref = btrfs_delayed_node_to_tree_ref(node);
2095         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2096                 parent = ref->parent;
2097         else
2098                 ref_root = ref->root;
2099
2100         BUG_ON(node->ref_mod != 1);
2101         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2102                 BUG_ON(!extent_op || !extent_op->update_flags ||
2103                        !extent_op->update_key);
2104                 ret = alloc_reserved_tree_block(trans, root,
2105                                                 parent, ref_root,
2106                                                 extent_op->flags_to_set,
2107                                                 &extent_op->key,
2108                                                 ref->level, &ins);
2109         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2110                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2111                                              node->num_bytes, parent, ref_root,
2112                                              ref->level, 0, 1, extent_op);
2113         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2114                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2115                                           node->num_bytes, parent, ref_root,
2116                                           ref->level, 0, 1, extent_op);
2117         } else {
2118                 BUG();
2119         }
2120         return ret;
2121 }
2122
2123 /* helper function to actually process a single delayed ref entry */
2124 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2125                                struct btrfs_root *root,
2126                                struct btrfs_delayed_ref_node *node,
2127                                struct btrfs_delayed_extent_op *extent_op,
2128                                int insert_reserved)
2129 {
2130         int ret;
2131         if (btrfs_delayed_ref_is_head(node)) {
2132                 struct btrfs_delayed_ref_head *head;
2133                 /*
2134                  * we've hit the end of the chain and we were supposed
2135                  * to insert this extent into the tree.  But, it got
2136                  * deleted before we ever needed to insert it, so all
2137                  * we have to do is clean up the accounting
2138                  */
2139                 BUG_ON(extent_op);
2140                 head = btrfs_delayed_node_to_head(node);
2141                 if (insert_reserved) {
2142                         btrfs_pin_extent(root, node->bytenr,
2143                                          node->num_bytes, 1);
2144                         if (head->is_data) {
2145                                 ret = btrfs_del_csums(trans, root,
2146                                                       node->bytenr,
2147                                                       node->num_bytes);
2148                                 BUG_ON(ret);
2149                         }
2150                 }
2151                 mutex_unlock(&head->mutex);
2152                 return 0;
2153         }
2154
2155         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2156             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2157                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2158                                            insert_reserved);
2159         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2160                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2161                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2162                                            insert_reserved);
2163         else
2164                 BUG();
2165         return ret;
2166 }
2167
2168 static noinline struct btrfs_delayed_ref_node *
2169 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2170 {
2171         struct rb_node *node;
2172         struct btrfs_delayed_ref_node *ref;
2173         int action = BTRFS_ADD_DELAYED_REF;
2174 again:
2175         /*
2176          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2177          * this prevents ref count from going down to zero when
2178          * there still are pending delayed ref.
2179          */
2180         node = rb_prev(&head->node.rb_node);
2181         while (1) {
2182                 if (!node)
2183                         break;
2184                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2185                                 rb_node);
2186                 if (ref->bytenr != head->node.bytenr)
2187                         break;
2188                 if (ref->action == action)
2189                         return ref;
2190                 node = rb_prev(node);
2191         }
2192         if (action == BTRFS_ADD_DELAYED_REF) {
2193                 action = BTRFS_DROP_DELAYED_REF;
2194                 goto again;
2195         }
2196         return NULL;
2197 }
2198
2199 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2200                                        struct btrfs_root *root,
2201                                        struct list_head *cluster)
2202 {
2203         struct btrfs_delayed_ref_root *delayed_refs;
2204         struct btrfs_delayed_ref_node *ref;
2205         struct btrfs_delayed_ref_head *locked_ref = NULL;
2206         struct btrfs_delayed_extent_op *extent_op;
2207         int ret;
2208         int count = 0;
2209         int must_insert_reserved = 0;
2210
2211         delayed_refs = &trans->transaction->delayed_refs;
2212         while (1) {
2213                 if (!locked_ref) {
2214                         /* pick a new head ref from the cluster list */
2215                         if (list_empty(cluster))
2216                                 break;
2217
2218                         locked_ref = list_entry(cluster->next,
2219                                      struct btrfs_delayed_ref_head, cluster);
2220
2221                         /* grab the lock that says we are going to process
2222                          * all the refs for this head */
2223                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2224
2225                         /*
2226                          * we may have dropped the spin lock to get the head
2227                          * mutex lock, and that might have given someone else
2228                          * time to free the head.  If that's true, it has been
2229                          * removed from our list and we can move on.
2230                          */
2231                         if (ret == -EAGAIN) {
2232                                 locked_ref = NULL;
2233                                 count++;
2234                                 continue;
2235                         }
2236                 }
2237
2238                 /*
2239                  * locked_ref is the head node, so we have to go one
2240                  * node back for any delayed ref updates
2241                  */
2242                 ref = select_delayed_ref(locked_ref);
2243
2244                 if (ref && ref->seq &&
2245                     btrfs_check_delayed_seq(delayed_refs, ref->seq)) {
2246                         /*
2247                          * there are still refs with lower seq numbers in the
2248                          * process of being added. Don't run this ref yet.
2249                          */
2250                         list_del_init(&locked_ref->cluster);
2251                         mutex_unlock(&locked_ref->mutex);
2252                         locked_ref = NULL;
2253                         delayed_refs->num_heads_ready++;
2254                         spin_unlock(&delayed_refs->lock);
2255                         cond_resched();
2256                         spin_lock(&delayed_refs->lock);
2257                         continue;
2258                 }
2259
2260                 /*
2261                  * record the must insert reserved flag before we
2262                  * drop the spin lock.
2263                  */
2264                 must_insert_reserved = locked_ref->must_insert_reserved;
2265                 locked_ref->must_insert_reserved = 0;
2266
2267                 extent_op = locked_ref->extent_op;
2268                 locked_ref->extent_op = NULL;
2269
2270                 if (!ref) {
2271                         /* All delayed refs have been processed, Go ahead
2272                          * and send the head node to run_one_delayed_ref,
2273                          * so that any accounting fixes can happen
2274                          */
2275                         ref = &locked_ref->node;
2276
2277                         if (extent_op && must_insert_reserved) {
2278                                 kfree(extent_op);
2279                                 extent_op = NULL;
2280                         }
2281
2282                         if (extent_op) {
2283                                 spin_unlock(&delayed_refs->lock);
2284
2285                                 ret = run_delayed_extent_op(trans, root,
2286                                                             ref, extent_op);
2287                                 BUG_ON(ret);
2288                                 kfree(extent_op);
2289
2290                                 goto next;
2291                         }
2292
2293                         list_del_init(&locked_ref->cluster);
2294                         locked_ref = NULL;
2295                 }
2296
2297                 ref->in_tree = 0;
2298                 rb_erase(&ref->rb_node, &delayed_refs->root);
2299                 delayed_refs->num_entries--;
2300                 /*
2301                  * we modified num_entries, but as we're currently running
2302                  * delayed refs, skip
2303                  *     wake_up(&delayed_refs->seq_wait);
2304                  * here.
2305                  */
2306                 spin_unlock(&delayed_refs->lock);
2307
2308                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2309                                           must_insert_reserved);
2310                 BUG_ON(ret);
2311
2312                 btrfs_put_delayed_ref(ref);
2313                 kfree(extent_op);
2314                 count++;
2315 next:
2316                 do_chunk_alloc(trans, root->fs_info->extent_root,
2317                                2 * 1024 * 1024,
2318                                btrfs_get_alloc_profile(root, 0),
2319                                CHUNK_ALLOC_NO_FORCE);
2320                 cond_resched();
2321                 spin_lock(&delayed_refs->lock);
2322         }
2323         return count;
2324 }
2325
2326
2327 static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs,
2328                         unsigned long num_refs)
2329 {
2330         struct list_head *first_seq = delayed_refs->seq_head.next;
2331
2332         spin_unlock(&delayed_refs->lock);
2333         pr_debug("waiting for more refs (num %ld, first %p)\n",
2334                  num_refs, first_seq);
2335         wait_event(delayed_refs->seq_wait,
2336                    num_refs != delayed_refs->num_entries ||
2337                    delayed_refs->seq_head.next != first_seq);
2338         pr_debug("done waiting for more refs (num %ld, first %p)\n",
2339                  delayed_refs->num_entries, delayed_refs->seq_head.next);
2340         spin_lock(&delayed_refs->lock);
2341 }
2342
2343 /*
2344  * this starts processing the delayed reference count updates and
2345  * extent insertions we have queued up so far.  count can be
2346  * 0, which means to process everything in the tree at the start
2347  * of the run (but not newly added entries), or it can be some target
2348  * number you'd like to process.
2349  */
2350 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2351                            struct btrfs_root *root, unsigned long count)
2352 {
2353         struct rb_node *node;
2354         struct btrfs_delayed_ref_root *delayed_refs;
2355         struct btrfs_delayed_ref_node *ref;
2356         struct list_head cluster;
2357         int ret;
2358         u64 delayed_start;
2359         int run_all = count == (unsigned long)-1;
2360         int run_most = 0;
2361         unsigned long num_refs = 0;
2362         int consider_waiting;
2363
2364         if (root == root->fs_info->extent_root)
2365                 root = root->fs_info->tree_root;
2366
2367         do_chunk_alloc(trans, root->fs_info->extent_root,
2368                        2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2369                        CHUNK_ALLOC_NO_FORCE);
2370
2371         delayed_refs = &trans->transaction->delayed_refs;
2372         INIT_LIST_HEAD(&cluster);
2373 again:
2374         consider_waiting = 0;
2375         spin_lock(&delayed_refs->lock);
2376         if (count == 0) {
2377                 count = delayed_refs->num_entries * 2;
2378                 run_most = 1;
2379         }
2380         while (1) {
2381                 if (!(run_all || run_most) &&
2382                     delayed_refs->num_heads_ready < 64)
2383                         break;
2384
2385                 /*
2386                  * go find something we can process in the rbtree.  We start at
2387                  * the beginning of the tree, and then build a cluster
2388                  * of refs to process starting at the first one we are able to
2389                  * lock
2390                  */
2391                 delayed_start = delayed_refs->run_delayed_start;
2392                 ret = btrfs_find_ref_cluster(trans, &cluster,
2393                                              delayed_refs->run_delayed_start);
2394                 if (ret)
2395                         break;
2396
2397                 if (delayed_start >= delayed_refs->run_delayed_start) {
2398                         if (consider_waiting == 0) {
2399                                 /*
2400                                  * btrfs_find_ref_cluster looped. let's do one
2401                                  * more cycle. if we don't run any delayed ref
2402                                  * during that cycle (because we can't because
2403                                  * all of them are blocked) and if the number of
2404                                  * refs doesn't change, we avoid busy waiting.
2405                                  */
2406                                 consider_waiting = 1;
2407                                 num_refs = delayed_refs->num_entries;
2408                         } else {
2409                                 wait_for_more_refs(delayed_refs, num_refs);
2410                                 /*
2411                                  * after waiting, things have changed. we
2412                                  * dropped the lock and someone else might have
2413                                  * run some refs, built new clusters and so on.
2414                                  * therefore, we restart staleness detection.
2415                                  */
2416                                 consider_waiting = 0;
2417                         }
2418                 }
2419
2420                 ret = run_clustered_refs(trans, root, &cluster);
2421                 BUG_ON(ret < 0);
2422
2423                 count -= min_t(unsigned long, ret, count);
2424
2425                 if (count == 0)
2426                         break;
2427
2428                 if (ret || delayed_refs->run_delayed_start == 0) {
2429                         /* refs were run, let's reset staleness detection */
2430                         consider_waiting = 0;
2431                 }
2432         }
2433
2434         if (run_all) {
2435                 node = rb_first(&delayed_refs->root);
2436                 if (!node)
2437                         goto out;
2438                 count = (unsigned long)-1;
2439
2440                 while (node) {
2441                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2442                                        rb_node);
2443                         if (btrfs_delayed_ref_is_head(ref)) {
2444                                 struct btrfs_delayed_ref_head *head;
2445
2446                                 head = btrfs_delayed_node_to_head(ref);
2447                                 atomic_inc(&ref->refs);
2448
2449                                 spin_unlock(&delayed_refs->lock);
2450                                 /*
2451                                  * Mutex was contended, block until it's
2452                                  * released and try again
2453                                  */
2454                                 mutex_lock(&head->mutex);
2455                                 mutex_unlock(&head->mutex);
2456
2457                                 btrfs_put_delayed_ref(ref);
2458                                 cond_resched();
2459                                 goto again;
2460                         }
2461                         node = rb_next(node);
2462                 }
2463                 spin_unlock(&delayed_refs->lock);
2464                 schedule_timeout(1);
2465                 goto again;
2466         }
2467 out:
2468         spin_unlock(&delayed_refs->lock);
2469         return 0;
2470 }
2471
2472 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2473                                 struct btrfs_root *root,
2474                                 u64 bytenr, u64 num_bytes, u64 flags,
2475                                 int is_data)
2476 {
2477         struct btrfs_delayed_extent_op *extent_op;
2478         int ret;
2479
2480         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2481         if (!extent_op)
2482                 return -ENOMEM;
2483
2484         extent_op->flags_to_set = flags;
2485         extent_op->update_flags = 1;
2486         extent_op->update_key = 0;
2487         extent_op->is_data = is_data ? 1 : 0;
2488
2489         ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2490                                           num_bytes, extent_op);
2491         if (ret)
2492                 kfree(extent_op);
2493         return ret;
2494 }
2495
2496 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2497                                       struct btrfs_root *root,
2498                                       struct btrfs_path *path,
2499                                       u64 objectid, u64 offset, u64 bytenr)
2500 {
2501         struct btrfs_delayed_ref_head *head;
2502         struct btrfs_delayed_ref_node *ref;
2503         struct btrfs_delayed_data_ref *data_ref;
2504         struct btrfs_delayed_ref_root *delayed_refs;
2505         struct rb_node *node;
2506         int ret = 0;
2507
2508         ret = -ENOENT;
2509         delayed_refs = &trans->transaction->delayed_refs;
2510         spin_lock(&delayed_refs->lock);
2511         head = btrfs_find_delayed_ref_head(trans, bytenr);
2512         if (!head)
2513                 goto out;
2514
2515         if (!mutex_trylock(&head->mutex)) {
2516                 atomic_inc(&head->node.refs);
2517                 spin_unlock(&delayed_refs->lock);
2518
2519                 btrfs_release_path(path);
2520
2521                 /*
2522                  * Mutex was contended, block until it's released and let
2523                  * caller try again
2524                  */
2525                 mutex_lock(&head->mutex);
2526                 mutex_unlock(&head->mutex);
2527                 btrfs_put_delayed_ref(&head->node);
2528                 return -EAGAIN;
2529         }
2530
2531         node = rb_prev(&head->node.rb_node);
2532         if (!node)
2533                 goto out_unlock;
2534
2535         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2536
2537         if (ref->bytenr != bytenr)
2538                 goto out_unlock;
2539
2540         ret = 1;
2541         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2542                 goto out_unlock;
2543
2544         data_ref = btrfs_delayed_node_to_data_ref(ref);
2545
2546         node = rb_prev(node);
2547         if (node) {
2548                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2549                 if (ref->bytenr == bytenr)
2550                         goto out_unlock;
2551         }
2552
2553         if (data_ref->root != root->root_key.objectid ||
2554             data_ref->objectid != objectid || data_ref->offset != offset)
2555                 goto out_unlock;
2556
2557         ret = 0;
2558 out_unlock:
2559         mutex_unlock(&head->mutex);
2560 out:
2561         spin_unlock(&delayed_refs->lock);
2562         return ret;
2563 }
2564
2565 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2566                                         struct btrfs_root *root,
2567                                         struct btrfs_path *path,
2568                                         u64 objectid, u64 offset, u64 bytenr)
2569 {
2570         struct btrfs_root *extent_root = root->fs_info->extent_root;
2571         struct extent_buffer *leaf;
2572         struct btrfs_extent_data_ref *ref;
2573         struct btrfs_extent_inline_ref *iref;
2574         struct btrfs_extent_item *ei;
2575         struct btrfs_key key;
2576         u32 item_size;
2577         int ret;
2578
2579         key.objectid = bytenr;
2580         key.offset = (u64)-1;
2581         key.type = BTRFS_EXTENT_ITEM_KEY;
2582
2583         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2584         if (ret < 0)
2585                 goto out;
2586         BUG_ON(ret == 0);
2587
2588         ret = -ENOENT;
2589         if (path->slots[0] == 0)
2590                 goto out;
2591
2592         path->slots[0]--;
2593         leaf = path->nodes[0];
2594         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2595
2596         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2597                 goto out;
2598
2599         ret = 1;
2600         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2601 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2602         if (item_size < sizeof(*ei)) {
2603                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2604                 goto out;
2605         }
2606 #endif
2607         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2608
2609         if (item_size != sizeof(*ei) +
2610             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2611                 goto out;
2612
2613         if (btrfs_extent_generation(leaf, ei) <=
2614             btrfs_root_last_snapshot(&root->root_item))
2615                 goto out;
2616
2617         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2618         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2619             BTRFS_EXTENT_DATA_REF_KEY)
2620                 goto out;
2621
2622         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2623         if (btrfs_extent_refs(leaf, ei) !=
2624             btrfs_extent_data_ref_count(leaf, ref) ||
2625             btrfs_extent_data_ref_root(leaf, ref) !=
2626             root->root_key.objectid ||
2627             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2628             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2629                 goto out;
2630
2631         ret = 0;
2632 out:
2633         return ret;
2634 }
2635
2636 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2637                           struct btrfs_root *root,
2638                           u64 objectid, u64 offset, u64 bytenr)
2639 {
2640         struct btrfs_path *path;
2641         int ret;
2642         int ret2;
2643
2644         path = btrfs_alloc_path();
2645         if (!path)
2646                 return -ENOENT;
2647
2648         do {
2649                 ret = check_committed_ref(trans, root, path, objectid,
2650                                           offset, bytenr);
2651                 if (ret && ret != -ENOENT)
2652                         goto out;
2653
2654                 ret2 = check_delayed_ref(trans, root, path, objectid,
2655                                          offset, bytenr);
2656         } while (ret2 == -EAGAIN);
2657
2658         if (ret2 && ret2 != -ENOENT) {
2659                 ret = ret2;
2660                 goto out;
2661         }
2662
2663         if (ret != -ENOENT || ret2 != -ENOENT)
2664                 ret = 0;
2665 out:
2666         btrfs_free_path(path);
2667         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2668                 WARN_ON(ret > 0);
2669         return ret;
2670 }
2671
2672 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2673                            struct btrfs_root *root,
2674                            struct extent_buffer *buf,
2675                            int full_backref, int inc, int for_cow)
2676 {
2677         u64 bytenr;
2678         u64 num_bytes;
2679         u64 parent;
2680         u64 ref_root;
2681         u32 nritems;
2682         struct btrfs_key key;
2683         struct btrfs_file_extent_item *fi;
2684         int i;
2685         int level;
2686         int ret = 0;
2687         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2688                             u64, u64, u64, u64, u64, u64, int);
2689
2690         ref_root = btrfs_header_owner(buf);
2691         nritems = btrfs_header_nritems(buf);
2692         level = btrfs_header_level(buf);
2693
2694         if (!root->ref_cows && level == 0)
2695                 return 0;
2696
2697         if (inc)
2698                 process_func = btrfs_inc_extent_ref;
2699         else
2700                 process_func = btrfs_free_extent;
2701
2702         if (full_backref)
2703                 parent = buf->start;
2704         else
2705                 parent = 0;
2706
2707         for (i = 0; i < nritems; i++) {
2708                 if (level == 0) {
2709                         btrfs_item_key_to_cpu(buf, &key, i);
2710                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2711                                 continue;
2712                         fi = btrfs_item_ptr(buf, i,
2713                                             struct btrfs_file_extent_item);
2714                         if (btrfs_file_extent_type(buf, fi) ==
2715                             BTRFS_FILE_EXTENT_INLINE)
2716                                 continue;
2717                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2718                         if (bytenr == 0)
2719                                 continue;
2720
2721                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2722                         key.offset -= btrfs_file_extent_offset(buf, fi);
2723                         ret = process_func(trans, root, bytenr, num_bytes,
2724                                            parent, ref_root, key.objectid,
2725                                            key.offset, for_cow);
2726                         if (ret)
2727                                 goto fail;
2728                 } else {
2729                         bytenr = btrfs_node_blockptr(buf, i);
2730                         num_bytes = btrfs_level_size(root, level - 1);
2731                         ret = process_func(trans, root, bytenr, num_bytes,
2732                                            parent, ref_root, level - 1, 0,
2733                                            for_cow);
2734                         if (ret)
2735                                 goto fail;
2736                 }
2737         }
2738         return 0;
2739 fail:
2740         BUG();
2741         return ret;
2742 }
2743
2744 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2745                   struct extent_buffer *buf, int full_backref, int for_cow)
2746 {
2747         return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2748 }
2749
2750 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2751                   struct extent_buffer *buf, int full_backref, int for_cow)
2752 {
2753         return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2754 }
2755
2756 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2757                                  struct btrfs_root *root,
2758                                  struct btrfs_path *path,
2759                                  struct btrfs_block_group_cache *cache)
2760 {
2761         int ret;
2762         struct btrfs_root *extent_root = root->fs_info->extent_root;
2763         unsigned long bi;
2764         struct extent_buffer *leaf;
2765
2766         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2767         if (ret < 0)
2768                 goto fail;
2769         BUG_ON(ret);
2770
2771         leaf = path->nodes[0];
2772         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2773         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2774         btrfs_mark_buffer_dirty(leaf);
2775         btrfs_release_path(path);
2776 fail:
2777         if (ret)
2778                 return ret;
2779         return 0;
2780
2781 }
2782
2783 static struct btrfs_block_group_cache *
2784 next_block_group(struct btrfs_root *root,
2785                  struct btrfs_block_group_cache *cache)
2786 {
2787         struct rb_node *node;
2788         spin_lock(&root->fs_info->block_group_cache_lock);
2789         node = rb_next(&cache->cache_node);
2790         btrfs_put_block_group(cache);
2791         if (node) {
2792                 cache = rb_entry(node, struct btrfs_block_group_cache,
2793                                  cache_node);
2794                 btrfs_get_block_group(cache);
2795         } else
2796                 cache = NULL;
2797         spin_unlock(&root->fs_info->block_group_cache_lock);
2798         return cache;
2799 }
2800
2801 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2802                             struct btrfs_trans_handle *trans,
2803                             struct btrfs_path *path)
2804 {
2805         struct btrfs_root *root = block_group->fs_info->tree_root;
2806         struct inode *inode = NULL;
2807         u64 alloc_hint = 0;
2808         int dcs = BTRFS_DC_ERROR;
2809         int num_pages = 0;
2810         int retries = 0;
2811         int ret = 0;
2812
2813         /*
2814          * If this block group is smaller than 100 megs don't bother caching the
2815          * block group.
