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 "compat.h"
  27 #include "hash.h"
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "print-tree.h"
  31 #include "transaction.h"
  32 #include "volumes.h"
  33 #include "locking.h"
  34 #include "free-space-cache.h"
  35
  36 static int update_block_group(struct btrfs_trans_handle *trans,
  37                               struct btrfs_root *root,
  38                               u64 bytenr, u64 num_bytes, int alloc);
  39 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
  40                                  u64 num_bytes, int reserve, int sinfo);
  41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  42                                 struct btrfs_root *root,
  43                                 u64 bytenr, u64 num_bytes, u64 parent,
  44                                 u64 root_objectid, u64 owner_objectid,
  45                                 u64 owner_offset, int refs_to_drop,
  46                                 struct btrfs_delayed_extent_op *extra_op);
  47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  48                                     struct extent_buffer *leaf,
  49                                     struct btrfs_extent_item *ei);
  50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  51                                       struct btrfs_root *root,
  52                                       u64 parent, u64 root_objectid,
  53                                       u64 flags, u64 owner, u64 offset,
  54                                       struct btrfs_key *ins, int ref_mod);
  55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  56                                      struct btrfs_root *root,
  57                                      u64 parent, u64 root_objectid,
  58                                      u64 flags, struct btrfs_disk_key *key,
  59                                      int level, struct btrfs_key *ins);
  60 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  61                           struct btrfs_root *extent_root, u64 alloc_bytes,
  62                           u64 flags, int force);
  63 static int find_next_key(struct btrfs_path *path, int level,
  64                          struct btrfs_key *key);
  65 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  66                             int dump_block_groups);
  67
  68 static noinline int
  69 block_group_cache_done(struct btrfs_block_group_cache *cache)
  70 {
  71         smp_mb();
  72         return cache->cached == BTRFS_CACHE_FINISHED;
  73 }
  74
  75 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  76 {
  77         return (cache->flags & bits) == bits;
  78 }
  79
  80 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
  81 {
  82         atomic_inc(&cache->count);
  83 }
  84
  85 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
  86 {
  87         if (atomic_dec_and_test(&cache->count)) {
  88                 WARN_ON(cache->pinned > 0);
  89                 WARN_ON(cache->reserved > 0);
  90                 WARN_ON(cache->reserved_pinned > 0);
  91                 kfree(cache);
  92         }
  93 }
  94
  95 /*
  96  * this adds the block group to the fs_info rb tree for the block group
  97  * cache
  98  */
  99 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 100                                 struct btrfs_block_group_cache *block_group)
 101 {
 102         struct rb_node **p;
 103         struct rb_node *parent = NULL;
 104         struct btrfs_block_group_cache *cache;
 105
 106         spin_lock(&info->block_group_cache_lock);
 107         p = &info->block_group_cache_tree.rb_node;
 108
 109         while (*p) {
 110                 parent = *p;
 111                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 112                                  cache_node);
 113                 if (block_group->key.objectid < cache->key.objectid) {
 114                         p = &(*p)->rb_left;
 115                 } else if (block_group->key.objectid > cache->key.objectid) {
 116                         p = &(*p)->rb_right;
 117                 } else {
 118                         spin_unlock(&info->block_group_cache_lock);
 119                         return -EEXIST;
 120                 }
 121         }
 122
 123         rb_link_node(&block_group->cache_node, parent, p);
 124         rb_insert_color(&block_group->cache_node,
 125                         &info->block_group_cache_tree);
 126         spin_unlock(&info->block_group_cache_lock);
 127
 128         return 0;
 129 }
 130
 131 /*
 132  * This will return the block group at or after bytenr if contains is 0, else
 133  * it will return the block group that contains the bytenr
 134  */
 135 static struct btrfs_block_group_cache *
 136 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 137                               int contains)
 138 {
 139         struct btrfs_block_group_cache *cache, *ret = NULL;