2816          */
2817         if (block_group->key.offset < (100 * 1024 * 1024)) {
2818                 spin_lock(&block_group->lock);
2819                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2820                 spin_unlock(&block_group->lock);
2821                 return 0;
2822         }
2823
2824 again:
2825         inode = lookup_free_space_inode(root, block_group, path);
2826         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2827                 ret = PTR_ERR(inode);
2828                 btrfs_release_path(path);
2829                 goto out;
2830         }
2831
2832         if (IS_ERR(inode)) {
2833                 BUG_ON(retries);
2834                 retries++;
2835
2836                 if (block_group->ro)
2837                         goto out_free;
2838
2839                 ret = create_free_space_inode(root, trans, block_group, path);
2840                 if (ret)
2841                         goto out_free;
2842                 goto again;
2843         }
2844
2845         /* We've already setup this transaction, go ahead and exit */
2846         if (block_group->cache_generation == trans->transid &&
2847             i_size_read(inode)) {
2848                 dcs = BTRFS_DC_SETUP;
2849                 goto out_put;
2850         }
2851
2852         /*
2853          * We want to set the generation to 0, that way if anything goes wrong
2854          * from here on out we know not to trust this cache when we load up next
2855          * time.
2856          */
2857         BTRFS_I(inode)->generation = 0;
2858         ret = btrfs_update_inode(trans, root, inode);
2859         WARN_ON(ret);
2860
2861         if (i_size_read(inode) > 0) {
2862                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2863                                                       inode);
2864                 if (ret)
2865                         goto out_put;
2866         }
2867
2868         spin_lock(&block_group->lock);
2869         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2870                 /* We're not cached, don't bother trying to write stuff out */
2871                 dcs = BTRFS_DC_WRITTEN;
2872                 spin_unlock(&block_group->lock);
2873                 goto out_put;
2874         }
2875         spin_unlock(&block_group->lock);
2876
2877         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2878         if (!num_pages)
2879                 num_pages = 1;
2880
2881         /*
2882          * Just to make absolutely sure we have enough space, we're going to
2883          * preallocate 12 pages worth of space for each block group.  In
2884          * practice we ought to use at most 8, but we need extra space so we can
2885          * add our header and have a terminator between the extents and the
2886          * bitmaps.
2887          */
2888         num_pages *= 16;
2889         num_pages *= PAGE_CACHE_SIZE;
2890
2891         ret = btrfs_check_data_free_space(inode, num_pages);
2892         if (ret)
2893                 goto out_put;
2894
2895         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2896                                               num_pages, num_pages,
2897                                               &alloc_hint);
2898         if (!ret)
2899                 dcs = BTRFS_DC_SETUP;
2900         btrfs_free_reserved_data_space(inode, num_pages);
2901
2902 out_put:
2903         iput(inode);
2904 out_free:
2905         btrfs_release_path(path);
2906 out:
2907         spin_lock(&block_group->lock);
2908         if (!ret && dcs == BTRFS_DC_SETUP)
2909                 block_group->cache_generation = trans->transid;
2910         block_group->disk_cache_state = dcs;
2911         spin_unlock(&block_group->lock);
2912
2913         return ret;
2914 }
2915
2916 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2917                                    struct btrfs_root *root)
2918 {
2919         struct btrfs_block_group_cache *cache;
2920         int err = 0;
2921         struct btrfs_path *path;
2922         u64 last = 0;
2923
2924         path = btrfs_alloc_path();
2925         if (!path)
2926                 return -ENOMEM;
2927
2928 again:
2929         while (1) {
2930                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2931                 while (cache) {
2932                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2933                                 break;
2934                         cache = next_block_group(root, cache);
2935                 }
2936                 if (!cache) {
2937                         if (last == 0)
2938                                 break;
2939                         last = 0;
2940                         continue;
2941                 }
2942                 err = cache_save_setup(cache, trans, path);
2943                 last = cache->key.objectid + cache->key.offset;
2944                 btrfs_put_block_group(cache);
2945         }
2946
2947         while (1) {
2948                 if (last == 0) {
2949                         err = btrfs_run_delayed_refs(trans, root,
2950                                                      (unsigned long)-1);
2951                         BUG_ON(err);
2952                 }
2953
2954                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2955                 while (cache) {
2956                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2957                                 btrfs_put_block_group(cache);
2958                                 goto again;
2959                         }
2960
2961                         if (cache->dirty)
2962                                 break;
2963                         cache = next_block_group(root, cache);
2964                 }
2965                 if (!cache) {
2966                         if (last == 0)
2967                                 break;
2968                         last = 0;
2969                         continue;
2970                 }
2971
2972                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2973                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2974                 cache->dirty = 0;
2975                 last = cache->key.objectid + cache->key.offset;
2976
2977                 err = write_one_cache_group(trans, root, path, cache);
2978                 BUG_ON(err);
2979                 btrfs_put_block_group(cache);
2980         }
2981
2982         while (1) {
2983                 /*
2984                  * I don't think this is needed since we're just marking our
2985                  * preallocated extent as written, but just in case it can't
2986                  * hurt.
2987                  */
2988                 if (last == 0) {
2989                         err = btrfs_run_delayed_refs(trans, root,
2990                                                      (unsigned long)-1);
2991                         BUG_ON(err);
2992                 }
2993
2994                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2995                 while (cache) {
2996                         /*
2997                          * Really this shouldn't happen, but it could if we
2998                          * couldn't write the entire preallocated extent and
2999                          * splitting the extent resulted in a new block.
3000                          */
3001                         if (cache->dirty) {
3002                                 btrfs_put_block_group(cache);
3003                                 goto again;
3004                         }
3005                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3006                                 break;
3007                         cache = next_block_group(root, cache);
3008                 }
3009                 if (!cache) {
3010                         if (last == 0)
3011                                 break;
3012                         last = 0;
3013                         continue;
3014                 }
3015
3016                 btrfs_write_out_cache(root, trans, cache, path);
3017
3018                 /*
3019                  * If we didn't have an error then the cache state is still
3020                  * NEED_WRITE, so we can set it to WRITTEN.
3021                  */
3022                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3023                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
3024                 last = cache->key.objectid + cache->key.offset;
3025                 btrfs_put_block_group(cache);
3026         }
3027
3028         btrfs_free_path(path);
3029         return 0;
3030 }
3031
3032 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3033 {
3034         struct btrfs_block_group_cache *block_group;
3035         int readonly = 0;
3036
3037         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3038         if (!block_group || block_group->ro)
3039                 readonly = 1;
3040         if (block_group)
3041                 btrfs_put_block_group(block_group);
3042         return readonly;
3043 }
3044
3045 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3046                              u64 total_bytes, u64 bytes_used,
3047                              struct btrfs_space_info **space_info)
3048 {
3049         struct btrfs_space_info *found;
3050         int i;
3051         int factor;
3052
3053         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3054                      BTRFS_BLOCK_GROUP_RAID10))
3055                 factor = 2;
3056         else
3057                 factor = 1;
3058
3059         found = __find_space_info(info, flags);
3060         if (found) {
3061                 spin_lock(&found->lock);
3062                 found->total_bytes += total_bytes;
3063                 found->disk_total += total_bytes * factor;
3064                 found->bytes_used += bytes_used;
3065                 found->disk_used += bytes_used * factor;
3066                 found->full = 0;
3067                 spin_unlock(&found->lock);
3068                 *space_info = found;
3069                 return 0;
3070         }
3071         found = kzalloc(sizeof(*found), GFP_NOFS);
3072         if (!found)
3073                 return -ENOMEM;
3074
3075         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3076                 INIT_LIST_HEAD(&found->block_groups[i]);
3077         init_rwsem(&found->groups_sem);
3078         spin_lock_init(&found->lock);
3079         found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3080         found->total_bytes = total_bytes;
3081         found->disk_total = total_bytes * factor;
3082         found->bytes_used = bytes_used;
3083         found->disk_used = bytes_used * factor;
3084         found->bytes_pinned = 0;
3085         found->bytes_reserved = 0;
3086         found->bytes_readonly = 0;
3087         found->bytes_may_use = 0;
3088         found->full = 0;
3089         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3090         found->chunk_alloc = 0;
3091         found->flush = 0;
3092         init_waitqueue_head(&found->wait);
3093         *space_info = found;
3094         list_add_rcu(&found->list, &info->space_info);
3095         return 0;
3096 }
3097
3098 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3099 {
3100         u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
3101
3102         /* chunk -> extended profile */
3103         if (extra_flags == 0)
3104                 extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3105
3106         if (flags & BTRFS_BLOCK_GROUP_DATA)
3107                 fs_info->avail_data_alloc_bits |= extra_flags;
3108         if (flags & BTRFS_BLOCK_GROUP_METADATA)
3109                 fs_info->avail_metadata_alloc_bits |= extra_flags;
3110         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3111                 fs_info->avail_system_alloc_bits |= extra_flags;
3112 }
3113
3114 /*
3115  * @flags: available profiles in extended format (see ctree.h)
3116  *
3117  * Returns reduced profile in chunk format.  If profile changing is in
3118  * progress (either running or paused) picks the target profile (if it's
3119  * already available), otherwise falls back to plain reducing.
3120  */
3121 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3122 {
3123         /*
3124          * we add in the count of missing devices because we want
3125          * to make sure that any RAID levels on a degraded FS
3126          * continue to be honored.
3127          */
3128         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3129                 root->fs_info->fs_devices->missing_devices;
3130
3131         /* pick restriper's target profile if it's available */
3132         spin_lock(&root->fs_info->balance_lock);
3133         if (root->fs_info->balance_ctl) {
3134                 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3135                 u64 tgt = 0;
3136
3137                 if ((flags & BTRFS_BLOCK_GROUP_DATA) &&
3138                     (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3139                     (flags & bctl->data.target)) {
3140                         tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3141                 } else if ((flags & BTRFS_BLOCK_GROUP_SYSTEM) &&
3142                            (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3143                            (flags & bctl->sys.target)) {
3144                         tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3145                 } else if ((flags & BTRFS_BLOCK_GROUP_METADATA) &&
3146                            (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3147                            (flags & bctl->meta.target)) {
3148                         tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3149                 }
3150
3151                 if (tgt) {
3152                         spin_unlock(&root->fs_info->balance_lock);
3153                         flags = tgt;
3154                         goto out;
3155                 }
3156         }
3157         spin_unlock(&root->fs_info->balance_lock);
3158
3159         if (num_devices == 1)
3160                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3161         if (num_devices < 4)
3162                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3163
3164         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3165             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3166                       BTRFS_BLOCK_GROUP_RAID10))) {
3167                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3168         }
3169
3170         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3171             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3172                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3173         }
3174
3175         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3176             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3177              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3178              (flags & BTRFS_BLOCK_GROUP_DUP))) {
3179                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3180         }
3181
3182 out:
3183         /* extended -> chunk profile */
3184         flags &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3185         return flags;
3186 }
3187
3188 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3189 {
3190         if (flags & BTRFS_BLOCK_GROUP_DATA)
3191                 flags |= root->fs_info->avail_data_alloc_bits;
3192         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3193                 flags |= root->fs_info->avail_system_alloc_bits;
3194         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3195                 flags |= root->fs_info->avail_metadata_alloc_bits;
3196
3197         return btrfs_reduce_alloc_profile(root, flags);
3198 }
3199
3200 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3201 {
3202         u64 flags;
3203
3204         if (data)
3205                 flags = BTRFS_BLOCK_GROUP_DATA;
3206         else if (root == root->fs_info->chunk_root)
3207                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3208         else
3209                 flags = BTRFS_BLOCK_GROUP_METADATA;
3210
3211         return get_alloc_profile(root, flags);
3212 }
3213
3214 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3215 {
3216         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3217                                                        BTRFS_BLOCK_GROUP_DATA);
3218 }
3219
3220 /*
3221  * This will check the space that the inode allocates from to make sure we have
3222  * enough space for bytes.
3223  */
3224 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3225 {
3226         struct btrfs_space_info *data_sinfo;
3227         struct btrfs_root *root = BTRFS_I(inode)->root;
3228         u64 used;
3229         int ret = 0, committed = 0, alloc_chunk = 1;
3230
3231         /* make sure bytes are sectorsize aligned */
3232         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3233
3234         if (root == root->fs_info->tree_root ||
3235             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3236                 alloc_chunk = 0;
3237                 committed = 1;
3238         }
3239
3240         data_sinfo = BTRFS_I(inode)->space_info;
3241         if (!data_sinfo)
3242                 goto alloc;
3243
3244 again:
3245         /* make sure we have enough space to handle the data first */
3246         spin_lock(&data_sinfo->lock);
3247         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3248                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3249                 data_sinfo->bytes_may_use;
3250
3251         if (used + bytes > data_sinfo->total_bytes) {
3252                 struct btrfs_trans_handle *trans;
3253
3254                 /*
3255                  * if we don't have enough free bytes in this space then we need
3256                  * to alloc a new chunk.
3257                  */
3258                 if (!data_sinfo->full && alloc_chunk) {
3259                         u64 alloc_target;
3260
3261                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3262                         spin_unlock(&data_sinfo->lock);
3263 alloc:
3264                         alloc_target = btrfs_get_alloc_profile(root, 1);
3265                         trans = btrfs_join_transaction(root);
3266                         if (IS_ERR(trans))
3267                                 return PTR_ERR(trans);
3268
3269                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3270                                              bytes + 2 * 1024 * 1024,
3271                                              alloc_target,
3272                                              CHUNK_ALLOC_NO_FORCE);
3273                         btrfs_end_transaction(trans, root);
3274                         if (ret < 0) {
3275                                 if (ret != -ENOSPC)
3276                                         return ret;
3277                                 else
3278                                         goto commit_trans;
3279                         }
3280
3281                         if (!data_sinfo) {
3282                                 btrfs_set_inode_space_info(root, inode);
3283                                 data_sinfo = BTRFS_I(inode)->space_info;
3284                         }
3285                         goto again;
3286                 }
3287
3288                 /*
3289                  * If we have less pinned bytes than we want to allocate then
3290                  * don't bother committing the transaction, it won't help us.
3291                  */
3292                 if (data_sinfo->bytes_pinned < bytes)
3293                         committed = 1;
3294                 spin_unlock(&data_sinfo->lock);
3295
3296                 /* commit the current transaction and try again */
3297 commit_trans:
3298                 if (!committed &&
3299                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3300                         committed = 1;
3301                         trans = btrfs_join_transaction(root);
3302                         if (IS_ERR(trans))
3303                                 return PTR_ERR(trans);
3304                         ret = btrfs_commit_transaction(trans, root);
3305                         if (ret)
3306                                 return ret;
3307                         goto again;
3308                 }
3309
3310                 return -ENOSPC;
3311         }
3312         data_sinfo->bytes_may_use += bytes;
3313         trace_btrfs_space_reservation(root->fs_info, "space_info",
3314                                       (u64)data_sinfo, bytes, 1);
3315         spin_unlock(&data_sinfo->lock);
3316
3317         return 0;
3318 }
3319
3320 /*
3321  * Called if we need to clear a data reservation for this inode.
3322  */
3323 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3324 {
3325         struct btrfs_root *root = BTRFS_I(inode)->root;
3326         struct btrfs_space_info *data_sinfo;
3327
3328         /* make sure bytes are sectorsize aligned */
3329         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3330
3331         data_sinfo = BTRFS_I(inode)->space_info;
3332         spin_lock(&data_sinfo->lock);
3333         data_sinfo->bytes_may_use -= bytes;
3334         trace_btrfs_space_reservation(root->fs_info, "space_info",
3335                                       (u64)data_sinfo, bytes, 0);
3336         spin_unlock(&data_sinfo->lock);
3337 }
3338
3339 static void force_metadata_allocation(struct btrfs_fs_info *info)
3340 {
3341         struct list_head *head = &info->space_info;
3342         struct btrfs_space_info *found;
3343
3344         rcu_read_lock();
3345         list_for_each_entry_rcu(found, head, list) {
3346                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3347                         found->force_alloc = CHUNK_ALLOC_FORCE;
3348         }
3349         rcu_read_unlock();
3350 }
3351
3352 static int should_alloc_chunk(struct btrfs_root *root,
3353                               struct btrfs_space_info *sinfo, u64 alloc_bytes,
3354                               int force)
3355 {
3356         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3357         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3358         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3359         u64 thresh;
3360
3361         if (force == CHUNK_ALLOC_FORCE)
3362                 return 1;
3363
3364         /*
3365          * We need to take into account the global rsv because for all intents
3366          * and purposes it's used space.  Don't worry about locking the
3367          * global_rsv, it doesn't change except when the transaction commits.
3368          */
3369         num_allocated += global_rsv->size;
3370
3371         /*
3372          * in limited mode, we want to have some free space up to
3373          * about 1% of the FS size.
3374          */
3375         if (force == CHUNK_ALLOC_LIMITED) {
3376                 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3377                 thresh = max_t(u64, 64 * 1024 * 1024,
3378                                div_factor_fine(thresh, 1));
3379
3380                 if (num_bytes - num_allocated < thresh)
3381                         return 1;
3382         }
3383         thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3384
3385         /* 256MB or 2% of the FS */
3386         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3387         /* system chunks need a much small threshold */
3388         if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM)
3389                 thresh = 32 * 1024 * 1024;
3390
3391         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3392                 return 0;
3393         return 1;
3394 }
3395
3396 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3397                           struct btrfs_root *extent_root, u64 alloc_bytes,
3398                           u64 flags, int force)
3399 {
3400         struct btrfs_space_info *space_info;
3401         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3402         int wait_for_alloc = 0;
3403         int ret = 0;
3404
3405         BUG_ON(!profile_is_valid(flags, 0));
3406
3407         space_info = __find_space_info(extent_root->fs_info, flags);
3408         if (!space_info) {
3409                 ret = update_space_info(extent_root->fs_info, flags,
3410                                         0, 0, &space_info);
3411                 BUG_ON(ret);
3412         }
3413         BUG_ON(!space_info);
3414
3415 again:
3416         spin_lock(&space_info->lock);
3417         if (space_info->force_alloc)
3418                 force = space_info->force_alloc;
3419         if (space_info->full) {
3420                 spin_unlock(&space_info->lock);
3421                 return 0;
3422         }
3423
3424         if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3425                 spin_unlock(&space_info->lock);
3426                 return 0;
3427         } else if (space_info->chunk_alloc) {
3428                 wait_for_alloc = 1;
3429         } else {
3430                 space_info->chunk_alloc = 1;
3431         }
3432
3433         spin_unlock(&space_info->lock);
3434
3435         mutex_lock(&fs_info->chunk_mutex);
3436
3437         /*
3438          * The chunk_mutex is held throughout the entirety of a chunk
3439          * allocation, so once we've acquired the chunk_mutex we know that the
3440          * other guy is done and we need to recheck and see if we should
3441          * allocate.
3442          */
3443         if (wait_for_alloc) {
3444                 mutex_unlock(&fs_info->chunk_mutex);
3445                 wait_for_alloc = 0;
3446                 goto again;
3447         }
3448
3449         /*
3450          * If we have mixed data/metadata chunks we want to make sure we keep
3451          * allocating mixed chunks instead of individual chunks.
3452          */
3453         if (btrfs_mixed_space_info(space_info))
3454                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3455
3456         /*
3457          * if we're doing a data chunk, go ahead and make sure that
3458          * we keep a reasonable number of metadata chunks allocated in the
3459          * FS as well.
3460          */
3461         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3462                 fs_info->data_chunk_allocations++;
3463                 if (!(fs_info->data_chunk_allocations %
3464                       fs_info->metadata_ratio))
3465                         force_metadata_allocation(fs_info);
3466         }
3467
3468         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3469         if (ret < 0 && ret != -ENOSPC)
3470                 goto out;
3471
3472         spin_lock(&space_info->lock);
3473         if (ret)
3474                 space_info->full = 1;
3475         else
3476                 ret = 1;
3477
3478         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3479         space_info->chunk_alloc = 0;
3480         spin_unlock(&space_info->lock);
3481 out:
3482         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3483         return ret;
3484 }
3485
3486 /*
3487  * shrink metadata reservation for delalloc
3488  */
3489 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3490                            bool wait_ordered)
3491 {
3492         struct btrfs_block_rsv *block_rsv;
3493         struct btrfs_space_info *space_info;
3494         struct btrfs_trans_handle *trans;
3495         u64 reserved;
3496         u64 max_reclaim;
3497         u64 reclaimed = 0;
3498         long time_left;
3499         unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3500         int loops = 0;
3501         unsigned long progress;
3502
3503         trans = (struct btrfs_trans_handle *)current->journal_info;
3504         block_rsv = &root->fs_info->delalloc_block_rsv;
3505         space_info = block_rsv->space_info;
3506
3507         smp_mb();
3508         reserved = space_info->bytes_may_use;
3509         progress = space_info->reservation_progress;
3510
3511         if (reserved == 0)
3512                 return 0;
3513
3514         smp_mb();
3515         if (root->fs_info->delalloc_bytes == 0) {
3516                 if (trans)
3517                         return 0;
3518                 btrfs_wait_ordered_extents(root, 0, 0);
3519                 return 0;
3520         }
3521
3522         max_reclaim = min(reserved, to_reclaim);
3523         nr_pages = max_t(unsigned long, nr_pages,
3524                          max_reclaim >> PAGE_CACHE_SHIFT);
3525         while (loops < 1024) {
3526                 /* have the flusher threads jump in and do some IO */
3527                 smp_mb();
3528                 nr_pages = min_t(unsigned long, nr_pages,
3529                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3530                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3531                                                 WB_REASON_FS_FREE_SPACE);
3532
3533                 spin_lock(&space_info->lock);
3534                 if (reserved > space_info->bytes_may_use)
3535                         reclaimed += reserved - space_info->bytes_may_use;
3536                 reserved = space_info->bytes_may_use;
3537                 spin_unlock(&space_info->lock);
3538
3539                 loops++;
3540
3541                 if (reserved == 0 || reclaimed >= max_reclaim)
3542                         break;
3543
3544                 if (trans && trans->transaction->blocked)
3545                         return -EAGAIN;
3546
3547                 if (wait_ordered && !trans) {
3548                         btrfs_wait_ordered_extents(root, 0, 0);
3549                 } else {
3550                         time_left = schedule_timeout_interruptible(1);
3551
3552                         /* We were interrupted, exit */
3553                         if (time_left)
3554                                 break;
3555                 }
3556
3557                 /* we've kicked the IO a few times, if anything has been freed,
3558                  * exit.  There is no sense in looping here for a long time
3559                  * when we really need to commit the transaction, or there are
3560                  * just too many writers without enough free space
3561                  */
3562
3563                 if (loops > 3) {
3564                         smp_mb();
3565                         if (progress != space_info->reservation_progress)
3566                                 break;
3567                 }
3568
3569         }
3570
3571         return reclaimed >= to_reclaim;
3572 }
3573
3574 /**
3575  * maybe_commit_transaction - possibly commit the transaction if its ok to
3576  * @root - the root we're allocating for
3577  * @bytes - the number of bytes we want to reserve
3578  * @force - force the commit
3579  *
3580  * This will check to make sure that committing the transaction will actually
3581  * get us somewhere and then commit the transaction if it does.  Otherwise it
3582  * will return -ENOSPC.
3583  */
3584 static int may_commit_transaction(struct btrfs_root *root,
3585                                   struct btrfs_space_info *space_info,
3586                                   u64 bytes, int force)
3587 {
3588         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3589         struct btrfs_trans_handle *trans;
3590
3591         trans = (struct btrfs_trans_handle *)current->journal_info;
3592         if (trans)
3593                 return -EAGAIN;
3594
3595         if (force)
3596                 goto commit;
3597
3598         /* See if there is enough pinned space to make this reservation */
3599         spin_lock(&space_info->lock);
3600         if (space_info->bytes_pinned >= bytes) {
3601                 spin_unlock(&space_info->lock);
3602                 goto commit;
3603         }
3604         spin_unlock(&space_info->lock);
3605
3606         /*
3607          * See if there is some space in the delayed insertion reservation for
3608          * this reservation.
3609          */
3610         if (space_info != delayed_rsv->space_info)
3611                 return -ENOSPC;
3612
3613         spin_lock(&delayed_rsv->lock);
3614         if (delayed_rsv->size < bytes) {
3615                 spin_unlock(&delayed_rsv->lock);
3616                 return -ENOSPC;
3617         }
3618         spin_unlock(&delayed_rsv->lock);
3619
3620 commit:
3621         trans = btrfs_join_transaction(root);
3622         if (IS_ERR(trans))
3623                 return -ENOSPC;
3624
3625         return btrfs_commit_transaction(trans, root);
3626 }
3627
3628 /**
3629  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3630  * @root - the root we're allocating for
3631  * @block_rsv - the block_rsv we're allocating for
3632  * @orig_bytes - the number of bytes we want
3633  * @flush - wether or not we can flush to make our reservation
3634  *
3635  * This will reserve orgi_bytes number of bytes from the space info associated
3636  * with the block_rsv.  If there is not enough space it will make an attempt to
3637  * flush out space to make room.  It will do this by flushing delalloc if
3638  * possible or committing the transaction.  If flush is 0 then no attempts to
3639  * regain reservations will be made and this will fail if there is not enough
3640  * space already.
3641  */
3642 static int reserve_metadata_bytes(struct btrfs_root *root,
3643                                   struct btrfs_block_rsv *block_rsv,
3644                                   u64 orig_bytes, int flush)
3645 {
3646         struct btrfs_space_info *space_info = block_rsv->space_info;
3647         u64 used;
3648         u64 num_bytes = orig_bytes;
3649         int retries = 0;
3650         int ret = 0;
3651         bool committed = false;
3652         bool flushing = false;
3653         bool wait_ordered = false;
3654
3655 again:
3656         ret = 0;
3657         spin_lock(&space_info->lock);
3658         /*
3659          * We only want to wait if somebody other than us is flushing and we are
3660          * actually alloed to flush.
3661          */
3662         while (flush && !flushing && space_info->flush) {
3663                 spin_unlock(&space_info->lock);
3664                 /*
3665                  * If we have a trans handle we can't wait because the flusher
3666                  * may have to commit the transaction, which would mean we would
3667                  * deadlock since we are waiting for the flusher to finish, but
3668                  * hold the current transaction open.
3669                  */
3670                 if (current->journal_info)
3671                         return -EAGAIN;
3672                 ret = wait_event_interruptible(space_info->wait,
3673                                                !space_info->flush);
3674                 /* Must have been interrupted, return */
3675                 if (ret)
3676                         return -EINTR;
3677
3678                 spin_lock(&space_info->lock);
3679         }
3680
3681         ret = -ENOSPC;
3682         used = space_info->bytes_used + space_info->bytes_reserved +
3683                 space_info->bytes_pinned + space_info->bytes_readonly +
3684                 space_info->bytes_may_use;
3685
3686         /*
3687          * The idea here is that we've not already over-reserved the block group
3688          * then we can go ahead and save our reservation first and then start
3689          * flushing if we need to.  Otherwise if we've already overcommitted
3690          * lets start flushing stuff first and then come back and try to make
3691          * our reservation.
3692          */
3693         if (used <= space_info->total_bytes) {
3694                 if (used + orig_bytes <= space_info->total_bytes) {
3695                         space_info->bytes_may_use += orig_bytes;
3696                         trace_btrfs_space_reservation(root->fs_info,
3697                                                       "space_info",
3698                                                       (u64)space_info,
3699                                                       orig_bytes, 1);
3700                         ret = 0;
3701                 } else {
3702                         /*
3703                          * Ok set num_bytes to orig_bytes since we aren't
3704                          * overocmmitted, this way we only try and reclaim what
3705                          * we need.
3706                          */
3707                         num_bytes = orig_bytes;
3708                 }
3709         } else {
3710                 /*
3711                  * Ok we're over committed, set num_bytes to the overcommitted
3712                  * amount plus the amount of bytes that we need for this
3713                  * reservation.
3714                  */
3715                 wait_ordered = true;
3716                 num_bytes = used - space_info->total_bytes +
3717                         (orig_bytes * (retries + 1));
3718         }
3719
3720         if (ret) {
3721                 u64 profile = btrfs_get_alloc_profile(root, 0);
3722                 u64 avail;
3723
3724                 /*
3725                  * If we have a lot of space that's pinned, don't bother doing
3726                  * the overcommit dance yet and just commit the transaction.
3727                  */
3728                 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3729                 do_div(avail, 10);
3730                 if (space_info->bytes_pinned >= avail && flush && !committed) {
3731                         space_info->flush = 1;
3732                         flushing = true;
3733                         spin_unlock(&space_info->lock);
3734                         ret = may_commit_transaction(root, space_info,
3735                                                      orig_bytes, 1);
3736                         if (ret)
3737                                 goto out;
3738                         committed = true;
3739                         goto again;
3740                 }
3741
3742                 spin_lock(&root->fs_info->free_chunk_lock);
3743                 avail = root->fs_info->free_chunk_space;
3744
3745                 /*
3746                  * If we have dup, raid1 or raid10 then only half of the free
3747                  * space is actually useable.
3748                  */
3749                 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3750                                BTRFS_BLOCK_GROUP_RAID1 |
3751                                BTRFS_BLOCK_GROUP_RAID10))
3752                         avail >>= 1;
3753
3754                 /*
3755                  * If we aren't flushing don't let us overcommit too much, say
3756                  * 1/8th of the space.  If we can flush, let it overcommit up to
3757                  * 1/2 of the space.
3758                  */
3759                 if (flush)
3760                         avail >>= 3;
3761                 else
3762                         avail >>= 1;
3763                  spin_unlock(&root->fs_info->free_chunk_lock);
3764
3765                 if (used + num_bytes < space_info->total_bytes + avail) {
3766                         space_info->bytes_may_use += orig_bytes;
3767                         trace_btrfs_space_reservation(root->fs_info,
3768                                                       "space_info",
3769                                                       (u64)space_info,
3770                                                       orig_bytes, 1);
3771                         ret = 0;
3772                 } else {
3773                         wait_ordered = true;
3774                 }
3775         }
3776
3777         /*
3778          * Couldn't make our reservation, save our place so while we're trying
3779          * to reclaim space we can actually use it instead of somebody else
3780          * stealing it from us.
3781          */
3782         if (ret && flush) {
3783                 flushing = true;
3784                 space_info->flush = 1;
3785         }
3786
3787         spin_unlock(&space_info->lock);
3788
3789         if (!ret || !flush)
3790                 goto out;
3791
3792         /*
3793          * We do synchronous shrinking since we don't actually unreserve
3794          * metadata until after the IO is completed.
3795          */
3796         ret = shrink_delalloc(root, num_bytes, wait_ordered);
3797         if (ret < 0)
3798                 goto out;
3799
3800         ret = 0;
3801
3802         /*
3803          * So if we were overcommitted it's possible that somebody else flushed
3804          * out enough space and we simply didn't have enough space to reclaim,
3805          * so go back around and try again.
3806          */
3807         if (retries < 2) {
3808                 wait_ordered = true;
3809                 retries++;
3810                 goto again;
3811         }
3812
3813         ret = -ENOSPC;
3814         if (committed)
3815                 goto out;
3816
3817         ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3818         if (!ret) {
3819                 committed = true;
3820                 goto again;
3821         }
3822
3823 out:
3824         if (flushing) {
3825                 spin_lock(&space_info->lock);
3826                 space_info->flush = 0;
3827                 wake_up_all(&space_info->wait);
3828                 spin_unlock(&space_info->lock);
3829         }
3830         return ret;
3831 }
3832
3833 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3834                                              struct btrfs_root *root)
3835 {
3836         struct btrfs_block_rsv *block_rsv = NULL;
3837
3838         if (root->ref_cows || root == root->fs_info->csum_root)
3839                 block_rsv = trans->block_rsv;
3840
3841         if (!block_rsv)
3842                 block_rsv = root->block_rsv;
3843
3844         if (!block_rsv)
3845                 block_rsv = &root->fs_info->empty_block_rsv;
3846
3847         return block_rsv;
3848 }
3849
3850 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3851                                u64 num_bytes)
3852 {
3853         int ret = -ENOSPC;
3854         spin_lock(&block_rsv->lock);
3855         if (block_rsv->reserved >= num_bytes) {
3856                 block_rsv->reserved -= num_bytes;
3857                 if (block_rsv->reserved < block_rsv->size)
3858                         block_rsv->full = 0;
3859                 ret = 0;
3860         }
3861         spin_unlock(&block_rsv->lock);
3862         return ret;
3863 }
3864
3865 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3866                                 u64 num_bytes, int update_size)
3867 {
3868         spin_lock(&block_rsv->lock);
3869         block_rsv->reserved += num_bytes;
3870         if (update_size)
3871                 block_rsv->size += num_bytes;
3872         else if (block_rsv->reserved >= block_rsv->size)
3873                 block_rsv->full = 1;
3874         spin_unlock(&block_rsv->lock);
3875 }
3876
3877 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
3878                                     struct btrfs_block_rsv *block_rsv,
3879                                     struct btrfs_block_rsv *dest, u64 num_bytes)
3880 {
3881         struct btrfs_space_info *space_info = block_rsv->space_info;
3882
3883         spin_lock(&block_rsv->lock);
3884         if (num_bytes == (u64)-1)
3885                 num_bytes = block_rsv->size;
3886         block_rsv->size -= num_bytes;
3887         if (block_rsv->reserved >= block_rsv->size) {
3888                 num_bytes = block_rsv->reserved - block_rsv->size;
3889                 block_rsv->reserved = block_rsv->size;
3890                 block_rsv->full = 1;
3891         } else {
3892                 num_bytes = 0;
3893         }
3894         spin_unlock(&block_rsv->lock);
3895
3896         if (num_bytes > 0) {
3897                 if (dest) {
3898                         spin_lock(&dest->lock);
3899                         if (!dest->full) {
3900                                 u64 bytes_to_add;
3901
3902                                 bytes_to_add = dest->size - dest->reserved;
3903                                 bytes_to_add = min(num_bytes, bytes_to_add);
3904                                 dest->reserved += bytes_to_add;
3905                                 if (dest->reserved >= dest->size)
3906                                         dest->full = 1;
3907                                 num_bytes -= bytes_to_add;
3908                         }
3909                         spin_unlock(&dest->lock);
3910                 }
3911                 if (num_bytes) {
3912                         spin_lock(&space_info->lock);
3913                         space_info->bytes_may_use -= num_bytes;
3914                         trace_btrfs_space_reservation(fs_info, "space_info",
3915                                                       (u64)space_info,
3916                                                       num_bytes, 0);
3917                         space_info->reservation_progress++;
3918                         spin_unlock(&space_info->lock);
3919                 }
3920         }
3921 }
3922
3923 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3924                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3925 {
3926         int ret;
3927
3928         ret = block_rsv_use_bytes(src, num_bytes);
3929         if (ret)
3930                 return ret;
3931
3932         block_rsv_add_bytes(dst, num_bytes, 1);
3933         return 0;
3934 }
3935
3936 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3937 {
3938         memset(rsv, 0, sizeof(*rsv));
3939         spin_lock_init(&rsv->lock);
3940 }
3941
3942 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3943 {
3944         struct btrfs_block_rsv *block_rsv;
3945         struct btrfs_fs_info *fs_info = root->fs_info;
3946
3947         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3948         if (!block_rsv)
3949                 return NULL;
3950
3951         btrfs_init_block_rsv(block_rsv);
3952         block_rsv->space_info = __find_space_info(fs_info,
3953                                                   BTRFS_BLOCK_GROUP_METADATA);
3954         return block_rsv;
3955 }
3956
3957 void btrfs_free_block_rsv(struct btrfs_root *root,
3958                           struct btrfs_block_rsv *rsv)
3959 {
3960         btrfs_block_rsv_release(root, rsv, (u64)-1);
3961         kfree(rsv);
3962 }
3963
3964 static inline int __block_rsv_add(struct btrfs_root *root,
3965                                   struct btrfs_block_rsv *block_rsv,
3966                                   u64 num_bytes, int flush)
3967 {
3968         int ret;
3969
3970         if (num_bytes == 0)
3971                 return 0;
3972
3973         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3974         if (!ret) {
3975                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3976                 return 0;
3977         }
3978
3979         return ret;
3980 }
3981
3982 int btrfs_block_rsv_add(struct btrfs_root *root,
3983                         struct btrfs_block_rsv *block_rsv,
3984                         u64 num_bytes)
3985 {
3986         return __block_rsv_add(root, block_rsv, num_bytes, 1);
3987 }
3988
3989 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3990                                 struct btrfs_block_rsv *block_rsv,
3991                                 u64 num_bytes)
3992 {
3993         return __block_rsv_add(root, block_rsv, num_bytes, 0);
3994 }
3995
3996 int btrfs_block_rsv_check(struct btrfs_root *root,
3997                           struct btrfs_block_rsv *block_rsv, int min_factor)
3998 {
3999         u64 num_bytes = 0;
4000         int ret = -ENOSPC;
4001
4002         if (!block_rsv)
4003                 return 0;
4004
4005         spin_lock(&block_rsv->lock);
4006         num_bytes = div_factor(block_rsv->size, min_factor);
4007         if (block_rsv->reserved >= num_bytes)
4008                 ret = 0;
4009         spin_unlock(&block_rsv->lock);
4010
4011         return ret;
4012 }
4013
4014 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4015                                            struct btrfs_block_rsv *block_rsv,
4016                                            u64 min_reserved, int flush)
4017 {
4018         u64 num_bytes = 0;
4019         int ret = -ENOSPC;
4020
4021         if (!block_rsv)
4022                 return 0;
4023
4024         spin_lock(&block_rsv->lock);
4025         num_bytes = min_reserved;
4026         if (block_rsv->reserved >= num_bytes)
4027                 ret = 0;
4028         else
4029                 num_bytes -= block_rsv->reserved;
4030         spin_unlock(&block_rsv->lock);
4031
4032         if (!ret)
4033                 return 0;
4034
4035         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4036         if (!ret) {
4037                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4038                 return 0;
4039         }
4040
4041         return ret;
4042 }
4043
4044 int btrfs_block_rsv_refill(struct btrfs_root *root,
4045                            struct btrfs_block_rsv *block_rsv,
4046                            u64 min_reserved)
4047 {
4048         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4049 }
4050
4051 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4052                                    struct btrfs_block_rsv *block_rsv,
4053                                    u64 min_reserved)
4054 {
4055         return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4056 }
4057
4058 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4059                             struct btrfs_block_rsv *dst_rsv,
4060                             u64 num_bytes)
4061 {
4062         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4063 }
4064
4065 void btrfs_block_rsv_release(struct btrfs_root *root,
4066                              struct btrfs_block_rsv *block_rsv,
4067                              u64 num_bytes)
4068 {
4069         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4070         if (global_rsv->full || global_rsv == block_rsv ||
4071             block_rsv->space_info != global_rsv->space_info)
4072                 global_rsv = NULL;
4073         block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4074                                 num_bytes);
4075 }
4076
4077 /*
4078  * helper to calculate size of global block reservation.
4079  * the desired value is sum of space used by extent tree,
4080  * checksum tree and root tree
4081  */
4082 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4083 {
4084         struct btrfs_space_info *sinfo;
4085         u64 num_bytes;
4086         u64 meta_used;
4087         u64 data_used;
4088         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4089
4090         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4091         spin_lock(&sinfo->lock);
4092         data_used = sinfo->bytes_used;
4093         spin_unlock(&sinfo->lock);
4094
4095         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4096         spin_lock(&sinfo->lock);
4097         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4098                 data_used = 0;
4099         meta_used = sinfo->bytes_used;
4100         spin_unlock(&sinfo->lock);
4101
4102         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4103                     csum_size * 2;
4104         num_bytes += div64_u64(data_used + meta_used, 50);
4105
4106         if (num_bytes * 3 > meta_used)
4107                 num_bytes = div64_u64(meta_used, 3);
4108
4109         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4110 }
4111
4112 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4113 {
4114         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4115         struct btrfs_space_info *sinfo = block_rsv->space_info;
4116         u64 num_bytes;
4117
4118         num_bytes = calc_global_metadata_size(fs_info);
4119
4120         spin_lock(&block_rsv->lock);
4121         spin_lock(&sinfo->lock);
4122
4123         block_rsv->size = num_bytes;
4124
4125         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4126                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4127                     sinfo->bytes_may_use;
4128
4129         if (sinfo->total_bytes > num_bytes) {
4130                 num_bytes = sinfo->total_bytes - num_bytes;
4131                 block_rsv->reserved += num_bytes;
4132                 sinfo->bytes_may_use += num_bytes;
4133                 trace_btrfs_space_reservation(fs_info, "space_info",
4134                                               (u64)sinfo, num_bytes, 1);
4135         }
4136
4137         if (block_rsv->reserved >= block_rsv->size) {
4138                 num_bytes = block_rsv->reserved - block_rsv->size;
4139                 sinfo->bytes_may_use -= num_bytes;
4140                 trace_btrfs_space_reservation(fs_info, "space_info",
4141                                               (u64)sinfo, num_bytes, 0);
4142                 sinfo->reservation_progress++;
4143                 block_rsv->reserved = block_rsv->size;
4144                 block_rsv->full = 1;
4145         }
4146
4147         spin_unlock(&sinfo->lock);
4148         spin_unlock(&block_rsv->lock);
4149 }
4150
4151 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4152 {
4153         struct btrfs_space_info *space_info;
4154
4155         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4156         fs_info->chunk_block_rsv.space_info = space_info;
4157
4158         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4159         fs_info->global_block_rsv.space_info = space_info;
4160         fs_info->delalloc_block_rsv.space_info = space_info;
4161         fs_info->trans_block_rsv.space_info = space_info;
4162         fs_info->empty_block_rsv.space_info = space_info;
4163         fs_info->delayed_block_rsv.space_info = space_info;
4164
4165         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4166         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4167         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4168         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4169         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4170
4171         update_global_block_rsv(fs_info);
4172 }
4173
4174 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4175 {
4176         block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4177                                 (u64)-1);
4178         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4179         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4180         WARN_ON(fs_info->trans_block_rsv.size > 0);
4181         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4182         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4183         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4184         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4185         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4186 }
4187
4188 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4189                                   struct btrfs_root *root)
4190 {
4191         if (!trans->bytes_reserved)
4192                 return;
4193
4194         trace_btrfs_space_reservation(root->fs_info, "transaction", (u64)trans,
4195                                       trans->bytes_reserved, 0);
4196         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4197         trans->bytes_reserved = 0;
4198 }
4199
4200 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4201                                   struct inode *inode)
4202 {
4203         struct btrfs_root *root = BTRFS_I(inode)->root;
4204         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4205         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4206
4207         /*
4208          * We need to hold space in order to delete our orphan item once we've
4209          * added it, so this takes the reservation so we can release it later
4210          * when we are truly done with the orphan item.
4211          */
4212         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4213         trace_btrfs_space_reservation(root->fs_info, "orphan",
4214                                       btrfs_ino(inode), num_bytes, 1);
4215         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4216 }
4217
4218 void btrfs_orphan_release_metadata(struct inode *inode)
4219 {
4220         struct btrfs_root *root = BTRFS_I(inode)->root;
4221         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4222         trace_btrfs_space_reservation(root->fs_info, "orphan",
4223                                       btrfs_ino(inode), num_bytes, 0);
4224         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4225 }
4226
4227 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4228                                 struct btrfs_pending_snapshot *pending)
4229 {
4230         struct btrfs_root *root = pending->root;
4231         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4232         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4233         /*
4234          * two for root back/forward refs, two for directory entries
4235          * and one for root of the snapshot.
4236          */
4237         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4238         dst_rsv->space_info = src_rsv->space_info;
4239         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4240 }
4241
4242 /**
4243  * drop_outstanding_extent - drop an outstanding extent
4244  * @inode: the inode we're dropping the extent for
4245  *
4246  * This is called when we are freeing up an outstanding extent, either called
4247  * after an error or after an extent is written.  This will return the number of
4248  * reserved extents that need to be freed.  This must be called with
4249  * BTRFS_I(inode)->lock held.
4250  */
4251 static unsigned drop_outstanding_extent(struct inode *inode)
4252 {
4253         unsigned drop_inode_space = 0;
4254         unsigned dropped_extents = 0;
4255
4256         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4257         BTRFS_I(inode)->outstanding_extents--;
4258
4259         if (BTRFS_I(inode)->outstanding_extents == 0 &&
4260             BTRFS_I(inode)->delalloc_meta_reserved) {
4261                 drop_inode_space = 1;
4262                 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4263         }
4264
4265         /*
4266          * If we have more or the same amount of outsanding extents than we have
4267          * reserved then we need to leave the reserved extents count alone.
4268          */
4269         if (BTRFS_I(inode)->outstanding_extents >=
4270             BTRFS_I(inode)->reserved_extents)
4271                 return drop_inode_space;
4272
4273         dropped_extents = BTRFS_I(inode)->reserved_extents -
4274                 BTRFS_I(inode)->outstanding_extents;
4275         BTRFS_I(inode)->reserved_extents -= dropped_extents;
4276         return dropped_extents + drop_inode_space;
4277 }
4278
4279 /**
4280  * calc_csum_metadata_size - return the amount of metada space that must be
4281  *      reserved/free'd for the given bytes.
4282  * @inode: the inode we're manipulating
4283  * @num_bytes: the number of bytes in question
4284  * @reserve: 1 if we are reserving space, 0 if we are freeing space
4285  *
4286  * This adjusts the number of csum_bytes in the inode and then returns the
4287  * correct amount of metadata that must either be reserved or freed.  We
4288  * calculate how many checksums we can fit into one leaf and then divide the
4289  * number of bytes that will need to be checksumed by this value to figure out
4290  * how many checksums will be required.  If we are adding bytes then the number
4291  * may go up and we will return the number of additional bytes that must be
4292  * reserved.  If it is going down we will return the number of bytes that must
4293  * be freed.
4294  *
4295  * This must be called with BTRFS_I(inode)->lock held.