 140         struct rb_node *n;
 141         u64 end, start;
 142
 143         spin_lock(&info->block_group_cache_lock);
 144         n = info->block_group_cache_tree.rb_node;
 145
 146         while (n) {
 147                 cache = rb_entry(n, struct btrfs_block_group_cache,
 148                                  cache_node);
 149                 end = cache->key.objectid + cache->key.offset - 1;
 150                 start = cache->key.objectid;
 151
 152                 if (bytenr < start) {
 153                         if (!contains && (!ret || start < ret->key.objectid))
 154                                 ret = cache;
 155                         n = n->rb_left;
 156                 } else if (bytenr > start) {
 157                         if (contains && bytenr <= end) {
 158                                 ret = cache;
 159                                 break;
 160                         }
 161                         n = n->rb_right;
 162                 } else {
 163                         ret = cache;
 164                         break;
 165                 }
 166         }
 167         if (ret)
 168                 btrfs_get_block_group(ret);
 169         spin_unlock(&info->block_group_cache_lock);
 170
 171         return ret;
 172 }
 173
 174 static int add_excluded_extent(struct btrfs_root *root,
 175                                u64 start, u64 num_bytes)
 176 {
 177         u64 end = start + num_bytes - 1;
 178         set_extent_bits(&root->fs_info->freed_extents[0],
 179                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 180         set_extent_bits(&root->fs_info->freed_extents[1],
 181                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 182         return 0;
 183 }
 184
 185 static void free_excluded_extents(struct btrfs_root *root,
 186                                   struct btrfs_block_group_cache *cache)
 187 {
 188         u64 start, end;
 189
 190         start = cache->key.objectid;
 191         end = start + cache->key.offset - 1;
 192
 193         clear_extent_bits(&root->fs_info->freed_extents[0],
 194                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 195         clear_extent_bits(&root->fs_info->freed_extents[1],
 196                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 197 }
 198
 199 static int exclude_super_stripes(struct btrfs_root *root,
 200                                  struct btrfs_block_group_cache *cache)
 201 {
 202         u64 bytenr;
 203         u64 *logical;
 204         int stripe_len;
 205         int i, nr, ret;
 206
 207         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 208                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 209                 cache->bytes_super += stripe_len;
 210                 ret = add_excluded_extent(root, cache->key.objectid,
 211                                           stripe_len);
 212                 BUG_ON(ret);
 213         }
 214
 215         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 216                 bytenr = btrfs_sb_offset(i);
 217                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 218                                        cache->key.objectid, bytenr,
 219                                        0, &logical, &nr, &stripe_len);
 220                 BUG_ON(ret);
 221
 222                 while (nr--) {
 223                         cache->bytes_super += stripe_len;
 224                         ret = add_excluded_extent(root, logical[nr],
 225                                                   stripe_len);
 226                         BUG_ON(ret);
 227                 }
 228
 229                 kfree(logical);
 230         }
 231         return 0;
 232 }
 233
 234 static struct btrfs_caching_control *
 235 get_caching_control(struct btrfs_block_group_cache *cache)
 236 {
 237         struct btrfs_caching_control *ctl;
 238
 239         spin_lock(&cache->lock);
 240         if (cache->cached != BTRFS_CACHE_STARTED) {
 241                 spin_unlock(&cache->lock);
 242                 return NULL;
 243         }
 244
 245         /* We're loading it the fast way, so we don't have a caching_ctl. */
 246         if (!cache->caching_ctl) {
 247                 spin_unlock(&cache->lock);
 248                 return NULL;
 249         }
 250
 251         ctl = cache->caching_ctl;
 252         atomic_inc(&ctl->count);
 253         spin_unlock(&cache->lock);
 254         return ctl;
 255 }
 256
 257 static void put_caching_control(struct btrfs_caching_control *ctl)
 258 {
 259         if (atomic_dec_and_test(&ctl->count))
 260                 kfree(ctl);
 261 }
 262
 263 /*
 264  * this is only called by cache_block_group, since we could have freed extents
 265  * we need to check the pinned_extents for any extents that can't be used yet
 266  * since their free space will be released as soon as the transaction commits.