4296  */
4297 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4298                                    int reserve)
4299 {
4300         struct btrfs_root *root = BTRFS_I(inode)->root;
4301         u64 csum_size;
4302         int num_csums_per_leaf;
4303         int num_csums;
4304         int old_csums;
4305
4306         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4307             BTRFS_I(inode)->csum_bytes == 0)
4308                 return 0;
4309
4310         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4311         if (reserve)
4312                 BTRFS_I(inode)->csum_bytes += num_bytes;
4313         else
4314                 BTRFS_I(inode)->csum_bytes -= num_bytes;
4315         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4316         num_csums_per_leaf = (int)div64_u64(csum_size,
4317                                             sizeof(struct btrfs_csum_item) +
4318                                             sizeof(struct btrfs_disk_key));
4319         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4320         num_csums = num_csums + num_csums_per_leaf - 1;
4321         num_csums = num_csums / num_csums_per_leaf;
4322
4323         old_csums = old_csums + num_csums_per_leaf - 1;
4324         old_csums = old_csums / num_csums_per_leaf;
4325
4326         /* No change, no need to reserve more */
4327         if (old_csums == num_csums)
4328                 return 0;
4329
4330         if (reserve)
4331                 return btrfs_calc_trans_metadata_size(root,
4332                                                       num_csums - old_csums);
4333
4334         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4335 }
4336
4337 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4338 {
4339         struct btrfs_root *root = BTRFS_I(inode)->root;
4340         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4341         u64 to_reserve = 0;
4342         u64 csum_bytes;
4343         unsigned nr_extents = 0;
4344         int extra_reserve = 0;
4345         int flush = 1;
4346         int ret;
4347
4348         /* Need to be holding the i_mutex here if we aren't free space cache */
4349         if (btrfs_is_free_space_inode(root, inode))
4350                 flush = 0;
4351
4352         if (flush && btrfs_transaction_in_commit(root->fs_info))
4353                 schedule_timeout(1);
4354
4355         mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4356         num_bytes = ALIGN(num_bytes, root->sectorsize);
4357
4358         spin_lock(&BTRFS_I(inode)->lock);
4359         BTRFS_I(inode)->outstanding_extents++;
4360
4361         if (BTRFS_I(inode)->outstanding_extents >
4362             BTRFS_I(inode)->reserved_extents)
4363                 nr_extents = BTRFS_I(inode)->outstanding_extents -
4364                         BTRFS_I(inode)->reserved_extents;
4365
4366         /*
4367          * Add an item to reserve for updating the inode when we complete the
4368          * delalloc io.
4369          */
4370         if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4371                 nr_extents++;
4372                 extra_reserve = 1;
4373         }
4374
4375         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4376         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4377         csum_bytes = BTRFS_I(inode)->csum_bytes;
4378         spin_unlock(&BTRFS_I(inode)->lock);
4379
4380         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4381         if (ret) {
4382                 u64 to_free = 0;
4383                 unsigned dropped;
4384
4385                 spin_lock(&BTRFS_I(inode)->lock);
4386                 dropped = drop_outstanding_extent(inode);
4387                 /*
4388                  * If the inodes csum_bytes is the same as the original
4389                  * csum_bytes then we know we haven't raced with any free()ers
4390                  * so we can just reduce our inodes csum bytes and carry on.
4391                  * Otherwise we have to do the normal free thing to account for
4392                  * the case that the free side didn't free up its reserve
4393                  * because of this outstanding reservation.
4394                  */
4395                 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4396                         calc_csum_metadata_size(inode, num_bytes, 0);
4397                 else
4398                         to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4399                 spin_unlock(&BTRFS_I(inode)->lock);
4400                 if (dropped)
4401                         to_free += btrfs_calc_trans_metadata_size(root, dropped);
4402
4403                 if (to_free) {
4404                         btrfs_block_rsv_release(root, block_rsv, to_free);
4405                         trace_btrfs_space_reservation(root->fs_info,
4406                                                       "delalloc",
4407                                                       btrfs_ino(inode),
4408                                                       to_free, 0);
4409                 }
4410                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4411                 return ret;
4412         }
4413
4414         spin_lock(&BTRFS_I(inode)->lock);
4415         if (extra_reserve) {
4416                 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4417                 nr_extents--;
4418         }
4419         BTRFS_I(inode)->reserved_extents += nr_extents;
4420         spin_unlock(&BTRFS_I(inode)->lock);
4421         mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4422
4423         if (to_reserve)
4424                 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4425                                               btrfs_ino(inode), to_reserve, 1);
4426         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4427
4428         return 0;
4429 }
4430
4431 /**
4432  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4433  * @inode: the inode to release the reservation for
4434  * @num_bytes: the number of bytes we're releasing
4435  *
4436  * This will release the metadata reservation for an inode.  This can be called
4437  * once we complete IO for a given set of bytes to release their metadata
4438  * reservations.
4439  */
4440 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4441 {
4442         struct btrfs_root *root = BTRFS_I(inode)->root;
4443         u64 to_free = 0;
4444         unsigned dropped;
4445
4446         num_bytes = ALIGN(num_bytes, root->sectorsize);
4447         spin_lock(&BTRFS_I(inode)->lock);
4448         dropped = drop_outstanding_extent(inode);
4449
4450         to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4451         spin_unlock(&BTRFS_I(inode)->lock);
4452         if (dropped > 0)
4453                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4454
4455         trace_btrfs_space_reservation(root->fs_info, "delalloc",
4456                                       btrfs_ino(inode), to_free, 0);
4457         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4458                                 to_free);
4459 }
4460
4461 /**
4462  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4463  * @inode: inode we're writing to
4464  * @num_bytes: the number of bytes we want to allocate
4465  *
4466  * This will do the following things
4467  *
4468  * o reserve space in the data space info for num_bytes
4469  * o reserve space in the metadata space info based on number of outstanding
4470  *   extents and how much csums will be needed
4471  * o add to the inodes ->delalloc_bytes
4472  * o add it to the fs_info's delalloc inodes list.
4473  *
4474  * This will return 0 for success and -ENOSPC if there is no space left.
4475  */
4476 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4477 {
4478         int ret;
4479
4480         ret = btrfs_check_data_free_space(inode, num_bytes);
4481         if (ret)
4482                 return ret;
4483
4484         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4485         if (ret) {
4486                 btrfs_free_reserved_data_space(inode, num_bytes);
4487                 return ret;
4488         }
4489
4490         return 0;
4491 }
4492
4493 /**
4494  * btrfs_delalloc_release_space - release data and metadata space for delalloc
4495  * @inode: inode we're releasing space for
4496  * @num_bytes: the number of bytes we want to free up
4497  *
4498  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
4499  * called in the case that we don't need the metadata AND data reservations
4500  * anymore.  So if there is an error or we insert an inline extent.
4501  *
4502  * This function will release the metadata space that was not used and will
4503  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4504  * list if there are no delalloc bytes left.
4505  */
4506 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4507 {
4508         btrfs_delalloc_release_metadata(inode, num_bytes);
4509         btrfs_free_reserved_data_space(inode, num_bytes);
4510 }
4511
4512 static int update_block_group(struct btrfs_trans_handle *trans,
4513                               struct btrfs_root *root,
4514                               u64 bytenr, u64 num_bytes, int alloc)
4515 {
4516         struct btrfs_block_group_cache *cache = NULL;
4517         struct btrfs_fs_info *info = root->fs_info;
4518         u64 total = num_bytes;
4519         u64 old_val;
4520         u64 byte_in_group;
4521         int factor;
4522
4523         /* block accounting for super block */
4524         spin_lock(&info->delalloc_lock);
4525         old_val = btrfs_super_bytes_used(info->super_copy);
4526         if (alloc)
4527                 old_val += num_bytes;
4528         else
4529                 old_val -= num_bytes;
4530         btrfs_set_super_bytes_used(info->super_copy, old_val);
4531         spin_unlock(&info->delalloc_lock);
4532
4533         while (total) {
4534                 cache = btrfs_lookup_block_group(info, bytenr);
4535                 if (!cache)
4536                         return -1;
4537                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4538                                     BTRFS_BLOCK_GROUP_RAID1 |
4539                                     BTRFS_BLOCK_GROUP_RAID10))
4540                         factor = 2;
4541                 else
4542                         factor = 1;
4543                 /*
4544                  * If this block group has free space cache written out, we
4545                  * need to make sure to load it if we are removing space.  This
4546                  * is because we need the unpinning stage to actually add the
4547                  * space back to the block group, otherwise we will leak space.
4548                  */
4549                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4550                         cache_block_group(cache, trans, NULL, 1);
4551
4552                 byte_in_group = bytenr - cache->key.objectid;
4553                 WARN_ON(byte_in_group > cache->key.offset);
4554
4555                 spin_lock(&cache->space_info->lock);
4556                 spin_lock(&cache->lock);
4557
4558                 if (btrfs_test_opt(root, SPACE_CACHE) &&
4559                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4560                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4561
4562                 cache->dirty = 1;
4563                 old_val = btrfs_block_group_used(&cache->item);
4564                 num_bytes = min(total, cache->key.offset - byte_in_group);
4565                 if (alloc) {
4566                         old_val += num_bytes;
4567                         btrfs_set_block_group_used(&cache->item, old_val);
4568                         cache->reserved -= num_bytes;
4569                         cache->space_info->bytes_reserved -= num_bytes;
4570                         cache->space_info->bytes_used += num_bytes;
4571                         cache->space_info->disk_used += num_bytes * factor;
4572                         spin_unlock(&cache->lock);
4573                         spin_unlock(&cache->space_info->lock);
4574                 } else {
4575                         old_val -= num_bytes;
4576                         btrfs_set_block_group_used(&cache->item, old_val);
4577                         cache->pinned += num_bytes;
4578                         cache->space_info->bytes_pinned += num_bytes;
4579                         cache->space_info->bytes_used -= num_bytes;
4580                         cache->space_info->disk_used -= num_bytes * factor;
4581                         spin_unlock(&cache->lock);
4582                         spin_unlock(&cache->space_info->lock);
4583
4584                         set_extent_dirty(info->pinned_extents,
4585                                          bytenr, bytenr + num_bytes - 1,
4586                                          GFP_NOFS | __GFP_NOFAIL);
4587                 }
4588                 btrfs_put_block_group(cache);
4589                 total -= num_bytes;
4590                 bytenr += num_bytes;
4591         }
4592         return 0;
4593 }
4594
4595 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4596 {
4597         struct btrfs_block_group_cache *cache;
4598         u64 bytenr;
4599
4600         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4601         if (!cache)
4602                 return 0;
4603
4604         bytenr = cache->key.objectid;
4605         btrfs_put_block_group(cache);
4606
4607         return bytenr;
4608 }
4609
4610 static int pin_down_extent(struct btrfs_root *root,
4611                            struct btrfs_block_group_cache *cache,
4612                            u64 bytenr, u64 num_bytes, int reserved)
4613 {
4614         spin_lock(&cache->space_info->lock);
4615         spin_lock(&cache->lock);
4616         cache->pinned += num_bytes;
4617         cache->space_info->bytes_pinned += num_bytes;
4618         if (reserved) {
4619                 cache->reserved -= num_bytes;
4620                 cache->space_info->bytes_reserved -= num_bytes;
4621         }
4622         spin_unlock(&cache->lock);
4623         spin_unlock(&cache->space_info->lock);
4624
4625         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4626                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4627         return 0;
4628 }
4629
4630 /*
4631  * this function must be called within transaction
4632  */
4633 int btrfs_pin_extent(struct btrfs_root *root,
4634                      u64 bytenr, u64 num_bytes, int reserved)
4635 {
4636         struct btrfs_block_group_cache *cache;
4637
4638         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4639         BUG_ON(!cache);
4640
4641         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4642
4643         btrfs_put_block_group(cache);
4644         return 0;
4645 }
4646
4647 /*
4648  * this function must be called within transaction
4649  */
4650 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4651                                     struct btrfs_root *root,
4652                                     u64 bytenr, u64 num_bytes)
4653 {
4654         struct btrfs_block_group_cache *cache;
4655
4656         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4657         BUG_ON(!cache);
4658
4659         /*
4660          * pull in the free space cache (if any) so that our pin
4661          * removes the free space from the cache.  We have load_only set
4662          * to one because the slow code to read in the free extents does check
4663          * the pinned extents.
4664          */
4665         cache_block_group(cache, trans, root, 1);
4666
4667         pin_down_extent(root, cache, bytenr, num_bytes, 0);
4668
4669         /* remove us from the free space cache (if we're there at all) */
4670         btrfs_remove_free_space(cache, bytenr, num_bytes);
4671         btrfs_put_block_group(cache);
4672         return 0;
4673 }
4674
4675 /**
4676  * btrfs_update_reserved_bytes - update the block_group and space info counters
4677  * @cache:      The cache we are manipulating
4678  * @num_bytes:  The number of bytes in question
4679  * @reserve:    One of the reservation enums
4680  *
4681  * This is called by the allocator when it reserves space, or by somebody who is
4682  * freeing space that was never actually used on disk.  For example if you
4683  * reserve some space for a new leaf in transaction A and before transaction A
4684  * commits you free that leaf, you call this with reserve set to 0 in order to
4685  * clear the reservation.
4686  *
4687  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4688  * ENOSPC accounting.  For data we handle the reservation through clearing the
4689  * delalloc bits in the io_tree.  We have to do this since we could end up
4690  * allocating less disk space for the amount of data we have reserved in the
4691  * case of compression.
4692  *
4693  * If this is a reservation and the block group has become read only we cannot
4694  * make the reservation and return -EAGAIN, otherwise this function always
4695  * succeeds.
4696  */
4697 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4698                                        u64 num_bytes, int reserve)
4699 {
4700         struct btrfs_space_info *space_info = cache->space_info;
4701         int ret = 0;
4702         spin_lock(&space_info->lock);
4703         spin_lock(&cache->lock);
4704         if (reserve != RESERVE_FREE) {
4705                 if (cache->ro) {
4706                         ret = -EAGAIN;
4707                 } else {
4708                         cache->reserved += num_bytes;
4709                         space_info->bytes_reserved += num_bytes;
4710                         if (reserve == RESERVE_ALLOC) {
4711                                 trace_btrfs_space_reservation(cache->fs_info,
4712                                                               "space_info",
4713                                                               (u64)space_info,
4714                                                               num_bytes, 0);
4715                                 space_info->bytes_may_use -= num_bytes;
4716                         }
4717                 }
4718         } else {
4719                 if (cache->ro)
4720                         space_info->bytes_readonly += num_bytes;
4721                 cache->reserved -= num_bytes;
4722                 space_info->bytes_reserved -= num_bytes;
4723                 space_info->reservation_progress++;
4724         }
4725         spin_unlock(&cache->lock);
4726         spin_unlock(&space_info->lock);
4727         return ret;
4728 }
4729
4730 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4731                                 struct btrfs_root *root)
4732 {
4733         struct btrfs_fs_info *fs_info = root->fs_info;
4734         struct btrfs_caching_control *next;
4735         struct btrfs_caching_control *caching_ctl;
4736         struct btrfs_block_group_cache *cache;
4737
4738         down_write(&fs_info->extent_commit_sem);
4739
4740         list_for_each_entry_safe(caching_ctl, next,
4741                                  &fs_info->caching_block_groups, list) {
4742                 cache = caching_ctl->block_group;
4743                 if (block_group_cache_done(cache)) {
4744                         cache->last_byte_to_unpin = (u64)-1;
4745                         list_del_init(&caching_ctl->list);
4746                         put_caching_control(caching_ctl);
4747                 } else {
4748                         cache->last_byte_to_unpin = caching_ctl->progress;
4749                 }
4750         }
4751
4752         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4753                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4754         else
4755                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4756
4757         up_write(&fs_info->extent_commit_sem);
4758
4759         update_global_block_rsv(fs_info);
4760         return 0;
4761 }
4762
4763 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4764 {
4765         struct btrfs_fs_info *fs_info = root->fs_info;
4766         struct btrfs_block_group_cache *cache = NULL;
4767         u64 len;
4768
4769         while (start <= end) {
4770                 if (!cache ||
4771                     start >= cache->key.objectid + cache->key.offset) {
4772                         if (cache)
4773                                 btrfs_put_block_group(cache);
4774                         cache = btrfs_lookup_block_group(fs_info, start);
4775                         BUG_ON(!cache);
4776                 }
4777
4778                 len = cache->key.objectid + cache->key.offset - start;
4779                 len = min(len, end + 1 - start);
4780
4781                 if (start < cache->last_byte_to_unpin) {
4782                         len = min(len, cache->last_byte_to_unpin - start);
4783                         btrfs_add_free_space(cache, start, len);
4784                 }
4785
4786                 start += len;
4787
4788                 spin_lock(&cache->space_info->lock);
4789                 spin_lock(&cache->lock);
4790                 cache->pinned -= len;
4791                 cache->space_info->bytes_pinned -= len;
4792                 if (cache->ro)
4793                         cache->space_info->bytes_readonly += len;
4794                 spin_unlock(&cache->lock);
4795                 spin_unlock(&cache->space_info->lock);
4796         }
4797
4798         if (cache)
4799                 btrfs_put_block_group(cache);
4800         return 0;
4801 }
4802
4803 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4804                                struct btrfs_root *root)
4805 {
4806         struct btrfs_fs_info *fs_info = root->fs_info;
4807         struct extent_io_tree *unpin;
4808         u64 start;
4809         u64 end;
4810         int ret;
4811
4812         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4813                 unpin = &fs_info->freed_extents[1];
4814         else
4815                 unpin = &fs_info->freed_extents[0];
4816
4817         while (1) {
4818                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4819                                             EXTENT_DIRTY);
4820                 if (ret)
4821                         break;
4822
4823                 if (btrfs_test_opt(root, DISCARD))
4824                         ret = btrfs_discard_extent(root, start,
4825                                                    end + 1 - start, NULL);
4826
4827                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4828                 unpin_extent_range(root, start, end);
4829                 cond_resched();
4830         }
4831
4832         return 0;
4833 }
4834
4835 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4836                                 struct btrfs_root *root,
4837                                 u64 bytenr, u64 num_bytes, u64 parent,
4838                                 u64 root_objectid, u64 owner_objectid,
4839                                 u64 owner_offset, int refs_to_drop,
4840                                 struct btrfs_delayed_extent_op *extent_op)
4841 {
4842         struct btrfs_key key;
4843         struct btrfs_path *path;
4844         struct btrfs_fs_info *info = root->fs_info;
4845         struct btrfs_root *extent_root = info->extent_root;
4846         struct extent_buffer *leaf;
4847         struct btrfs_extent_item *ei;
4848         struct btrfs_extent_inline_ref *iref;
4849         int ret;
4850         int is_data;
4851         int extent_slot = 0;
4852         int found_extent = 0;
4853         int num_to_del = 1;
4854         u32 item_size;
4855         u64 refs;
4856
4857         path = btrfs_alloc_path();
4858         if (!path)
4859                 return -ENOMEM;
4860
4861         path->reada = 1;
4862         path->leave_spinning = 1;
4863
4864         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4865         BUG_ON(!is_data && refs_to_drop != 1);
4866
4867         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4868                                     bytenr, num_bytes, parent,
4869                                     root_objectid, owner_objectid,
4870                                     owner_offset);
4871         if (ret == 0) {
4872                 extent_slot = path->slots[0];
4873                 while (extent_slot >= 0) {
4874                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4875                                               extent_slot);
4876                         if (key.objectid != bytenr)
4877                                 break;
4878                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4879                             key.offset == num_bytes) {
4880                                 found_extent = 1;
4881                                 break;
4882                         }
4883                         if (path->slots[0] - extent_slot > 5)
4884                                 break;
4885                         extent_slot--;
4886                 }
4887 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4888                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4889                 if (found_extent && item_size < sizeof(*ei))
4890                         found_extent = 0;
4891 #endif
4892                 if (!found_extent) {
4893                         BUG_ON(iref);
4894                         ret = remove_extent_backref(trans, extent_root, path,
4895                                                     NULL, refs_to_drop,
4896                                                     is_data);
4897                         BUG_ON(ret);
4898                         btrfs_release_path(path);
4899                         path->leave_spinning = 1;
4900
4901                         key.objectid = bytenr;
4902                         key.type = BTRFS_EXTENT_ITEM_KEY;
4903                         key.offset = num_bytes;
4904
4905                         ret = btrfs_search_slot(trans, extent_root,
4906                                                 &key, path, -1, 1);
4907                         if (ret) {
4908                                 printk(KERN_ERR "umm, got %d back from search"
4909                                        ", was looking for %llu\n", ret,
4910                                        (unsigned long long)bytenr);
4911                                 if (ret > 0)
4912                                         btrfs_print_leaf(extent_root,
4913                                                          path->nodes[0]);
4914                         }
4915                         BUG_ON(ret);
4916                         extent_slot = path->slots[0];
4917                 }
4918         } else {
4919                 btrfs_print_leaf(extent_root, path->nodes[0]);
4920                 WARN_ON(1);
4921                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4922                        "parent %llu root %llu  owner %llu offset %llu\n",
4923                        (unsigned long long)bytenr,
4924                        (unsigned long long)parent,
4925                        (unsigned long long)root_objectid,
4926                        (unsigned long long)owner_objectid,
4927                        (unsigned long long)owner_offset);
4928         }
4929
4930         leaf = path->nodes[0];
4931         item_size = btrfs_item_size_nr(leaf, extent_slot);
4932 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4933         if (item_size < sizeof(*ei)) {
4934                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4935                 ret = convert_extent_item_v0(trans, extent_root, path,
4936                                              owner_objectid, 0);
4937                 BUG_ON(ret < 0);
4938
4939                 btrfs_release_path(path);
4940                 path->leave_spinning = 1;
4941
4942                 key.objectid = bytenr;
4943                 key.type = BTRFS_EXTENT_ITEM_KEY;
4944                 key.offset = num_bytes;
4945
4946                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4947                                         -1, 1);
4948                 if (ret) {
4949                         printk(KERN_ERR "umm, got %d back from search"
4950                                ", was looking for %llu\n", ret,
4951                                (unsigned long long)bytenr);
4952                         btrfs_print_leaf(extent_root, path->nodes[0]);
4953                 }
4954                 BUG_ON(ret);
4955                 extent_slot = path->slots[0];
4956                 leaf = path->nodes[0];
4957                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4958         }
4959 #endif
4960         BUG_ON(item_size < sizeof(*ei));
4961         ei = btrfs_item_ptr(leaf, extent_slot,
4962                             struct btrfs_extent_item);
4963         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4964                 struct btrfs_tree_block_info *bi;
4965                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4966                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4967                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4968         }
4969
4970         refs = btrfs_extent_refs(leaf, ei);
4971         BUG_ON(refs < refs_to_drop);
4972         refs -= refs_to_drop;
4973
4974         if (refs > 0) {
4975                 if (extent_op)
4976                         __run_delayed_extent_op(extent_op, leaf, ei);
4977                 /*
4978                  * In the case of inline back ref, reference count will
4979                  * be updated by remove_extent_backref
4980                  */
4981                 if (iref) {
4982                         BUG_ON(!found_extent);
4983                 } else {
4984                         btrfs_set_extent_refs(leaf, ei, refs);
4985                         btrfs_mark_buffer_dirty(leaf);
4986                 }
4987                 if (found_extent) {
4988                         ret = remove_extent_backref(trans, extent_root, path,
4989                                                     iref, refs_to_drop,
4990                                                     is_data);
4991                         BUG_ON(ret);
4992                 }
4993         } else {
4994                 if (found_extent) {
4995                         BUG_ON(is_data && refs_to_drop !=
4996                                extent_data_ref_count(root, path, iref));
4997                         if (iref) {
4998                                 BUG_ON(path->slots[0] != extent_slot);
4999                         } else {
5000                                 BUG_ON(path->slots[0] != extent_slot + 1);
5001                                 path->slots[0] = extent_slot;
5002                                 num_to_del = 2;
5003                         }
5004                 }
5005
5006                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5007                                       num_to_del);
5008                 BUG_ON(ret);
5009                 btrfs_release_path(path);
5010
5011                 if (is_data) {
5012                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5013                         BUG_ON(ret);
5014                 } else {
5015                         invalidate_mapping_pages(info->btree_inode->i_mapping,
5016                              bytenr >> PAGE_CACHE_SHIFT,
5017                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
5018                 }
5019
5020                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5021                 BUG_ON(ret);
5022         }
5023         btrfs_free_path(path);
5024         return ret;
5025 }
5026
5027 /*
5028  * when we free an block, it is possible (and likely) that we free the last
5029  * delayed ref for that extent as well.  This searches the delayed ref tree for
5030  * a given extent, and if there are no other delayed refs to be processed, it
5031  * removes it from the tree.
5032  */
5033 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5034                                       struct btrfs_root *root, u64 bytenr)
5035 {
5036         struct btrfs_delayed_ref_head *head;
5037         struct btrfs_delayed_ref_root *delayed_refs;
5038         struct btrfs_delayed_ref_node *ref;
5039         struct rb_node *node;
5040         int ret = 0;
5041
5042         delayed_refs = &trans->transaction->delayed_refs;
5043         spin_lock(&delayed_refs->lock);
5044         head = btrfs_find_delayed_ref_head(trans, bytenr);
5045         if (!head)
5046                 goto out;
5047
5048         node = rb_prev(&head->node.rb_node);
5049         if (!node)
5050                 goto out;
5051
5052         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5053
5054         /* there are still entries for this ref, we can't drop it */
5055         if (ref->bytenr == bytenr)
5056                 goto out;
5057
5058         if (head->extent_op) {
5059                 if (!head->must_insert_reserved)
5060                         goto out;
5061                 kfree(head->extent_op);
5062                 head->extent_op = NULL;
5063         }
5064
5065         /*
5066          * waiting for the lock here would deadlock.  If someone else has it
5067          * locked they are already in the process of dropping it anyway
5068          */
5069         if (!mutex_trylock(&head->mutex))
5070                 goto out;
5071
5072         /*
5073          * at this point we have a head with no other entries.  Go
5074          * ahead and process it.
5075          */
5076         head->node.in_tree = 0;
5077         rb_erase(&head->node.rb_node, &delayed_refs->root);
5078
5079         delayed_refs->num_entries--;
5080         if (waitqueue_active(&delayed_refs->seq_wait))
5081                 wake_up(&delayed_refs->seq_wait);
5082
5083         /*
5084          * we don't take a ref on the node because we're removing it from the
5085          * tree, so we just steal the ref the tree was holding.
5086          */
5087         delayed_refs->num_heads--;
5088         if (list_empty(&head->cluster))
5089                 delayed_refs->num_heads_ready--;
5090
5091         list_del_init(&head->cluster);
5092         spin_unlock(&delayed_refs->lock);
5093
5094         BUG_ON(head->extent_op);
5095         if (head->must_insert_reserved)
5096                 ret = 1;
5097
5098         mutex_unlock(&head->mutex);
5099         btrfs_put_delayed_ref(&head->node);
5100         return ret;
5101 out:
5102         spin_unlock(&delayed_refs->lock);
5103         return 0;
5104 }
5105
5106 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5107                            struct btrfs_root *root,
5108                            struct extent_buffer *buf,
5109                            u64 parent, int last_ref, int for_cow)
5110 {
5111         struct btrfs_block_group_cache *cache = NULL;
5112         int ret;
5113
5114         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5115                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5116                                         buf->start, buf->len,
5117                                         parent, root->root_key.objectid,
5118                                         btrfs_header_level(buf),
5119                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5120                 BUG_ON(ret);
5121         }
5122
5123         if (!last_ref)
5124                 return;
5125
5126         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5127
5128         if (btrfs_header_generation(buf) == trans->transid) {
5129                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5130                         ret = check_ref_cleanup(trans, root, buf->start);
5131                         if (!ret)
5132                                 goto out;
5133                 }
5134
5135                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5136                         pin_down_extent(root, cache, buf->start, buf->len, 1);
5137                         goto out;
5138                 }
5139
5140                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5141
5142                 btrfs_add_free_space(cache, buf->start, buf->len);
5143                 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5144         }
5145 out:
5146         /*
5147          * Deleting the buffer, clear the corrupt flag since it doesn't matter
5148          * anymore.
5149          */
5150         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5151         btrfs_put_block_group(cache);
5152 }
5153
5154 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5155                       u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5156                       u64 owner, u64 offset, int for_cow)
5157 {
5158         int ret;
5159         struct btrfs_fs_info *fs_info = root->fs_info;
5160
5161         /*
5162          * tree log blocks never actually go into the extent allocation
5163          * tree, just update pinning info and exit early.
5164          */
5165         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5166                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5167                 /* unlocks the pinned mutex */
5168                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5169                 ret = 0;
5170         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5171                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5172                                         num_bytes,
5173                                         parent, root_objectid, (int)owner,
5174                                         BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5175                 BUG_ON(ret);
5176         } else {
5177                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5178                                                 num_bytes,
5179                                                 parent, root_objectid, owner,
5180                                                 offset, BTRFS_DROP_DELAYED_REF,
5181                                                 NULL, for_cow);
5182                 BUG_ON(ret);
5183         }
5184         return ret;
5185 }
5186
5187 static u64 stripe_align(struct btrfs_root *root, u64 val)
5188 {
5189         u64 mask = ((u64)root->stripesize - 1);
5190         u64 ret = (val + mask) & ~mask;
5191         return ret;
5192 }
5193
5194 /*
5195  * when we wait for progress in the block group caching, its because
5196  * our allocation attempt failed at least once.  So, we must sleep
5197  * and let some progress happen before we try again.
5198  *
5199  * This function will sleep at least once waiting for new free space to
5200  * show up, and then it will check the block group free space numbers
5201  * for our min num_bytes.  Another option is to have it go ahead
5202  * and look in the rbtree for a free extent of a given size, but this
5203  * is a good start.
5204  */
5205 static noinline int
5206 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5207                                 u64 num_bytes)
5208 {
5209         struct btrfs_caching_control *caching_ctl;
5210         DEFINE_WAIT(wait);
5211
5212         caching_ctl = get_caching_control(cache);
5213         if (!caching_ctl)
5214                 return 0;
5215
5216         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5217                    (cache->free_space_ctl->free_space >= num_bytes));
5218
5219         put_caching_control(caching_ctl);
5220         return 0;
5221 }
5222
5223 static noinline int
5224 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5225 {
5226         struct btrfs_caching_control *caching_ctl;
5227         DEFINE_WAIT(wait);
5228
5229         caching_ctl = get_caching_control(cache);
5230         if (!caching_ctl)
5231                 return 0;
5232
5233         wait_event(caching_ctl->wait, block_group_cache_done(cache));
5234
5235         put_caching_control(caching_ctl);
5236         return 0;
5237 }
5238
5239 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5240 {
5241         int index;
5242         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5243                 index = 0;
5244         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5245                 index = 1;
5246         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5247                 index = 2;
5248         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5249                 index = 3;
5250         else
5251                 index = 4;
5252         return index;
5253 }
5254
5255 enum btrfs_loop_type {
5256         LOOP_FIND_IDEAL = 0,
5257         LOOP_CACHING_NOWAIT = 1,
5258         LOOP_CACHING_WAIT = 2,
5259         LOOP_ALLOC_CHUNK = 3,
5260         LOOP_NO_EMPTY_SIZE = 4,
5261 };
5262
5263 /*
5264  * walks the btree of allocated extents and find a hole of a given size.
5265  * The key ins is changed to record the hole:
5266  * ins->objectid == block start
5267  * ins->flags = BTRFS_EXTENT_ITEM_KEY
5268  * ins->offset == number of blocks
5269  * Any available blocks before search_start are skipped.
5270  */
5271 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5272                                      struct btrfs_root *orig_root,
5273                                      u64 num_bytes, u64 empty_size,
5274                                      u64 search_start, u64 search_end,
5275                                      u64 hint_byte, struct btrfs_key *ins,
5276                                      u64 data)
5277 {
5278         int ret = 0;
5279         struct btrfs_root *root = orig_root->fs_info->extent_root;
5280         struct btrfs_free_cluster *last_ptr = NULL;
5281         struct btrfs_block_group_cache *block_group = NULL;
5282         struct btrfs_block_group_cache *used_block_group;
5283         int empty_cluster = 2 * 1024 * 1024;
5284         int allowed_chunk_alloc = 0;
5285         int done_chunk_alloc = 0;
5286         struct btrfs_space_info *space_info;
5287         int loop = 0;
5288         int index = 0;
5289         int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5290                 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5291         bool found_uncached_bg = false;
5292         bool failed_cluster_refill = false;
5293         bool failed_alloc = false;
5294         bool use_cluster = true;
5295         bool have_caching_bg = false;
5296         u64 ideal_cache_percent = 0;
5297         u64 ideal_cache_offset = 0;
5298
5299         WARN_ON(num_bytes < root->sectorsize);
5300         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5301         ins->objectid = 0;
5302         ins->offset = 0;
5303
5304         trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5305
5306         space_info = __find_space_info(root->fs_info, data);
5307         if (!space_info) {
5308                 printk(KERN_ERR "No space info for %llu\n", data);
5309                 return -ENOSPC;
5310         }
5311
5312         /*
5313          * If the space info is for both data and metadata it means we have a
5314          * small filesystem and we can't use the clustering stuff.
5315          */
5316         if (btrfs_mixed_space_info(space_info))
5317                 use_cluster = false;
5318
5319         if (orig_root->ref_cows || empty_size)
5320                 allowed_chunk_alloc = 1;
5321
5322         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5323                 last_ptr = &root->fs_info->meta_alloc_cluster;
5324                 if (!btrfs_test_opt(root, SSD))
5325                         empty_cluster = 64 * 1024;
5326         }
5327
5328         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5329             btrfs_test_opt(root, SSD)) {
5330                 last_ptr = &root->fs_info->data_alloc_cluster;
5331         }
5332
5333         if (last_ptr) {
5334                 spin_lock(&last_ptr->lock);
5335                 if (last_ptr->block_group)
5336                         hint_byte = last_ptr->window_start;
5337                 spin_unlock(&last_ptr->lock);
5338         }
5339
5340         search_start = max(search_start, first_logical_byte(root, 0));
5341         search_start = max(search_start, hint_byte);
5342
5343         if (!last_ptr)
5344                 empty_cluster = 0;
5345
5346         if (search_start == hint_byte) {
5347 ideal_cache:
5348                 block_group = btrfs_lookup_block_group(root->fs_info,
5349                                                        search_start);
5350                 used_block_group = block_group;
5351                 /*
5352                  * we don't want to use the block group if it doesn't match our
5353                  * allocation bits, or if its not cached.
5354                  *
5355                  * However if we are re-searching with an ideal block group
5356                  * picked out then we don't care that the block group is cached.
5357                  */
5358                 if (block_group && block_group_bits(block_group, data) &&
5359                     (block_group->cached != BTRFS_CACHE_NO ||
5360                      search_start == ideal_cache_offset)) {
5361                         down_read(&space_info->groups_sem);
5362                         if (list_empty(&block_group->list) ||
5363                             block_group->ro) {
5364                                 /*
5365                                  * someone is removing this block group,
5366                                  * we can't jump into the have_block_group
5367                                  * target because our list pointers are not
5368                                  * valid
5369                                  */
5370                                 btrfs_put_block_group(block_group);
5371                                 up_read(&space_info->groups_sem);
5372                         } else {
5373                                 index = get_block_group_index(block_group);
5374                                 goto have_block_group;
5375                         }
5376                 } else if (block_group) {
5377                         btrfs_put_block_group(block_group);
5378                 }
5379         }
5380 search:
5381         have_caching_bg = false;
5382         down_read(&space_info->groups_sem);
5383         list_for_each_entry(block_group, &space_info->block_groups[index],
5384                             list) {
5385                 u64 offset;
5386                 int cached;
5387
5388                 used_block_group = block_group;
5389                 btrfs_get_block_group(block_group);
5390                 search_start = block_group->key.objectid;
5391
5392                 /*
5393                  * this can happen if we end up cycling through all the
5394                  * raid types, but we want to make sure we only allocate
5395                  * for the proper type.
5396                  */
5397                 if (!block_group_bits(block_group, data)) {
5398                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
5399                                 BTRFS_BLOCK_GROUP_RAID1 |
5400                                 BTRFS_BLOCK_GROUP_RAID10;
5401
5402                         /*
5403                          * if they asked for extra copies and this block group
5404                          * doesn't provide them, bail.  This does allow us to
5405                          * fill raid0 from raid1.
5406                          */
5407                         if ((data & extra) && !(block_group->flags & extra))
5408                                 goto loop;
5409                 }
5410
5411 have_block_group:
5412                 cached = block_group_cache_done(block_group);
5413                 if (unlikely(!cached)) {
5414                         u64 free_percent;
5415
5416                         found_uncached_bg = true;
5417                         ret = cache_block_group(block_group, trans,
5418                                                 orig_root, 1);
5419                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5420                                 goto alloc;
5421
5422                         free_percent = btrfs_block_group_used(&block_group->item);
5423                         free_percent *= 100;
5424                         free_percent = div64_u64(free_percent,
5425                                                  block_group->key.offset);
5426                         free_percent = 100 - free_percent;
5427                         if (free_percent > ideal_cache_percent &&
5428                             likely(!block_group->ro)) {
5429                                 ideal_cache_offset = block_group->key.objectid;
5430                                 ideal_cache_percent = free_percent;
5431                         }
5432
5433                         /*
5434                          * The caching workers are limited to 2 threads, so we
5435                          * can queue as much work as we care to.
5436                          */
5437                         if (loop > LOOP_FIND_IDEAL) {
5438                                 ret = cache_block_group(block_group, trans,
5439                                                         orig_root, 0);
5440                                 BUG_ON(ret);
5441                         }
5442
5443                         /*
5444                          * If loop is set for cached only, try the next block
5445                          * group.
5446                          */
5447                         if (loop == LOOP_FIND_IDEAL)
5448                                 goto loop;
5449                 }
5450
5451 alloc:
5452                 if (unlikely(block_group->ro))
5453                         goto loop;
5454
5455                 /*
5456                  * Ok we want to try and use the cluster allocator, so
5457                  * lets look there
5458                  */
5459                 if (last_ptr) {
5460                         /*
5461                          * the refill lock keeps out other
5462                          * people trying to start a new cluster
5463                          */
5464                         spin_lock(&last_ptr->refill_lock);
5465                         used_block_group = last_ptr->block_group;
5466                         if (used_block_group != block_group &&
5467                             (!used_block_group ||
5468                              used_block_group->ro ||
5469                              !block_group_bits(used_block_group, data))) {
5470                                 used_block_group = block_group;
5471                                 goto refill_cluster;
5472                         }
5473
5474                         if (used_block_group != block_group)
5475                                 btrfs_get_block_group(used_block_group);
5476
5477                         offset = btrfs_alloc_from_cluster(used_block_group,
5478                           last_ptr, num_bytes, used_block_group->key.objectid);
5479                         if (offset) {
5480                                 /* we have a block, we're done */
5481                                 spin_unlock(&last_ptr->refill_lock);
5482                                 trace_btrfs_reserve_extent_cluster(root,
5483                                         block_group, search_start, num_bytes);
5484                                 goto checks;
5485                         }
5486
5487                         WARN_ON(last_ptr->block_group != used_block_group);
5488                         if (used_block_group != block_group) {
5489                                 btrfs_put_block_group(used_block_group);
5490                                 used_block_group = block_group;
5491                         }
5492 refill_cluster:
5493                         BUG_ON(used_block_group != block_group);
5494                         /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5495                          * set up a new clusters, so lets just skip it
5496                          * and let the allocator find whatever block
5497                          * it can find.  If we reach this point, we
5498                          * will have tried the cluster allocator
5499                          * plenty of times and not have found
5500                          * anything, so we are likely way too
5501                          * fragmented for the clustering stuff to find
5502                          * anything.
5503                          *
5504                          * However, if the cluster is taken from the
5505                          * current block group, release the cluster
5506                          * first, so that we stand a better chance of
5507                          * succeeding in the unclustered
5508                          * allocation.  */
5509                         if (loop >= LOOP_NO_EMPTY_SIZE &&
5510                             last_ptr->block_group != block_group) {
5511                                 spin_unlock(&last_ptr->refill_lock);
5512                                 goto unclustered_alloc;
5513                         }
5514
5515                         /*
5516                          * this cluster didn't work out, free it and
5517                          * start over
5518                          */
5519                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5520
5521                         if (loop >= LOOP_NO_EMPTY_SIZE) {
5522                                 spin_unlock(&last_ptr->refill_lock);
5523                                 goto unclustered_alloc;
5524                         }
5525
5526                         /* allocate a cluster in this block group */
5527                         ret = btrfs_find_space_cluster(trans, root,
5528                                                block_group, last_ptr,
5529                                                search_start, num_bytes,
5530                                                empty_cluster + empty_size);
5531                         if (ret == 0) {
5532                                 /*
5533                                  * now pull our allocation out of this
5534                                  * cluster
5535                                  */
5536                                 offset = btrfs_alloc_from_cluster(block_group,
5537                                                   last_ptr, num_bytes,
5538                                                   search_start);
5539                                 if (offset) {
5540                                         /* we found one, proceed */
5541                                         spin_unlock(&last_ptr->refill_lock);
5542                                         trace_btrfs_reserve_extent_cluster(root,
5543                                                 block_group, search_start,
5544                                                 num_bytes);
5545                                         goto checks;
5546                                 }
5547                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5548                                    && !failed_cluster_refill) {
5549                                 spin_unlock(&last_ptr->refill_lock);
5550
5551                                 failed_cluster_refill = true;
5552                                 wait_block_group_cache_progress(block_group,
5553                                        num_bytes + empty_cluster + empty_size);
5554                                 goto have_block_group;
5555                         }
5556
5557                         /*
5558                          * at this point we either didn't find a cluster
5559                          * or we weren't able to allocate a block from our
5560                          * cluster.  Free the cluster we've been trying
5561                          * to use, and go to the next block group
5562                          */
5563                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5564                         spin_unlock(&last_ptr->refill_lock);
5565                         goto loop;
5566                 }
5567
5568 unclustered_alloc:
5569                 spin_lock(&block_group->free_space_ctl->tree_lock);
5570                 if (cached &&
5571                     block_group->free_space_ctl->free_space <
5572                     num_bytes + empty_cluster + empty_size) {
5573                         spin_unlock(&block_group->free_space_ctl->tree_lock);
5574                         goto loop;
5575                 }
5576                 spin_unlock(&block_group->free_space_ctl->tree_lock);
5577
5578                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5579                                                     num_bytes, empty_size);
5580                 /*
5581                  * If we didn't find a chunk, and we haven't failed on this
5582                  * block group before, and this block group is in the middle of
5583                  * caching and we are ok with waiting, then go ahead and wait
5584                  * for progress to be made, and set failed_alloc to true.
5585                  *
5586                  * If failed_alloc is true then we've already waited on this
5587                  * block group once and should move on to the next block group.
5588                  */
5589                 if (!offset && !failed_alloc && !cached &&
5590                     loop > LOOP_CACHING_NOWAIT) {
5591                         wait_block_group_cache_progress(block_group,
5592                                                 num_bytes + empty_size);
5593                         failed_alloc = true;
5594                         goto have_block_group;
5595                 } else if (!offset) {
5596                         if (!cached)
5597                                 have_caching_bg = true;
5598                         goto loop;
5599                 }
5600 checks:
5601                 search_start = stripe_align(root, offset);
5602                 /* move on to the next group */
5603                 if (search_start + num_bytes >= search_end) {
5604                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5605                         goto loop;
5606                 }
5607
5608                 /* move on to the next group */
5609                 if (search_start + num_bytes >
5610                     used_block_group->key.objectid + used_block_group->key.offset) {
5611                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5612                         goto loop;
5613                 }
5614
5615                 if (offset < search_start)
5616                         btrfs_add_free_space(used_block_group, offset,
5617                                              search_start - offset);
5618                 BUG_ON(offset > search_start);
5619
5620                 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5621                                                   alloc_type);
5622                 if (ret == -EAGAIN) {
5623                         btrfs_add_free_space(used_block_group, offset, num_bytes);
5624                         goto loop;
5625                 }
5626
5627                 /* we are all good, lets return */
5628                 ins->objectid = search_start;
5629                 ins->offset = num_bytes;
5630
5631                 trace_btrfs_reserve_extent(orig_root, block_group,
5632                                            search_start, num_bytes);
5633                 if (offset < search_start)
5634                         btrfs_add_free_space(used_block_group, offset,
5635                                              search_start - offset);
5636                 BUG_ON(offset > search_start);
5637                 if (used_block_group != block_group)
5638                         btrfs_put_block_group(used_block_group);
5639                 btrfs_put_block_group(block_group);
5640                 break;
5641 loop:
5642                 failed_cluster_refill = false;
5643                 failed_alloc = false;
5644                 BUG_ON(index != get_block_group_index(block_group));
5645                 if (used_block_group != block_group)
5646                         btrfs_put_block_group(used_block_group);
5647                 btrfs_put_block_group(block_group);
5648         }
5649         up_read(&space_info->groups_sem);
5650
5651         if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5652                 goto search;
5653
5654         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5655                 goto search;
5656
5657         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5658          *                      for them to make caching progress.  Also
5659          *                      determine the best possible bg to cache
5660          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5661          *                      caching kthreads as we move along
5662          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5663          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5664          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5665          *                      again
5666          */
5667         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5668                 index = 0;
5669                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5670                         found_uncached_bg = false;
5671                         loop++;
5672                         if (!ideal_cache_percent)
5673                                 goto search;
5674
5675                         /*
5676                          * 1 of the following 2 things have happened so far
5677                          *
5678                          * 1) We found an ideal block group for caching that
5679                          * is mostly full and will cache quickly, so we might
5680                          * as well wait for it.
5681                          *
5682                          * 2) We searched for cached only and we didn't find
5683                          * anything, and we didn't start any caching kthreads
5684                          * either, so chances are we will loop through and
5685                          * start a couple caching kthreads, and then come back
5686                          * around and just wait for them.  This will be slower
5687                          * because we will have 2 caching kthreads reading at
5688                          * the same time when we could have just started one
5689                          * and waited for it to get far enough to give us an
5690                          * allocation, so go ahead and go to the wait caching
5691                          * loop.
5692                          */
5693                         loop = LOOP_CACHING_WAIT;
5694                         search_start = ideal_cache_offset;
5695                         ideal_cache_percent = 0;
5696                         goto ideal_cache;
5697                 } else if (loop == LOOP_FIND_IDEAL) {
5698                         /*
5699                          * Didn't find a uncached bg, wait on anything we find
5700                          * next.