 267  */
 268 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 269                               struct btrfs_fs_info *info, u64 start, u64 end)
 270 {
 271         u64 extent_start, extent_end, size, total_added = 0;
 272         int ret;
 273
 274         while (start < end) {
 275                 ret = find_first_extent_bit(info->pinned_extents, start,
 276                                             &extent_start, &extent_end,
 277                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 278                 if (ret)
 279                         break;
 280
 281                 if (extent_start <= start) {
 282                         start = extent_end + 1;
 283                 } else if (extent_start > start && extent_start < end) {
 284                         size = extent_start - start;
 285                         total_added += size;
 286                         ret = btrfs_add_free_space(block_group, start,
 287                                                    size);
 288                         BUG_ON(ret);
 289                         start = extent_end + 1;
 290                 } else {
 291                         break;
 292                 }
 293         }
 294
 295         if (start < end) {
 296                 size = end - start;
 297                 total_added += size;
 298                 ret = btrfs_add_free_space(block_group, start, size);
 299                 BUG_ON(ret);
 300         }
 301
 302         return total_added;
 303 }
 304
 305 static int caching_kthread(void *data)
 306 {
 307         struct btrfs_block_group_cache *block_group = data;
 308         struct btrfs_fs_info *fs_info = block_group->fs_info;
 309         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 310         struct btrfs_root *extent_root = fs_info->extent_root;
 311         struct btrfs_path *path;
 312         struct extent_buffer *leaf;
 313         struct btrfs_key key;
 314         u64 total_found = 0;
 315         u64 last = 0;
 316         u32 nritems;
 317         int ret = 0;
 318
 319         path = btrfs_alloc_path();
 320         if (!path)
 321                 return -ENOMEM;
 322
 323         exclude_super_stripes(extent_root, block_group);
 324         spin_lock(&block_group->space_info->lock);
 325         block_group->space_info->bytes_readonly += block_group->bytes_super;
 326         spin_unlock(&block_group->space_info->lock);
 327
 328         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 329
 330         /*
 331          * We don't want to deadlock with somebody trying to allocate a new
 332          * extent for the extent root while also trying to search the extent
 333          * root to add free space.  So we skip locking and search the commit
 334          * root, since its read-only
 335          */
 336         path->skip_locking = 1;
 337         path->search_commit_root = 1;
 338         path->reada = 2;
 339
 340         key.objectid = last;
 341         key.offset = 0;
 342         key.type = BTRFS_EXTENT_ITEM_KEY;
 343 again:
 344         mutex_lock(&caching_ctl->mutex);
 345         /* need to make sure the commit_root doesn't disappear */
 346         down_read(&fs_info->extent_commit_sem);
 347
 348         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 349         if (ret < 0)
 350                 goto err;
 351
 352         leaf = path->nodes[0];
 353         nritems = btrfs_header_nritems(leaf);
 354
 355         while (1) {
 356                 smp_mb();
 357                 if (fs_info->closing > 1) {
 358                         last = (u64)-1;
 359                         break;
 360                 }
 361
 362                 if (path->slots[0] < nritems) {
 363                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 364                 } else {
 365                         ret = find_next_key(path, 0, &key);
 366                         if (ret)
 367                                 break;
 368
 369                         caching_ctl->progress = last;
 370                         btrfs_release_path(extent_root, path);
 371                         up_read(&fs_info->extent_commit_sem);
 372                         mutex_unlock(&caching_ctl->mutex);
 373                         if (btrfs_transaction_in_commit(fs_info))
 374                                 schedule_timeout(1);
 375                         else
 376                                 cond_resched();
 377                         goto again;
 378                 }
 379
 380                 if (key.objectid < block_group->key.objectid) {
 381                         path->slots[0]++;
 382                         continue;
 383                 }
 384
 385                 if (key.objectid >= block_group->key.objectid +
 386                     block_group->key.offset)
 387                         break;
 388
 389                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 390                         total_found += add_new_free_space(block_group,
 391                                                           fs_info, last,
 392                                                           key.objectid);
 393                         last = key.objectid + key.offset;
 394
 395                         if (total_found > (1024 * 1024 * 2)) {
 396                                 total_found = 0;
 397                                 wake_up(&caching_ctl->wait);
 398                         }
 399                 }
 400                 path->slots[0]++;
 401         }
 402         ret = 0;
 403
 404         total_found += add_new_free_space(block_group, fs_info, last,
 405                                           block_group->key.objectid +
 406                                           block_group->key.offset);
 407         caching_ctl->progress = (u64)-1;
 408
 409         spin_lock(&block_group->lock);
 410         block_group->caching_ctl = NULL;
 411         block_group->cached = BTRFS_CACHE_FINISHED;
 412         spin_unlock(&block_group->lock);
 413
 414 err:
 415         btrfs_free_path(path);
 416         up_read(&fs_info->extent_commit_sem);
 417
 418         free_excluded_extents(extent_root, block_group);
 419
 420         mutex_unlock(&caching_ctl->mutex);
 421         wake_up(&caching_ctl->wait);
 422
 423         put_caching_control(caching_ctl);
 424         atomic_dec(&block_group->space_info->caching_threads);
 425         btrfs_put_block_group(block_group);
 426
 427         return 0;
 428 }
 429
 430 static int cache_block_group(struct btrfs_block_group_cache *cache,
 431                              struct btrfs_trans_handle *trans,
 432                              int load_cache_only)
 433 {
 434         struct btrfs_fs_info *fs_info = cache->fs_info;
 435         struct btrfs_caching_control *caching_ctl;
 436         struct task_struct *tsk;
 437         int ret = 0;
 438
 439         smp_mb();
 440         if (cache->cached != BTRFS_CACHE_NO)
 441                 return 0;
 442
 443         /*
 444          * We can't do the read from on-disk cache during a commit since we need
 445          * to have the normal tree locking.