5701                          */
5702                         loop = LOOP_CACHING_WAIT;
5703                         goto search;
5704                 }
5705
5706                 loop++;
5707
5708                 if (loop == LOOP_ALLOC_CHUNK) {
5709                        if (allowed_chunk_alloc) {
5710                                 ret = do_chunk_alloc(trans, root, num_bytes +
5711                                                      2 * 1024 * 1024, data,
5712                                                      CHUNK_ALLOC_LIMITED);
5713                                 allowed_chunk_alloc = 0;
5714                                 if (ret == 1)
5715                                         done_chunk_alloc = 1;
5716                         } else if (!done_chunk_alloc &&
5717                                    space_info->force_alloc ==
5718                                    CHUNK_ALLOC_NO_FORCE) {
5719                                 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5720                         }
5721
5722                        /*
5723                         * We didn't allocate a chunk, go ahead and drop the
5724                         * empty size and loop again.
5725                         */
5726                        if (!done_chunk_alloc)
5727                                loop = LOOP_NO_EMPTY_SIZE;
5728                 }
5729
5730                 if (loop == LOOP_NO_EMPTY_SIZE) {
5731                         empty_size = 0;
5732                         empty_cluster = 0;
5733                 }
5734
5735                 goto search;
5736         } else if (!ins->objectid) {
5737                 ret = -ENOSPC;
5738         } else if (ins->objectid) {
5739                 ret = 0;
5740         }
5741
5742         return ret;
5743 }
5744
5745 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5746                             int dump_block_groups)
5747 {
5748         struct btrfs_block_group_cache *cache;
5749         int index = 0;
5750
5751         spin_lock(&info->lock);
5752         printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5753                (unsigned long long)info->flags,
5754                (unsigned long long)(info->total_bytes - info->bytes_used -
5755                                     info->bytes_pinned - info->bytes_reserved -
5756                                     info->bytes_readonly),
5757                (info->full) ? "" : "not ");
5758         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5759                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5760                (unsigned long long)info->total_bytes,
5761                (unsigned long long)info->bytes_used,
5762                (unsigned long long)info->bytes_pinned,
5763                (unsigned long long)info->bytes_reserved,
5764                (unsigned long long)info->bytes_may_use,
5765                (unsigned long long)info->bytes_readonly);
5766         spin_unlock(&info->lock);
5767
5768         if (!dump_block_groups)
5769                 return;
5770
5771         down_read(&info->groups_sem);
5772 again:
5773         list_for_each_entry(cache, &info->block_groups[index], list) {
5774                 spin_lock(&cache->lock);
5775                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5776                        "%llu pinned %llu reserved\n",
5777                        (unsigned long long)cache->key.objectid,
5778                        (unsigned long long)cache->key.offset,
5779                        (unsigned long long)btrfs_block_group_used(&cache->item),
5780                        (unsigned long long)cache->pinned,
5781                        (unsigned long long)cache->reserved);
5782                 btrfs_dump_free_space(cache, bytes);
5783                 spin_unlock(&cache->lock);
5784         }
5785         if (++index < BTRFS_NR_RAID_TYPES)
5786                 goto again;
5787         up_read(&info->groups_sem);
5788 }
5789
5790 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5791                          struct btrfs_root *root,
5792                          u64 num_bytes, u64 min_alloc_size,
5793                          u64 empty_size, u64 hint_byte,
5794                          u64 search_end, struct btrfs_key *ins,
5795                          u64 data)
5796 {
5797         int ret;
5798         u64 search_start = 0;
5799
5800         data = btrfs_get_alloc_profile(root, data);
5801 again:
5802         /*
5803          * the only place that sets empty_size is btrfs_realloc_node, which
5804          * is not called recursively on allocations
5805          */
5806         if (empty_size || root->ref_cows)
5807                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5808                                      num_bytes + 2 * 1024 * 1024, data,
5809                                      CHUNK_ALLOC_NO_FORCE);
5810
5811         WARN_ON(num_bytes < root->sectorsize);
5812         ret = find_free_extent(trans, root, num_bytes, empty_size,
5813                                search_start, search_end, hint_byte,
5814                                ins, data);
5815
5816         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5817                 num_bytes = num_bytes >> 1;
5818                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5819                 num_bytes = max(num_bytes, min_alloc_size);
5820                 do_chunk_alloc(trans, root->fs_info->extent_root,
5821                                num_bytes, data, CHUNK_ALLOC_FORCE);
5822                 goto again;
5823         }
5824         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5825                 struct btrfs_space_info *sinfo;
5826
5827                 sinfo = __find_space_info(root->fs_info, data);
5828                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5829                        "wanted %llu\n", (unsigned long long)data,
5830                        (unsigned long long)num_bytes);
5831                 dump_space_info(sinfo, num_bytes, 1);
5832         }
5833
5834         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5835
5836         return ret;
5837 }
5838
5839 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5840                                         u64 start, u64 len, int pin)
5841 {
5842         struct btrfs_block_group_cache *cache;
5843         int ret = 0;
5844
5845         cache = btrfs_lookup_block_group(root->fs_info, start);
5846         if (!cache) {
5847                 printk(KERN_ERR "Unable to find block group for %llu\n",
5848                        (unsigned long long)start);
5849                 return -ENOSPC;
5850         }
5851
5852         if (btrfs_test_opt(root, DISCARD))
5853                 ret = btrfs_discard_extent(root, start, len, NULL);
5854
5855         if (pin)
5856                 pin_down_extent(root, cache, start, len, 1);
5857         else {
5858                 btrfs_add_free_space(cache, start, len);
5859                 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5860         }
5861         btrfs_put_block_group(cache);
5862
5863         trace_btrfs_reserved_extent_free(root, start, len);
5864
5865         return ret;
5866 }
5867
5868 int btrfs_free_reserved_extent(struct btrfs_root *root,
5869                                         u64 start, u64 len)
5870 {
5871         return __btrfs_free_reserved_extent(root, start, len, 0);
5872 }
5873
5874 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5875                                        u64 start, u64 len)
5876 {
5877         return __btrfs_free_reserved_extent(root, start, len, 1);
5878 }
5879
5880 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5881                                       struct btrfs_root *root,
5882                                       u64 parent, u64 root_objectid,
5883                                       u64 flags, u64 owner, u64 offset,
5884                                       struct btrfs_key *ins, int ref_mod)
5885 {
5886         int ret;
5887         struct btrfs_fs_info *fs_info = root->fs_info;
5888         struct btrfs_extent_item *extent_item;
5889         struct btrfs_extent_inline_ref *iref;
5890         struct btrfs_path *path;
5891         struct extent_buffer *leaf;
5892         int type;
5893         u32 size;
5894
5895         if (parent > 0)
5896                 type = BTRFS_SHARED_DATA_REF_KEY;
5897         else
5898                 type = BTRFS_EXTENT_DATA_REF_KEY;
5899
5900         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5901
5902         path = btrfs_alloc_path();
5903         if (!path)
5904                 return -ENOMEM;
5905
5906         path->leave_spinning = 1;
5907         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5908                                       ins, size);
5909         BUG_ON(ret);
5910
5911         leaf = path->nodes[0];
5912         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5913                                      struct btrfs_extent_item);
5914         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5915         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5916         btrfs_set_extent_flags(leaf, extent_item,
5917                                flags | BTRFS_EXTENT_FLAG_DATA);
5918
5919         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5920         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5921         if (parent > 0) {
5922                 struct btrfs_shared_data_ref *ref;
5923                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5924                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5925                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5926         } else {
5927                 struct btrfs_extent_data_ref *ref;
5928                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5929                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5930                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5931                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5932                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5933         }
5934
5935         btrfs_mark_buffer_dirty(path->nodes[0]);
5936         btrfs_free_path(path);
5937
5938         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5939         if (ret) {
5940                 printk(KERN_ERR "btrfs update block group failed for %llu "
5941                        "%llu\n", (unsigned long long)ins->objectid,
5942                        (unsigned long long)ins->offset);
5943                 BUG();
5944         }
5945         return ret;
5946 }
5947
5948 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5949                                      struct btrfs_root *root,
5950                                      u64 parent, u64 root_objectid,
5951                                      u64 flags, struct btrfs_disk_key *key,
5952                                      int level, struct btrfs_key *ins)
5953 {
5954         int ret;
5955         struct btrfs_fs_info *fs_info = root->fs_info;
5956         struct btrfs_extent_item *extent_item;
5957         struct btrfs_tree_block_info *block_info;
5958         struct btrfs_extent_inline_ref *iref;
5959         struct btrfs_path *path;
5960         struct extent_buffer *leaf;
5961         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5962
5963         path = btrfs_alloc_path();
5964         if (!path)
5965                 return -ENOMEM;
5966
5967         path->leave_spinning = 1;
5968         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5969                                       ins, size);
5970         BUG_ON(ret);
5971
5972         leaf = path->nodes[0];
5973         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5974                                      struct btrfs_extent_item);
5975         btrfs_set_extent_refs(leaf, extent_item, 1);
5976         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5977         btrfs_set_extent_flags(leaf, extent_item,
5978                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5979         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5980
5981         btrfs_set_tree_block_key(leaf, block_info, key);
5982         btrfs_set_tree_block_level(leaf, block_info, level);
5983
5984         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5985         if (parent > 0) {
5986                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5987                 btrfs_set_extent_inline_ref_type(leaf, iref,
5988                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5989                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5990         } else {
5991                 btrfs_set_extent_inline_ref_type(leaf, iref,
5992                                                  BTRFS_TREE_BLOCK_REF_KEY);
5993                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5994         }
5995
5996         btrfs_mark_buffer_dirty(leaf);
5997         btrfs_free_path(path);
5998
5999         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6000         if (ret) {
6001                 printk(KERN_ERR "btrfs update block group failed for %llu "
6002                        "%llu\n", (unsigned long long)ins->objectid,
6003                        (unsigned long long)ins->offset);
6004                 BUG();
6005         }
6006         return ret;
6007 }
6008
6009 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6010                                      struct btrfs_root *root,
6011                                      u64 root_objectid, u64 owner,
6012                                      u64 offset, struct btrfs_key *ins)
6013 {
6014         int ret;
6015
6016         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6017
6018         ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6019                                          ins->offset, 0,
6020                                          root_objectid, owner, offset,
6021                                          BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6022         return ret;
6023 }
6024
6025 /*
6026  * this is used by the tree logging recovery code.  It records that
6027  * an extent has been allocated and makes sure to clear the free
6028  * space cache bits as well
6029  */
6030 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6031                                    struct btrfs_root *root,
6032                                    u64 root_objectid, u64 owner, u64 offset,
6033                                    struct btrfs_key *ins)
6034 {
6035         int ret;
6036         struct btrfs_block_group_cache *block_group;
6037         struct btrfs_caching_control *caching_ctl;
6038         u64 start = ins->objectid;
6039         u64 num_bytes = ins->offset;
6040
6041         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6042         cache_block_group(block_group, trans, NULL, 0);
6043         caching_ctl = get_caching_control(block_group);
6044
6045         if (!caching_ctl) {
6046                 BUG_ON(!block_group_cache_done(block_group));
6047                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6048                 BUG_ON(ret);
6049         } else {
6050                 mutex_lock(&caching_ctl->mutex);
6051
6052                 if (start >= caching_ctl->progress) {
6053                         ret = add_excluded_extent(root, start, num_bytes);
6054                         BUG_ON(ret);
6055                 } else if (start + num_bytes <= caching_ctl->progress) {
6056                         ret = btrfs_remove_free_space(block_group,
6057                                                       start, num_bytes);
6058                         BUG_ON(ret);
6059                 } else {
6060                         num_bytes = caching_ctl->progress - start;
6061                         ret = btrfs_remove_free_space(block_group,
6062                                                       start, num_bytes);
6063                         BUG_ON(ret);
6064
6065                         start = caching_ctl->progress;
6066                         num_bytes = ins->objectid + ins->offset -
6067                                     caching_ctl->progress;
6068                         ret = add_excluded_extent(root, start, num_bytes);
6069                         BUG_ON(ret);
6070                 }
6071
6072                 mutex_unlock(&caching_ctl->mutex);
6073                 put_caching_control(caching_ctl);
6074         }
6075
6076         ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6077                                           RESERVE_ALLOC_NO_ACCOUNT);
6078         BUG_ON(ret);
6079         btrfs_put_block_group(block_group);
6080         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6081                                          0, owner, offset, ins, 1);
6082         return ret;
6083 }
6084
6085 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6086                                             struct btrfs_root *root,
6087                                             u64 bytenr, u32 blocksize,
6088                                             int level)
6089 {
6090         struct extent_buffer *buf;
6091
6092         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6093         if (!buf)
6094                 return ERR_PTR(-ENOMEM);
6095         btrfs_set_header_generation(buf, trans->transid);
6096         btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6097         btrfs_tree_lock(buf);
6098         clean_tree_block(trans, root, buf);
6099
6100         btrfs_set_lock_blocking(buf);
6101         btrfs_set_buffer_uptodate(buf);
6102
6103         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6104                 /*
6105                  * we allow two log transactions at a time, use different
6106                  * EXENT bit to differentiate dirty pages.
6107                  */
6108                 if (root->log_transid % 2 == 0)
6109                         set_extent_dirty(&root->dirty_log_pages, buf->start,
6110                                         buf->start + buf->len - 1, GFP_NOFS);
6111                 else
6112                         set_extent_new(&root->dirty_log_pages, buf->start,
6113                                         buf->start + buf->len - 1, GFP_NOFS);
6114         } else {
6115                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6116                          buf->start + buf->len - 1, GFP_NOFS);
6117         }
6118         trans->blocks_used++;
6119         /* this returns a buffer locked for blocking */
6120         return buf;
6121 }
6122
6123 static struct btrfs_block_rsv *
6124 use_block_rsv(struct btrfs_trans_handle *trans,
6125               struct btrfs_root *root, u32 blocksize)
6126 {
6127         struct btrfs_block_rsv *block_rsv;
6128         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6129         int ret;
6130
6131         block_rsv = get_block_rsv(trans, root);
6132
6133         if (block_rsv->size == 0) {
6134                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6135                 /*
6136                  * If we couldn't reserve metadata bytes try and use some from
6137                  * the global reserve.
6138                  */
6139                 if (ret && block_rsv != global_rsv) {
6140                         ret = block_rsv_use_bytes(global_rsv, blocksize);
6141                         if (!ret)
6142                                 return global_rsv;
6143                         return ERR_PTR(ret);
6144                 } else if (ret) {
6145                         return ERR_PTR(ret);
6146                 }
6147                 return block_rsv;
6148         }
6149
6150         ret = block_rsv_use_bytes(block_rsv, blocksize);
6151         if (!ret)
6152                 return block_rsv;
6153         if (ret) {
6154                 static DEFINE_RATELIMIT_STATE(_rs,
6155                                 DEFAULT_RATELIMIT_INTERVAL,
6156                                 /*DEFAULT_RATELIMIT_BURST*/ 2);
6157                 if (__ratelimit(&_rs)) {
6158                         printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6159                         WARN_ON(1);
6160                 }
6161                 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6162                 if (!ret) {
6163                         return block_rsv;
6164                 } else if (ret && block_rsv != global_rsv) {
6165                         ret = block_rsv_use_bytes(global_rsv, blocksize);
6166                         if (!ret)
6167                                 return global_rsv;
6168                 }
6169         }
6170
6171         return ERR_PTR(-ENOSPC);
6172 }
6173
6174 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6175                             struct btrfs_block_rsv *block_rsv, u32 blocksize)
6176 {
6177         block_rsv_add_bytes(block_rsv, blocksize, 0);
6178         block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6179 }
6180
6181 /*
6182  * finds a free extent and does all the dirty work required for allocation
6183  * returns the key for the extent through ins, and a tree buffer for
6184  * the first block of the extent through buf.
6185  *
6186  * returns the tree buffer or NULL.
6187  */
6188 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6189                                         struct btrfs_root *root, u32 blocksize,
6190                                         u64 parent, u64 root_objectid,
6191                                         struct btrfs_disk_key *key, int level,
6192                                         u64 hint, u64 empty_size, int for_cow)
6193 {
6194         struct btrfs_key ins;
6195         struct btrfs_block_rsv *block_rsv;
6196         struct extent_buffer *buf;
6197         u64 flags = 0;
6198         int ret;
6199
6200
6201         block_rsv = use_block_rsv(trans, root, blocksize);
6202         if (IS_ERR(block_rsv))
6203                 return ERR_CAST(block_rsv);
6204
6205         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6206                                    empty_size, hint, (u64)-1, &ins, 0);
6207         if (ret) {
6208                 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6209                 return ERR_PTR(ret);
6210         }
6211
6212         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6213                                     blocksize, level);
6214         BUG_ON(IS_ERR(buf));
6215
6216         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6217                 if (parent == 0)
6218                         parent = ins.objectid;
6219                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6220         } else
6221                 BUG_ON(parent > 0);
6222
6223         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6224                 struct btrfs_delayed_extent_op *extent_op;
6225                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6226                 BUG_ON(!extent_op);
6227                 if (key)
6228                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
6229                 else
6230                         memset(&extent_op->key, 0, sizeof(extent_op->key));
6231                 extent_op->flags_to_set = flags;
6232                 extent_op->update_key = 1;
6233                 extent_op->update_flags = 1;
6234                 extent_op->is_data = 0;
6235
6236                 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6237                                         ins.objectid,
6238                                         ins.offset, parent, root_objectid,
6239                                         level, BTRFS_ADD_DELAYED_EXTENT,
6240                                         extent_op, for_cow);
6241                 BUG_ON(ret);
6242         }
6243         return buf;
6244 }
6245
6246 struct walk_control {
6247         u64 refs[BTRFS_MAX_LEVEL];
6248         u64 flags[BTRFS_MAX_LEVEL];
6249         struct btrfs_key update_progress;
6250         int stage;
6251         int level;
6252         int shared_level;
6253         int update_ref;
6254         int keep_locks;
6255         int reada_slot;
6256         int reada_count;
6257         int for_reloc;
6258 };
6259
6260 #define DROP_REFERENCE  1
6261 #define UPDATE_BACKREF  2
6262
6263 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6264                                      struct btrfs_root *root,
6265                                      struct walk_control *wc,
6266                                      struct btrfs_path *path)
6267 {
6268         u64 bytenr;
6269         u64 generation;
6270         u64 refs;
6271         u64 flags;
6272         u32 nritems;
6273         u32 blocksize;
6274         struct btrfs_key key;
6275         struct extent_buffer *eb;
6276         int ret;
6277         int slot;
6278         int nread = 0;
6279
6280         if (path->slots[wc->level] < wc->reada_slot) {
6281                 wc->reada_count = wc->reada_count * 2 / 3;
6282                 wc->reada_count = max(wc->reada_count, 2);
6283         } else {
6284                 wc->reada_count = wc->reada_count * 3 / 2;
6285                 wc->reada_count = min_t(int, wc->reada_count,
6286                                         BTRFS_NODEPTRS_PER_BLOCK(root));
6287         }
6288
6289         eb = path->nodes[wc->level];
6290         nritems = btrfs_header_nritems(eb);
6291         blocksize = btrfs_level_size(root, wc->level - 1);
6292
6293         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6294                 if (nread >= wc->reada_count)
6295                         break;
6296
6297                 cond_resched();
6298                 bytenr = btrfs_node_blockptr(eb, slot);
6299                 generation = btrfs_node_ptr_generation(eb, slot);
6300
6301                 if (slot == path->slots[wc->level])
6302                         goto reada;
6303
6304                 if (wc->stage == UPDATE_BACKREF &&
6305                     generation <= root->root_key.offset)
6306                         continue;
6307
6308                 /* We don't lock the tree block, it's OK to be racy here */
6309                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6310                                                &refs, &flags);
6311                 BUG_ON(ret);
6312                 BUG_ON(refs == 0);
6313
6314                 if (wc->stage == DROP_REFERENCE) {
6315                         if (refs == 1)
6316                                 goto reada;
6317
6318                         if (wc->level == 1 &&
6319                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6320                                 continue;
6321                         if (!wc->update_ref ||
6322                             generation <= root->root_key.offset)
6323                                 continue;
6324                         btrfs_node_key_to_cpu(eb, &key, slot);
6325                         ret = btrfs_comp_cpu_keys(&key,
6326                                                   &wc->update_progress);
6327                         if (ret < 0)
6328                                 continue;
6329                 } else {
6330                         if (wc->level == 1 &&
6331                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6332                                 continue;
6333                 }
6334 reada:
6335                 ret = readahead_tree_block(root, bytenr, blocksize,
6336                                            generation);
6337                 if (ret)
6338                         break;
6339                 nread++;
6340         }
6341         wc->reada_slot = slot;
6342 }
6343
6344 /*
6345  * hepler to process tree block while walking down the tree.
6346  *
6347  * when wc->stage == UPDATE_BACKREF, this function updates
6348  * back refs for pointers in the block.
6349  *
6350  * NOTE: return value 1 means we should stop walking down.
6351  */
6352 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6353                                    struct btrfs_root *root,
6354                                    struct btrfs_path *path,
6355                                    struct walk_control *wc, int lookup_info)
6356 {
6357         int level = wc->level;
6358         struct extent_buffer *eb = path->nodes[level];
6359         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6360         int ret;
6361
6362         if (wc->stage == UPDATE_BACKREF &&
6363             btrfs_header_owner(eb) != root->root_key.objectid)
6364                 return 1;
6365
6366         /*
6367          * when reference count of tree block is 1, it won't increase
6368          * again. once full backref flag is set, we never clear it.
6369          */
6370         if (lookup_info &&
6371             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6372              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6373                 BUG_ON(!path->locks[level]);
6374                 ret = btrfs_lookup_extent_info(trans, root,
6375                                                eb->start, eb->len,
6376                                                &wc->refs[level],
6377                                                &wc->flags[level]);
6378                 BUG_ON(ret);
6379                 BUG_ON(wc->refs[level] == 0);
6380         }
6381
6382         if (wc->stage == DROP_REFERENCE) {
6383                 if (wc->refs[level] > 1)
6384                         return 1;
6385
6386                 if (path->locks[level] && !wc->keep_locks) {
6387                         btrfs_tree_unlock_rw(eb, path->locks[level]);
6388                         path->locks[level] = 0;
6389                 }
6390                 return 0;
6391         }
6392
6393         /* wc->stage == UPDATE_BACKREF */
6394         if (!(wc->flags[level] & flag)) {
6395                 BUG_ON(!path->locks[level]);
6396                 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6397                 BUG_ON(ret);
6398                 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6399                 BUG_ON(ret);
6400                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6401                                                   eb->len, flag, 0);
6402                 BUG_ON(ret);
6403                 wc->flags[level] |= flag;
6404         }
6405
6406         /*
6407          * the block is shared by multiple trees, so it's not good to
6408          * keep the tree lock
6409          */
6410         if (path->locks[level] && level > 0) {
6411                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6412                 path->locks[level] = 0;
6413         }
6414         return 0;
6415 }
6416
6417 /*
6418  * hepler to process tree block pointer.