 446          */
 447         if (!trans->transaction->in_commit) {
 448                 spin_lock(&cache->lock);
 449                 if (cache->cached != BTRFS_CACHE_NO) {
 450                         spin_unlock(&cache->lock);
 451                         return 0;
 452                 }
 453                 cache->cached = BTRFS_CACHE_STARTED;
 454                 spin_unlock(&cache->lock);
 455
 456                 ret = load_free_space_cache(fs_info, cache);
 457
 458                 spin_lock(&cache->lock);
 459                 if (ret == 1) {
 460                         cache->cached = BTRFS_CACHE_FINISHED;
 461                         cache->last_byte_to_unpin = (u64)-1;
 462                 } else {
 463                         cache->cached = BTRFS_CACHE_NO;
 464                 }
 465                 spin_unlock(&cache->lock);
 466                 if (ret == 1)
 467                         return 0;
 468         }
 469
 470         if (load_cache_only)
 471                 return 0;
 472
 473         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
 474         BUG_ON(!caching_ctl);
 475
 476         INIT_LIST_HEAD(&caching_ctl->list);
 477         mutex_init(&caching_ctl->mutex);
 478         init_waitqueue_head(&caching_ctl->wait);
 479         caching_ctl->block_group = cache;
 480         caching_ctl->progress = cache->key.objectid;
 481         /* one for caching kthread, one for caching block group list */
 482         atomic_set(&caching_ctl->count, 2);
 483
 484         spin_lock(&cache->lock);
 485         if (cache->cached != BTRFS_CACHE_NO) {
 486                 spin_unlock(&cache->lock);
 487                 kfree(caching_ctl);
 488                 return 0;
 489         }
 490         cache->caching_ctl = caching_ctl;
 491         cache->cached = BTRFS_CACHE_STARTED;
 492         spin_unlock(&cache->lock);
 493
 494         down_write(&fs_info->extent_commit_sem);
 495         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 496         up_write(&fs_info->extent_commit_sem);
 497
 498         atomic_inc(&cache->space_info->caching_threads);
 499         btrfs_get_block_group(cache);
 500
 501         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 502                           cache->key.objectid);
 503         if (IS_ERR(tsk)) {
 504                 ret = PTR_ERR(tsk);
 505                 printk(KERN_ERR "error running thread %d\n", ret);
 506                 BUG();
 507         }
 508
 509         return ret;
 510 }
 511
 512 /*
 513  * return the block group that starts at or after bytenr
 514  */
 515 static struct btrfs_block_group_cache *
 516 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 517 {
 518         struct btrfs_block_group_cache *cache;
 519
 520         cache = block_group_cache_tree_search(info, bytenr, 0);
 521
 522         return cache;
 523 }
 524
 525 /*
 526  * return the block group that contains the given bytenr
 527  */
 528 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 529                                                  struct btrfs_fs_info *info,
 530                                                  u64 bytenr)
 531 {
 532         struct btrfs_block_group_cache *cache;
 533
 534         cache = block_group_cache_tree_search(info, bytenr, 1);
 535
 536         return cache;
 537 }
 538
 539 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 540                                                   u64 flags)
 541 {
 542         struct list_head *head = &info->space_info;
 543         struct btrfs_space_info *found;
 544
 545         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 546                  BTRFS_BLOCK_GROUP_METADATA;
 547
 548         rcu_read_lock();
 549         list_for_each_entry_rcu(found, head, list) {
 550                 if (found->flags & flags) {
 551                         rcu_read_unlock();
 552                         return found;
 553                 }
 554         }
 555         rcu_read_unlock();
 556         return NULL;
 557 }
 558
 559 /*
 560  * after adding space to the filesystem, we need to clear the full flags
 561  * on all the space infos.