6419  *
6420  * when wc->stage == DROP_REFERENCE, this function checks
6421  * reference count of the block pointed to. if the block
6422  * is shared and we need update back refs for the subtree
6423  * rooted at the block, this function changes wc->stage to
6424  * UPDATE_BACKREF. if the block is shared and there is no
6425  * need to update back, this function drops the reference
6426  * to the block.
6427  *
6428  * NOTE: return value 1 means we should stop walking down.
6429  */
6430 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6431                                  struct btrfs_root *root,
6432                                  struct btrfs_path *path,
6433                                  struct walk_control *wc, int *lookup_info)
6434 {
6435         u64 bytenr;
6436         u64 generation;
6437         u64 parent;
6438         u32 blocksize;
6439         struct btrfs_key key;
6440         struct extent_buffer *next;
6441         int level = wc->level;
6442         int reada = 0;
6443         int ret = 0;
6444
6445         generation = btrfs_node_ptr_generation(path->nodes[level],
6446                                                path->slots[level]);
6447         /*
6448          * if the lower level block was created before the snapshot
6449          * was created, we know there is no need to update back refs
6450          * for the subtree
6451          */
6452         if (wc->stage == UPDATE_BACKREF &&
6453             generation <= root->root_key.offset) {
6454                 *lookup_info = 1;
6455                 return 1;
6456         }
6457
6458         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6459         blocksize = btrfs_level_size(root, level - 1);
6460
6461         next = btrfs_find_tree_block(root, bytenr, blocksize);
6462         if (!next) {
6463                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6464                 if (!next)
6465                         return -ENOMEM;
6466                 reada = 1;
6467         }
6468         btrfs_tree_lock(next);
6469         btrfs_set_lock_blocking(next);
6470
6471         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6472                                        &wc->refs[level - 1],
6473                                        &wc->flags[level - 1]);
6474         BUG_ON(ret);
6475         BUG_ON(wc->refs[level - 1] == 0);
6476         *lookup_info = 0;
6477
6478         if (wc->stage == DROP_REFERENCE) {
6479                 if (wc->refs[level - 1] > 1) {
6480                         if (level == 1 &&
6481                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6482                                 goto skip;
6483
6484                         if (!wc->update_ref ||
6485                             generation <= root->root_key.offset)
6486                                 goto skip;
6487
6488                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6489                                               path->slots[level]);
6490                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6491                         if (ret < 0)
6492                                 goto skip;
6493
6494                         wc->stage = UPDATE_BACKREF;
6495                         wc->shared_level = level - 1;
6496                 }
6497         } else {
6498                 if (level == 1 &&
6499                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6500                         goto skip;
6501         }
6502
6503         if (!btrfs_buffer_uptodate(next, generation)) {
6504                 btrfs_tree_unlock(next);
6505                 free_extent_buffer(next);
6506                 next = NULL;
6507                 *lookup_info = 1;
6508         }
6509
6510         if (!next) {
6511                 if (reada && level == 1)
6512                         reada_walk_down(trans, root, wc, path);
6513                 next = read_tree_block(root, bytenr, blocksize, generation);
6514                 if (!next)
6515                         return -EIO;
6516                 btrfs_tree_lock(next);
6517                 btrfs_set_lock_blocking(next);
6518         }
6519
6520         level--;
6521         BUG_ON(level != btrfs_header_level(next));
6522         path->nodes[level] = next;
6523         path->slots[level] = 0;
6524         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6525         wc->level = level;
6526         if (wc->level == 1)
6527                 wc->reada_slot = 0;
6528         return 0;
6529 skip:
6530         wc->refs[level - 1] = 0;
6531         wc->flags[level - 1] = 0;
6532         if (wc->stage == DROP_REFERENCE) {
6533                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6534                         parent = path->nodes[level]->start;
6535                 } else {
6536                         BUG_ON(root->root_key.objectid !=
6537                                btrfs_header_owner(path->nodes[level]));
6538                         parent = 0;
6539                 }
6540
6541                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6542                                 root->root_key.objectid, level - 1, 0, 0);
6543                 BUG_ON(ret);
6544         }
6545         btrfs_tree_unlock(next);
6546         free_extent_buffer(next);
6547         *lookup_info = 1;
6548         return 1;
6549 }
6550
6551 /*
6552  * hepler to process tree block while walking up the tree.
6553  *
6554  * when wc->stage == DROP_REFERENCE, this function drops
6555  * reference count on the block.
6556  *
6557  * when wc->stage == UPDATE_BACKREF, this function changes
6558  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6559  * to UPDATE_BACKREF previously while processing the block.
6560  *
6561  * NOTE: return value 1 means we should stop walking up.
6562  */
6563 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6564                                  struct btrfs_root *root,
6565                                  struct btrfs_path *path,
6566                                  struct walk_control *wc)
6567 {
6568         int ret;
6569         int level = wc->level;
6570         struct extent_buffer *eb = path->nodes[level];
6571         u64 parent = 0;
6572
6573         if (wc->stage == UPDATE_BACKREF) {
6574                 BUG_ON(wc->shared_level < level);
6575                 if (level < wc->shared_level)
6576                         goto out;
6577
6578                 ret = find_next_key(path, level + 1, &wc->update_progress);
6579                 if (ret > 0)
6580                         wc->update_ref = 0;
6581
6582                 wc->stage = DROP_REFERENCE;
6583                 wc->shared_level = -1;
6584                 path->slots[level] = 0;
6585
6586                 /*
6587                  * check reference count again if the block isn't locked.
6588                  * we should start walking down the tree again if reference
6589                  * count is one.
6590                  */
6591                 if (!path->locks[level]) {
6592                         BUG_ON(level == 0);
6593                         btrfs_tree_lock(eb);
6594                         btrfs_set_lock_blocking(eb);
6595                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6596
6597                         ret = btrfs_lookup_extent_info(trans, root,
6598                                                        eb->start, eb->len,
6599                                                        &wc->refs[level],
6600                                                        &wc->flags[level]);
6601                         BUG_ON(ret);
6602                         BUG_ON(wc->refs[level] == 0);
6603                         if (wc->refs[level] == 1) {
6604                                 btrfs_tree_unlock_rw(eb, path->locks[level]);
6605                                 return 1;
6606                         }
6607                 }
6608         }
6609
6610         /* wc->stage == DROP_REFERENCE */
6611         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6612
6613         if (wc->refs[level] == 1) {
6614                 if (level == 0) {
6615                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6616                                 ret = btrfs_dec_ref(trans, root, eb, 1,
6617                                                     wc->for_reloc);
6618                         else
6619                                 ret = btrfs_dec_ref(trans, root, eb, 0,
6620                                                     wc->for_reloc);
6621                         BUG_ON(ret);
6622                 }
6623                 /* make block locked assertion in clean_tree_block happy */
6624                 if (!path->locks[level] &&
6625                     btrfs_header_generation(eb) == trans->transid) {
6626                         btrfs_tree_lock(eb);
6627                         btrfs_set_lock_blocking(eb);
6628                         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6629                 }
6630                 clean_tree_block(trans, root, eb);
6631         }
6632
6633         if (eb == root->node) {
6634                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6635                         parent = eb->start;
6636                 else
6637                         BUG_ON(root->root_key.objectid !=
6638                                btrfs_header_owner(eb));
6639         } else {
6640                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6641                         parent = path->nodes[level + 1]->start;
6642                 else
6643                         BUG_ON(root->root_key.objectid !=
6644                                btrfs_header_owner(path->nodes[level + 1]));
6645         }
6646
6647         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0);
6648 out:
6649         wc->refs[level] = 0;
6650         wc->flags[level] = 0;
6651         return 0;
6652 }
6653
6654 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6655                                    struct btrfs_root *root,
6656                                    struct btrfs_path *path,
6657                                    struct walk_control *wc)
6658 {
6659         int level = wc->level;
6660         int lookup_info = 1;
6661         int ret;
6662
6663         while (level >= 0) {
6664                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6665                 if (ret > 0)
6666                         break;
6667
6668                 if (level == 0)
6669                         break;
6670
6671                 if (path->slots[level] >=
6672                     btrfs_header_nritems(path->nodes[level]))
6673                         break;
6674
6675                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6676                 if (ret > 0) {
6677                         path->slots[level]++;
6678                         continue;
6679                 } else if (ret < 0)
6680                         return ret;
6681                 level = wc->level;
6682         }
6683         return 0;
6684 }
6685
6686 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6687                                  struct btrfs_root *root,
6688                                  struct btrfs_path *path,
6689                                  struct walk_control *wc, int max_level)
6690 {
6691         int level = wc->level;
6692         int ret;
6693
6694         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6695         while (level < max_level && path->nodes[level]) {
6696                 wc->level = level;
6697                 if (path->slots[level] + 1 <
6698                     btrfs_header_nritems(path->nodes[level])) {
6699                         path->slots[level]++;
6700                         return 0;
6701                 } else {
6702                         ret = walk_up_proc(trans, root, path, wc);
6703                         if (ret > 0)
6704                                 return 0;
6705
6706                         if (path->locks[level]) {
6707                                 btrfs_tree_unlock_rw(path->nodes[level],
6708                                                      path->locks[level]);
6709                                 path->locks[level] = 0;
6710                         }
6711                         free_extent_buffer(path->nodes[level]);
6712                         path->nodes[level] = NULL;
6713                         level++;
6714                 }
6715         }
6716         return 1;
6717 }
6718
6719 /*
6720  * drop a subvolume tree.
6721  *
6722  * this function traverses the tree freeing any blocks that only
6723  * referenced by the tree.
6724  *
6725  * when a shared tree block is found. this function decreases its
6726  * reference count by one. if update_ref is true, this function
6727  * also make sure backrefs for the shared block and all lower level
6728  * blocks are properly updated.
6729  */
6730 void btrfs_drop_snapshot(struct btrfs_root *root,
6731                          struct btrfs_block_rsv *block_rsv, int update_ref,
6732                          int for_reloc)
6733 {
6734         struct btrfs_path *path;
6735         struct btrfs_trans_handle *trans;
6736         struct btrfs_root *tree_root = root->fs_info->tree_root;
6737         struct btrfs_root_item *root_item = &root->root_item;
6738         struct walk_control *wc;
6739         struct btrfs_key key;
6740         int err = 0;
6741         int ret;
6742         int level;
6743
6744         path = btrfs_alloc_path();
6745         if (!path) {
6746                 err = -ENOMEM;
6747                 goto out;
6748         }
6749
6750         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6751         if (!wc) {
6752                 btrfs_free_path(path);
6753                 err = -ENOMEM;
6754                 goto out;
6755         }
6756
6757         trans = btrfs_start_transaction(tree_root, 0);
6758         BUG_ON(IS_ERR(trans));
6759
6760         if (block_rsv)
6761                 trans->block_rsv = block_rsv;
6762
6763         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6764                 level = btrfs_header_level(root->node);
6765                 path->nodes[level] = btrfs_lock_root_node(root);
6766                 btrfs_set_lock_blocking(path->nodes[level]);
6767                 path->slots[level] = 0;
6768                 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6769                 memset(&wc->update_progress, 0,
6770                        sizeof(wc->update_progress));
6771         } else {
6772                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6773                 memcpy(&wc->update_progress, &key,
6774                        sizeof(wc->update_progress));
6775
6776                 level = root_item->drop_level;
6777                 BUG_ON(level == 0);
6778                 path->lowest_level = level;
6779                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6780                 path->lowest_level = 0;
6781                 if (ret < 0) {
6782                         err = ret;
6783                         goto out_free;
6784                 }
6785                 WARN_ON(ret > 0);
6786
6787                 /*
6788                  * unlock our path, this is safe because only this
6789                  * function is allowed to delete this snapshot
6790                  */
6791                 btrfs_unlock_up_safe(path, 0);
6792
6793                 level = btrfs_header_level(root->node);
6794                 while (1) {
6795                         btrfs_tree_lock(path->nodes[level]);
6796                         btrfs_set_lock_blocking(path->nodes[level]);
6797
6798                         ret = btrfs_lookup_extent_info(trans, root,
6799                                                 path->nodes[level]->start,
6800                                                 path->nodes[level]->len,
6801                                                 &wc->refs[level],
6802                                                 &wc->flags[level]);
6803                         BUG_ON(ret);
6804                         BUG_ON(wc->refs[level] == 0);
6805
6806                         if (level == root_item->drop_level)
6807                                 break;
6808
6809                         btrfs_tree_unlock(path->nodes[level]);
6810                         WARN_ON(wc->refs[level] != 1);
6811                         level--;
6812                 }
6813         }
6814
6815         wc->level = level;
6816         wc->shared_level = -1;
6817         wc->stage = DROP_REFERENCE;
6818         wc->update_ref = update_ref;
6819         wc->keep_locks = 0;
6820         wc->for_reloc = for_reloc;
6821         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6822
6823         while (1) {
6824                 ret = walk_down_tree(trans, root, path, wc);
6825                 if (ret < 0) {
6826                         err = ret;
6827                         break;
6828                 }
6829
6830                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6831                 if (ret < 0) {
6832                         err = ret;
6833                         break;
6834                 }
6835
6836                 if (ret > 0) {
6837                         BUG_ON(wc->stage != DROP_REFERENCE);
6838                         break;
6839                 }
6840
6841                 if (wc->stage == DROP_REFERENCE) {
6842                         level = wc->level;
6843                         btrfs_node_key(path->nodes[level],
6844                                        &root_item->drop_progress,
6845                                        path->slots[level]);
6846                         root_item->drop_level = level;
6847                 }
6848
6849                 BUG_ON(wc->level == 0);
6850                 if (btrfs_should_end_transaction(trans, tree_root)) {
6851                         ret = btrfs_update_root(trans, tree_root,
6852                                                 &root->root_key,
6853                                                 root_item);
6854                         BUG_ON(ret);
6855
6856                         btrfs_end_transaction_throttle(trans, tree_root);
6857                         trans = btrfs_start_transaction(tree_root, 0);
6858                         BUG_ON(IS_ERR(trans));
6859                         if (block_rsv)
6860                                 trans->block_rsv = block_rsv;
6861                 }
6862         }
6863         btrfs_release_path(path);
6864         BUG_ON(err);
6865
6866         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6867         BUG_ON(ret);
6868
6869         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6870                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6871                                            NULL, NULL);
6872                 BUG_ON(ret < 0);
6873                 if (ret > 0) {
6874                         /* if we fail to delete the orphan item this time
6875                          * around, it'll get picked up the next time.
6876                          *
6877                          * The most common failure here is just -ENOENT.
6878                          */
6879                         btrfs_del_orphan_item(trans, tree_root,
6880                                               root->root_key.objectid);
6881                 }
6882         }
6883
6884         if (root->in_radix) {
6885                 btrfs_free_fs_root(tree_root->fs_info, root);
6886         } else {
6887                 free_extent_buffer(root->node);
6888                 free_extent_buffer(root->commit_root);
6889                 kfree(root);
6890         }
6891 out_free:
6892         btrfs_end_transaction_throttle(trans, tree_root);
6893         kfree(wc);
6894         btrfs_free_path(path);
6895 out:
6896         if (err)
6897                 btrfs_std_error(root->fs_info, err);
6898         return;
6899 }
6900
6901 /*
6902  * drop subtree rooted at tree block 'node'.
6903  *
6904  * NOTE: this function will unlock and release tree block 'node'
6905  * only used by relocation code
6906  */
6907 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6908                         struct btrfs_root *root,
6909                         struct extent_buffer *node,
6910                         struct extent_buffer *parent)
6911 {
6912         struct btrfs_path *path;
6913         struct walk_control *wc;
6914         int level;
6915         int parent_level;
6916         int ret = 0;
6917         int wret;
6918
6919         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6920
6921         path = btrfs_alloc_path();
6922         if (!path)
6923                 return -ENOMEM;
6924
6925         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6926         if (!wc) {
6927                 btrfs_free_path(path);
6928                 return -ENOMEM;
6929         }
6930
6931         btrfs_assert_tree_locked(parent);
6932         parent_level = btrfs_header_level(parent);
6933         extent_buffer_get(parent);
6934         path->nodes[parent_level] = parent;
6935         path->slots[parent_level] = btrfs_header_nritems(parent);
6936
6937         btrfs_assert_tree_locked(node);
6938         level = btrfs_header_level(node);
6939         path->nodes[level] = node;
6940         path->slots[level] = 0;
6941         path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6942
6943         wc->refs[parent_level] = 1;
6944         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6945         wc->level = level;
6946         wc->shared_level = -1;
6947         wc->stage = DROP_REFERENCE;
6948         wc->update_ref = 0;
6949         wc->keep_locks = 1;
6950         wc->for_reloc = 1;
6951         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6952
6953         while (1) {
6954                 wret = walk_down_tree(trans, root, path, wc);
6955                 if (wret < 0) {
6956                         ret = wret;
6957                         break;
6958                 }
6959
6960                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6961                 if (wret < 0)
6962                         ret = wret;
6963                 if (wret != 0)
6964                         break;
6965         }
6966
6967         kfree(wc);
6968         btrfs_free_path(path);
6969         return ret;
6970 }
6971
6972 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6973 {
6974         u64 num_devices;
6975         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6976                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6977
6978         if (root->fs_info->balance_ctl) {
6979                 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
6980                 u64 tgt = 0;
6981
6982                 /* pick restriper's target profile and return */
6983                 if (flags & BTRFS_BLOCK_GROUP_DATA &&
6984                     bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6985                         tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
6986                 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
6987                            bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6988                         tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
6989                 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
6990                            bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6991                         tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
6992                 }
6993
6994                 if (tgt) {
6995                         /* extended -> chunk profile */
6996                         tgt &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
6997                         return tgt;
6998                 }
6999         }
7000
7001         /*
7002          * we add in the count of missing devices because we want
7003          * to make sure that any RAID levels on a degraded FS
7004          * continue to be honored.
7005          */
7006         num_devices = root->fs_info->fs_devices->rw_devices +
7007                 root->fs_info->fs_devices->missing_devices;
7008
7009         if (num_devices == 1) {
7010                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7011                 stripped = flags & ~stripped;
7012
7013                 /* turn raid0 into single device chunks */
7014                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7015                         return stripped;
7016
7017                 /* turn mirroring into duplication */
7018                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7019                              BTRFS_BLOCK_GROUP_RAID10))
7020                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7021                 return flags;
7022         } else {
7023                 /* they already had raid on here, just return */
7024                 if (flags & stripped)
7025                         return flags;
7026
7027                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7028                 stripped = flags & ~stripped;
7029
7030                 /* switch duplicated blocks with raid1 */
7031                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7032                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7033
7034                 /* turn single device chunks into raid0 */
7035                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7036         }
7037         return flags;
7038 }
7039
7040 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7041 {
7042         struct btrfs_space_info *sinfo = cache->space_info;
7043         u64 num_bytes;
7044         u64 min_allocable_bytes;
7045         int ret = -ENOSPC;
7046
7047
7048         /*
7049          * We need some metadata space and system metadata space for
7050          * allocating chunks in some corner cases until we force to set
7051          * it to be readonly.
7052          */
7053         if ((sinfo->flags &
7054              (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7055             !force)
7056                 min_allocable_bytes = 1 * 1024 * 1024;
7057         else
7058                 min_allocable_bytes = 0;
7059
7060         spin_lock(&sinfo->lock);
7061         spin_lock(&cache->lock);
7062
7063         if (cache->ro) {
7064                 ret = 0;
7065                 goto out;
7066         }
7067
7068         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7069                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7070
7071         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7072             sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7073             min_allocable_bytes <= sinfo->total_bytes) {
7074                 sinfo->bytes_readonly += num_bytes;
7075                 cache->ro = 1;
7076                 ret = 0;
7077         }
7078 out:
7079         spin_unlock(&cache->lock);
7080         spin_unlock(&sinfo->lock);
7081         return ret;
7082 }
7083
7084 int btrfs_set_block_group_ro(struct btrfs_root *root,
7085                              struct btrfs_block_group_cache *cache)
7086
7087 {
7088         struct btrfs_trans_handle *trans;
7089         u64 alloc_flags;
7090         int ret;
7091
7092         BUG_ON(cache->ro);
7093
7094         trans = btrfs_join_transaction(root);
7095         BUG_ON(IS_ERR(trans));
7096
7097         alloc_flags = update_block_group_flags(root, cache->flags);
7098         if (alloc_flags != cache->flags)
7099                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7100                                CHUNK_ALLOC_FORCE);
7101
7102         ret = set_block_group_ro(cache, 0);
7103         if (!ret)
7104                 goto out;
7105         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7106         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7107                              CHUNK_ALLOC_FORCE);
7108         if (ret < 0)
7109                 goto out;
7110         ret = set_block_group_ro(cache, 0);
7111 out:
7112         btrfs_end_transaction(trans, root);
7113         return ret;
7114 }
7115
7116 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7117                             struct btrfs_root *root, u64 type)
7118 {
7119         u64 alloc_flags = get_alloc_profile(root, type);
7120         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7121                               CHUNK_ALLOC_FORCE);
7122 }
7123
7124 /*
7125  * helper to account the unused space of all the readonly block group in the
7126  * list. takes mirrors into account.
7127  */
7128 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7129 {
7130         struct btrfs_block_group_cache *block_group;
7131         u64 free_bytes = 0;
7132         int factor;
7133
7134         list_for_each_entry(block_group, groups_list, list) {
7135                 spin_lock(&block_group->lock);
7136
7137                 if (!block_group->ro) {
7138                         spin_unlock(&block_group->lock);
7139                         continue;
7140                 }
7141
7142                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7143                                           BTRFS_BLOCK_GROUP_RAID10 |
7144                                           BTRFS_BLOCK_GROUP_DUP))
7145                         factor = 2;
7146                 else
7147                         factor = 1;
7148
7149                 free_bytes += (block_group->key.offset -
7150                                btrfs_block_group_used(&block_group->item)) *
7151                                factor;
7152
7153                 spin_unlock(&block_group->lock);
7154         }
7155
7156         return free_bytes;
7157 }
7158
7159 /*
7160  * helper to account the unused space of all the readonly block group in the
7161  * space_info. takes mirrors into account.
7162  */
7163 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7164 {
7165         int i;
7166         u64 free_bytes = 0;
7167
7168         spin_lock(&sinfo->lock);
7169
7170         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7171                 if (!list_empty(&sinfo->block_groups[i]))
7172                         free_bytes += __btrfs_get_ro_block_group_free_space(
7173                                                 &sinfo->block_groups[i]);
7174
7175         spin_unlock(&sinfo->lock);
7176
7177         return free_bytes;
7178 }
7179
7180 int btrfs_set_block_group_rw(struct btrfs_root *root,
7181                               struct btrfs_block_group_cache *cache)
7182 {
7183         struct btrfs_space_info *sinfo = cache->space_info;
7184         u64 num_bytes;
7185
7186         BUG_ON(!cache->ro);
7187
7188         spin_lock(&sinfo->lock);
7189         spin_lock(&cache->lock);
7190         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7191                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7192         sinfo->bytes_readonly -= num_bytes;
7193         cache->ro = 0;
7194         spin_unlock(&cache->lock);
7195         spin_unlock(&sinfo->lock);
7196         return 0;
7197 }
7198
7199 /*
7200  * checks to see if its even possible to relocate this block group.
7201  *
7202  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7203  * ok to go ahead and try.
7204  */
7205 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7206 {
7207         struct btrfs_block_group_cache *block_group;
7208         struct btrfs_space_info *space_info;
7209         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7210         struct btrfs_device *device;
7211         u64 min_free;
7212         u64 dev_min = 1;
7213         u64 dev_nr = 0;
7214         int index;
7215         int full = 0;
7216         int ret = 0;
7217
7218         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7219
7220         /* odd, couldn't find the block group, leave it alone */
7221         if (!block_group)
7222                 return -1;
7223
7224         min_free = btrfs_block_group_used(&block_group->item);
7225
7226         /* no bytes used, we're good */
7227         if (!min_free)
7228                 goto out;
7229
7230         space_info = block_group->space_info;
7231         spin_lock(&space_info->lock);
7232
7233         full = space_info->full;
7234
7235         /*
7236          * if this is the last block group we have in this space, we can't
7237          * relocate it unless we're able to allocate a new chunk below.
7238          *
7239          * Otherwise, we need to make sure we have room in the space to handle
7240          * all of the extents from this block group.  If we can, we're good
7241          */
7242         if ((space_info->total_bytes != block_group->key.offset) &&
7243             (space_info->bytes_used + space_info->bytes_reserved +
7244              space_info->bytes_pinned + space_info->bytes_readonly +
7245              min_free < space_info->total_bytes)) {
7246                 spin_unlock(&space_info->lock);
7247                 goto out;
7248         }
7249         spin_unlock(&space_info->lock);
7250
7251         /*
7252          * ok we don't have enough space, but maybe we have free space on our
7253          * devices to allocate new chunks for relocation, so loop through our
7254          * alloc devices and guess if we have enough space.  However, if we
7255          * were marked as full, then we know there aren't enough chunks, and we
7256          * can just return.
7257          */
7258         ret = -1;
7259         if (full)
7260                 goto out;
7261
7262         /*
7263          * index:
7264          *      0: raid10
7265          *      1: raid1
7266          *      2: dup
7267          *      3: raid0
7268          *      4: single
7269          */
7270         index = get_block_group_index(block_group);
7271         if (index == 0) {
7272                 dev_min = 4;
7273                 /* Divide by 2 */
7274                 min_free >>= 1;
7275         } else if (index == 1) {
7276                 dev_min = 2;
7277         } else if (index == 2) {
7278                 /* Multiply by 2 */
7279                 min_free <<= 1;
7280         } else if (index == 3) {
7281                 dev_min = fs_devices->rw_devices;
7282                 do_div(min_free, dev_min);
7283         }
7284
7285         mutex_lock(&root->fs_info->chunk_mutex);
7286         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7287                 u64 dev_offset;
7288
7289                 /*
7290                  * check to make sure we can actually find a chunk with enough
7291                  * space to fit our block group in.
7292                  */
7293                 if (device->total_bytes > device->bytes_used + min_free) {
7294                         ret = find_free_dev_extent(device, min_free,
7295                                                    &dev_offset, NULL);
7296                         if (!ret)
7297                                 dev_nr++;
7298
7299                         if (dev_nr >= dev_min)
7300                                 break;
7301
7302                         ret = -1;
7303                 }
7304         }
7305         mutex_unlock(&root->fs_info->chunk_mutex);
7306 out:
7307         btrfs_put_block_group(block_group);
7308         return ret;
7309 }
7310
7311 static int find_first_block_group(struct btrfs_root *root,
7312                 struct btrfs_path *path, struct btrfs_key *key)
7313 {
7314         int ret = 0;
7315         struct btrfs_key found_key;
7316         struct extent_buffer *leaf;
7317         int slot;
7318
7319         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7320         if (ret < 0)
7321                 goto out;
7322
7323         while (1) {
7324                 slot = path->slots[0];
7325                 leaf = path->nodes[0];
7326                 if (slot >= btrfs_header_nritems(leaf)) {
7327                         ret = btrfs_next_leaf(root, path);
7328                         if (ret == 0)
7329                                 continue;
7330                         if (ret < 0)
7331                                 goto out;
7332                         break;
7333                 }
7334                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7335
7336                 if (found_key.objectid >= key->objectid &&
7337                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7338                         ret = 0;
7339                         goto out;
7340                 }
7341                 path->slots[0]++;
7342         }
7343 out:
7344         return ret;
7345 }
7346
7347 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7348 {
7349         struct btrfs_block_group_cache *block_group;
7350         u64 last = 0;
7351
7352         while (1) {
7353                 struct inode *inode;
7354
7355                 block_group = btrfs_lookup_first_block_group(info, last);
7356                 while (block_group) {
7357                         spin_lock(&block_group->lock);
7358                         if (block_group->iref)
7359                                 break;
7360                         spin_unlock(&block_group->lock);
7361                         block_group = next_block_group(info->tree_root,
7362                                                        block_group);
7363                 }
7364                 if (!block_group) {
7365                         if (last == 0)
7366                                 break;
7367                         last = 0;
7368                         continue;
7369                 }
7370
7371                 inode = block_group->inode;
7372                 block_group->iref = 0;
7373                 block_group->inode = NULL;
7374                 spin_unlock(&block_group->lock);
7375                 iput(inode);
7376                 last = block_group->key.objectid + block_group->key.offset;
7377                 btrfs_put_block_group(block_group);
7378         }
7379 }
7380
7381 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7382 {
7383         struct btrfs_block_group_cache *block_group;
7384         struct btrfs_space_info *space_info;
7385         struct btrfs_caching_control *caching_ctl;
7386         struct rb_node *n;
7387
7388         down_write(&info->extent_commit_sem);
7389         while (!list_empty(&info->caching_block_groups)) {
7390                 caching_ctl = list_entry(info->caching_block_groups.next,
7391                                          struct btrfs_caching_control, list);
7392                 list_del(&caching_ctl->list);
7393                 put_caching_control(caching_ctl);
7394         }
7395         up_write(&info->extent_commit_sem);
7396
7397         spin_lock(&info->block_group_cache_lock);
7398         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7399                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7400                                        cache_node);
7401                 rb_erase(&block_group->cache_node,
7402                          &info->block_group_cache_tree);
7403                 spin_unlock(&info->block_group_cache_lock);
7404
7405                 down_write(&block_group->space_info->groups_sem);
7406                 list_del(&block_group->list);
7407                 up_write(&block_group->space_info->groups_sem);
7408
7409                 if (block_group->cached == BTRFS_CACHE_STARTED)
7410                         wait_block_group_cache_done(block_group);
7411
7412                 /*
7413                  * We haven't cached this block group, which means we could
7414                  * possibly have excluded extents on this block group.
7415                  */
7416                 if (block_group->cached == BTRFS_CACHE_NO)
7417                         free_excluded_extents(info->extent_root, block_group);
7418
7419                 btrfs_remove_free_space_cache(block_group);
7420                 btrfs_put_block_group(block_group);
7421
7422                 spin_lock(&info->block_group_cache_lock);
7423         }
7424         spin_unlock(&info->block_group_cache_lock);
7425
7426         /* now that all the block groups are freed, go through and
7427          * free all the space_info structs.  This is only called during
7428          * the final stages of unmount, and so we know nobody is
7429          * using them.  We call synchronize_rcu() once before we start,
7430          * just to be on the safe side.
7431          */
7432         synchronize_rcu();
7433
7434         release_global_block_rsv(info);
7435
7436         while(!list_empty(&info->space_info)) {
7437                 space_info = list_entry(info->space_info.next,
7438                                         struct btrfs_space_info,
7439                                         list);
7440                 if (space_info->bytes_pinned > 0 ||
7441                     space_info->bytes_reserved > 0 ||
7442                     space_info->bytes_may_use > 0) {
7443                         WARN_ON(1);
7444                         dump_space_info(space_info, 0, 0);
7445                 }
7446                 list_del(&space_info->list);
7447                 kfree(space_info);
7448         }
7449         return 0;
7450 }
7451
7452 static void __link_block_group(struct btrfs_space_info *space_info,
7453                                struct btrfs_block_group_cache *cache)
7454 {
7455         int index = get_block_group_index(cache);
7456
7457         down_write(&space_info->groups_sem);
7458         list_add_tail(&cache->list, &space_info->block_groups[index]);
7459         up_write(&space_info->groups_sem);
7460 }
7461
7462 int btrfs_read_block_groups(struct btrfs_root *root)
7463 {
7464         struct btrfs_path *path;
7465         int ret;
7466         struct btrfs_block_group_cache *cache;
7467         struct btrfs_fs_info *info = root->fs_info;
7468         struct btrfs_space_info *space_info;
7469         struct btrfs_key key;
7470         struct btrfs_key found_key;
7471         struct extent_buffer *leaf;
7472         int need_clear = 0;
7473         u64 cache_gen;
7474
7475         root = info->extent_root;
7476         key.objectid = 0;
7477         key.offset = 0;
7478         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7479         path = btrfs_alloc_path();
7480         if (!path)
7481                 return -ENOMEM;
7482         path->reada = 1;
7483
7484         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7485         if (btrfs_test_opt(root, SPACE_CACHE) &&
7486             btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7487                 need_clear = 1;
7488         if (btrfs_test_opt(root, CLEAR_CACHE))
7489                 need_clear = 1;
7490
7491         while (1) {
7492                 ret = find_first_block_group(root, path, &key);
7493                 if (ret > 0)
7494                         break;
7495                 if (ret != 0)
7496                         goto error;
7497                 leaf = path->nodes[0];
7498                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7499                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7500                 if (!cache) {
7501                         ret = -ENOMEM;
7502                         goto error;
7503                 }
7504                 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7505                                                 GFP_NOFS);
7506                 if (!cache->free_space_ctl) {
7507                         kfree(cache);
7508                         ret = -ENOMEM;
7509                         goto error;
7510                 }
7511
7512                 atomic_set(&cache->count, 1);
7513                 spin_lock_init(&cache->lock);
7514                 cache->fs_info = info;
7515                 INIT_LIST_HEAD(&cache->list);
7516                 INIT_LIST_HEAD(&cache->cluster_list);
7517
7518                 if (need_clear)
7519                         cache->disk_cache_state = BTRFS_DC_CLEAR;
7520
7521                 read_extent_buffer(leaf, &cache->item,
7522                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7523                                    sizeof(cache->item));
7524                 memcpy(&cache->key, &found_key, sizeof(found_key));
7525
7526                 key.objectid = found_key.objectid + found_key.offset;
7527                 btrfs_release_path(path);
7528                 cache->flags = btrfs_block_group_flags(&cache->item);
7529                 cache->sectorsize = root->sectorsize;
7530
7531                 btrfs_init_free_space_ctl(cache);
7532
7533                 /*
7534                  * We need to exclude the super stripes now so that the space
7535                  * info has super bytes accounted for, otherwise we'll think
7536                  * we have more space than we actually do.
7537                  */
7538                 exclude_super_stripes(root, cache);
7539
7540                 /*
7541                  * check for two cases, either we are full, and therefore
7542                  * don't need to bother with the caching work since we won't
7543                  * find any space, or we are empty, and we can just add all
7544                  * the space in and be done with it.  This saves us _alot_ of
7545                  * time, particularly in the full case.
7546                  */
7547                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7548                         cache->last_byte_to_unpin = (u64)-1;
7549                         cache->cached = BTRFS_CACHE_FINISHED;
7550                         free_excluded_extents(root, cache);
7551                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7552                         cache->last_byte_to_unpin = (u64)-1;
7553                         cache->cached = BTRFS_CACHE_FINISHED;
7554                         add_new_free_space(cache, root->fs_info,
7555                                            found_key.objectid,
7556                                            found_key.objectid +
7557                                            found_key.offset);
7558                         free_excluded_extents(root, cache);
7559                 }
7560
7561                 ret = update_space_info(info, cache->flags, found_key.offset,
7562                                         btrfs_block_group_used(&cache->item),
7563                                         &space_info);
7564                 BUG_ON(ret);
7565                 cache->space_info = space_info;
7566                 spin_lock(&cache->space_info->lock);
7567                 cache->space_info->bytes_readonly += cache->bytes_super;
7568                 spin_unlock(&cache->space_info->lock);
7569
7570                 __link_block_group(space_info, cache);
7571
7572                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7573                 BUG_ON(ret);
7574
7575                 set_avail_alloc_bits(root->fs_info, cache->flags);
7576                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7577                         set_block_group_ro(cache, 1);
7578         }
7579
7580         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7581                 if (!(get_alloc_profile(root, space_info->flags) &
7582                       (BTRFS_BLOCK_GROUP_RAID10 |
7583                        BTRFS_BLOCK_GROUP_RAID1 |
7584                        BTRFS_BLOCK_GROUP_DUP)))
7585                         continue;
7586                 /*
7587                  * avoid allocating from un-mirrored block group if there are
7588                  * mirrored block groups.
7589                  */
7590                 list_for_each_entry(cache, &space_info->block_groups[3], list)
7591                         set_block_group_ro(cache, 1);
7592                 list_for_each_entry(cache, &space_info->block_groups[4], list)
7593                         set_block_group_ro(cache, 1);
7594         }
7595
7596         init_global_block_rsv(info);
7597         ret = 0;
7598 error:
7599         btrfs_free_path(path);
7600         return ret;
7601 }
7602
7603 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7604                            struct btrfs_root *root, u64 bytes_used,
7605                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7606                            u64 size)
7607 {
7608         int ret;
7609         struct btrfs_root *extent_root;
7610         struct btrfs_block_group_cache *cache;
7611
7612         extent_root = root->fs_info->extent_root;
7613
7614         root->fs_info->last_trans_log_full_commit = trans->transid;
7615
7616         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7617         if (!cache)
7618                 return -ENOMEM;
7619         cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7620                                         GFP_NOFS);
7621         if (!cache->free_space_ctl) {
7622                 kfree(cache);
7623                 return -ENOMEM;
7624         }
7625
7626         cache->key.objectid = chunk_offset;
7627         cache->key.offset = size;
7628         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7629         cache->sectorsize = root->sectorsize;
7630         cache->fs_info = root->fs_info;
7631
7632         atomic_set(&cache->count, 1);
7633         spin_lock_init(&cache->lock);
7634         INIT_LIST_HEAD(&cache->list);
7635         INIT_LIST_HEAD(&cache->cluster_list);
7636
7637         btrfs_init_free_space_ctl(cache);
7638
7639         btrfs_set_block_group_used(&cache->item, bytes_used);
7640         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7641         cache->flags = type;
7642         btrfs_set_block_group_flags(&cache->item, type);
7643
7644         cache->last_byte_to_unpin = (u64)-1;
7645         cache->cached = BTRFS_CACHE_FINISHED;
7646         exclude_super_stripes(root, cache);
7647
7648         add_new_free_space(cache, root->fs_info, chunk_offset,
7649                            chunk_offset + size);
7650
7651         free_excluded_extents(root, cache);
7652
7653         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7654                                 &cache->space_info);
7655         BUG_ON(ret);
7656         update_global_block_rsv(root->fs_info);
7657
7658         spin_lock(&cache->space_info->lock);
7659         cache->space_info->bytes_readonly += cache->bytes_super;
7660         spin_unlock(&cache->space_info->lock);
7661
7662         __link_block_group(cache->space_info, cache);
7663
7664         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7665         BUG_ON(ret);
7666
7667         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7668                                 sizeof(cache->item));
7669         BUG_ON(ret);
7670
7671         set_avail_alloc_bits(extent_root->fs_info, type);
7672
7673         return 0;
7674 }
7675
7676 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7677 {
7678         u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
7679
7680         /* chunk -> extended profile */
7681         if (extra_flags == 0)
7682                 extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
7683
7684         if (flags & BTRFS_BLOCK_GROUP_DATA)
7685                 fs_info->avail_data_alloc_bits &= ~extra_flags;
7686         if (flags & BTRFS_BLOCK_GROUP_METADATA)
7687                 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7688         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7689                 fs_info->avail_system_alloc_bits &= ~extra_flags;
7690 }
7691
7692 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7693                              struct btrfs_root *root, u64 group_start)
7694 {
7695         struct btrfs_path *path;
7696         struct btrfs_block_group_cache *block_group;
7697         struct btrfs_free_cluster *cluster;
7698         struct btrfs_root *tree_root = root->fs_info->tree_root;
7699         struct btrfs_key key;
7700         struct inode *inode;
7701         int ret;
7702         int index;
7703         int factor;
7704
7705         root = root->fs_info->extent_root;
7706
7707         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7708         BUG_ON(!block_group);
7709         BUG_ON(!block_group->ro);
7710
7711         /*
7712          * Free the reserved super bytes from this block group before
7713          * remove it.
7714          */
7715         free_excluded_extents(root, block_group);
7716
7717         memcpy(&key, &block_group->key, sizeof(key));
7718         index = get_block_group_index(block_group);
7719         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7720                                   BTRFS_BLOCK_GROUP_RAID1 |
7721                                   BTRFS_BLOCK_GROUP_RAID10))
7722                 factor = 2;
7723         else
7724                 factor = 1;
7725
7726         /* make sure this block group isn't part of an allocation cluster */
7727         cluster = &root->fs_info->data_alloc_cluster;
7728         spin_lock(&cluster->refill_lock);
7729         btrfs_return_cluster_to_free_space(block_group, cluster);
7730         spin_unlock(&cluster->refill_lock);
7731
7732         /*
7733          * make sure this block group isn't part of a metadata
7734          * allocation cluster
7735          */
7736         cluster = &root->fs_info->meta_alloc_cluster;
7737         spin_lock(&cluster->refill_lock);
7738         btrfs_return_cluster_to_free_space(block_group, cluster);
7739         spin_unlock(&cluster->refill_lock);
7740
7741         path = btrfs_alloc_path();
7742         if (!path) {
7743                 ret = -ENOMEM;
7744                 goto out;
7745         }
7746
7747         inode = lookup_free_space_inode(tree_root, block_group, path);
7748         if (!IS_ERR(inode)) {
7749                 ret = btrfs_orphan_add(trans, inode);
7750                 BUG_ON(ret);
7751                 clear_nlink(inode);
7752                 /* One for the block groups ref */
7753                 spin_lock(&block_group->lock);
7754                 if (block_group->iref) {
7755                         block_group->iref = 0;
7756                         block_group->inode = NULL;
7757                         spin_unlock(&block_group->lock);
7758                         iput(inode);
7759                 } else {
7760                         spin_unlock(&block_group->lock);
7761                 }
7762                 /* One for our lookup ref */
7763                 btrfs_add_delayed_iput(inode);
7764         }
7765
7766         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7767         key.offset = block_group->key.objectid;
7768         key.type = 0;
7769
7770         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7771         if (ret < 0)
7772                 goto out;
7773         if (ret > 0)
7774                 btrfs_release_path(path);
7775         if (ret == 0) {
7776                 ret = btrfs_del_item(trans, tree_root, path);
7777                 if (ret)
7778                         goto out;
7779                 btrfs_release_path(path);
7780         }
7781
7782         spin_lock(&root->fs_info->block_group_cache_lock);
7783         rb_erase(&block_group->cache_node,
7784                  &root->fs_info->block_group_cache_tree);
7785         spin_unlock(&root->fs_info->block_group_cache_lock);
7786
7787         down_write(&block_group->space_info->groups_sem);
7788         /*
7789          * we must use list_del_init so people can check to see if they
7790          * are still on the list after taking the semaphore
7791          */
7792         list_del_init(&block_group->list);
7793         if (list_empty(&block_group->space_info->block_groups[index]))
7794                 clear_avail_alloc_bits(root->fs_info, block_group->flags);
7795         up_write(&block_group->space_info->groups_sem);
7796
7797         if (block_group->cached == BTRFS_CACHE_STARTED)
7798                 wait_block_group_cache_done(block_group);
7799
7800         btrfs_remove_free_space_cache(block_group);
7801
7802         spin_lock(&block_group->space_info->lock);
7803         block_group->space_info->total_bytes -= block_group->key.offset;
7804         block_group->space_info->bytes_readonly -= block_group->key.offset;
7805         block_group->space_info->disk_total -= block_group->key.offset * factor;
7806         spin_unlock(&block_group->space_info->lock);
7807
7808         memcpy(&key, &block_group->key, sizeof(key));
7809
7810         btrfs_clear_space_info_full(root->fs_info);
7811
7812         btrfs_put_block_group(block_group);
7813         btrfs_put_block_group(block_group);
7814
7815         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7816         if (ret > 0)
7817                 ret = -EIO;
7818         if (ret < 0)
7819                 goto out;
7820
7821         ret = btrfs_del_item(trans, root, path);
7822 out:
7823         btrfs_free_path(path);
7824         return ret;
7825 }
7826
7827 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7828 {
7829         struct btrfs_space_info *space_info;
7830         struct btrfs_super_block *disk_super;
7831         u64 features;
7832         u64 flags;
7833         int mixed = 0;
7834         int ret;
7835
7836         disk_super = fs_info->super_copy;
7837         if (!btrfs_super_root(disk_super))
7838                 return 1;
7839
7840         features = btrfs_super_incompat_flags(disk_super);
7841         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7842                 mixed = 1;
7843
7844         flags = BTRFS_BLOCK_GROUP_SYSTEM;
7845         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7846         if (ret)
7847                 goto out;
7848
7849         if (mixed) {
7850                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7851                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7852         } else {
7853                 flags = BTRFS_BLOCK_GROUP_METADATA;
7854                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7855                 if (ret)
7856                         goto out;
7857
7858                 flags = BTRFS_BLOCK_GROUP_DATA;
7859                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7860         }
7861 out:
7862         return ret;
7863 }
7864
7865 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7866 {
7867         return unpin_extent_range(root, start, end);
7868 }
7869
7870 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7871                                u64 num_bytes, u64 *actual_bytes)
7872 {
7873         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7874 }
7875
7876 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7877 {
7878         struct btrfs_fs_info *fs_info = root->fs_info;
7879         struct btrfs_block_group_cache *cache = NULL;
7880         u64 group_trimmed;
7881         u64 start;
7882         u64 end;
7883         u64 trimmed = 0;
7884         int ret = 0;
7885
7886         cache = btrfs_lookup_block_group(fs_info, range->start);
7887
7888         while (cache) {
7889                 if (cache->key.objectid >= (range->start + range->len)) {
7890                         btrfs_put_block_group(cache);
7891                         break;
7892                 }
7893
7894                 start = max(range->start, cache->key.objectid);
7895                 end = min(range->start + range->len,
7896                                 cache->key.objectid + cache->key.offset);
7897
7898                 if (end - start >= range->minlen) {
7899                         if (!block_group_cache_done(cache)) {
7900                                 ret = cache_block_group(cache, NULL, root, 0);
7901                                 if (!ret)
7902                                         wait_block_group_cache_done(cache);
7903                         }
7904                         ret = btrfs_trim_block_group(cache,
7905                                                      &group_trimmed,
7906                                                      start,
7907                                                      end,
7908                                                      range->minlen);
7909
7910                         trimmed += group_trimmed;
7911                         if (ret) {
7912                                 btrfs_put_block_group(cache);
7913                                 break;
7914                         }
7915                 }
7916
7917                 cache = next_block_group(fs_info->tree_root, cache);
7918         }
7919
7920         range->len = trimmed;
7921         return ret;
7922 }