 562  */
 563 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 564 {
 565         struct list_head *head = &info->space_info;
 566         struct btrfs_space_info *found;
 567
 568         rcu_read_lock();
 569         list_for_each_entry_rcu(found, head, list)
 570                 found->full = 0;
 571         rcu_read_unlock();
 572 }
 573
 574 static u64 div_factor(u64 num, int factor)
 575 {
 576         if (factor == 10)
 577                 return num;
 578         num *= factor;
 579         do_div(num, 10);
 580         return num;
 581 }
 582
 583 u64 btrfs_find_block_group(struct btrfs_root *root,
 584                            u64 search_start, u64 search_hint, int owner)
 585 {
 586         struct btrfs_block_group_cache *cache;
 587         u64 used;
 588         u64 last = max(search_hint, search_start);
 589         u64 group_start = 0;
 590         int full_search = 0;
 591         int factor = 9;
 592         int wrapped = 0;
 593 again:
 594         while (1) {
 595                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 596                 if (!cache)
 597                         break;
 598
 599                 spin_lock(&cache->lock);
 600                 last = cache->key.objectid + cache->key.offset;
 601                 used = btrfs_block_group_used(&cache->item);
 602
 603                 if ((full_search || !cache->ro) &&
 604                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 605                         if (used + cache->pinned + cache->reserved <
 606                             div_factor(cache->key.offset, factor)) {
 607                                 group_start = cache->key.objectid;
 608                                 spin_unlock(&cache->lock);
 609                                 btrfs_put_block_group(cache);
 610                                 goto found;
 611                         }
 612                 }
 613                 spin_unlock(&cache->lock);
 614                 btrfs_put_block_group(cache);
 615                 cond_resched();
 616         }
 617         if (!wrapped) {
 618                 last = search_start;
 619                 wrapped = 1;
 620                 goto again;
 621         }
 622         if (!full_search && factor < 10) {
 623                 last = search_start;
 624                 full_search = 1;
 625                 factor = 10;
 626                 goto again;
 627         }
 628 found:
 629         return group_start;
 630 }
 631
 632 /* simple helper to search for an existing extent at a given offset */
 633 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 634 {
 635         int ret;
 636         struct btrfs_key key;
 637         struct btrfs_path *path;
 638
 639         path = btrfs_alloc_path();
 640         BUG_ON(!path);
 641         key.objectid = start;
 642         key.offset = len;
 643         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 644         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 645                                 0, 0);
 646         btrfs_free_path(path);
 647         return ret;
 648 }
 649
 650 /*
 651  * helper function to lookup reference count and flags of extent.
 652  *
 653  * the head node for delayed ref is used to store the sum of all the
 654  * reference count modifications queued up in the rbtree. the head
 655  * node may also store the extent flags to set. This way you can check
 656  * to see what the reference count and extent flags would be if all of
 657  * the delayed refs are not processed.
 658  */
 659 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 660                              struct btrfs_root *root, u64 bytenr,
 661                              u64 num_bytes, u64 *refs, u64 *flags)
 662 {
 663         struct btrfs_delayed_ref_head *head;
 664         struct btrfs_delayed_ref_root *delayed_refs;
 665         struct btrfs_path *path;
 666         struct btrfs_extent_item *ei;
 667         struct extent_buffer *leaf;
 668         struct btrfs_key key;
 669         u32 item_size;
 670         u64 num_refs;
 671         u64 extent_flags;
 672         int ret;
 673
 674         path = btrfs_alloc_path();
 675         if (!path)
 676                 return -ENOMEM;
 677
 678         key.objectid = bytenr;
 679         key.type = BTRFS_EXTENT_ITEM_KEY;
 680         key.offset = num_bytes;
 681         if (!trans) {
 682                 path->skip_locking = 1;
 683                 path->search_commit_root = 1;
 684         }
 685 again:
 686         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 687                                 &key, path, 0, 0);
 688         if (ret < 0)
 689                 goto out_free;
 690
 691         if (ret == 0) {
 692                 leaf = path->nodes[0];
 693                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 694                 if (item_size >= sizeof(*ei)) {
 695                         ei = btrfs_item_ptr(leaf, path->slots[0],
 696                                             struct btrfs_extent_item);
 697                         num_refs = btrfs_extent_refs(leaf, ei);
 698                         extent_flags = btrfs_extent_flags(leaf, ei);
 699                 } else {
 700 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 701                         struct btrfs_extent_item_v0 *ei0;
 702                         BUG_ON(item_size != sizeof(*ei0));
 703                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 704                                              struct btrfs_extent_item_v0);
 705                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 706                         /* FIXME: this isn't correct for data */
 707                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 708 #else
 709                         BUG();
 710 #endif
 711                 }
 712                 BUG_ON(num_refs == 0);
 713         } else {
 714                 num_refs = 0;
 715                 extent_flags = 0;
 716                 ret = 0;
 717         }
 718
 719         if (!trans)
 720                 goto out;
 721
 722         delayed_refs = &trans->transaction->delayed_refs;
 723         spin_lock(&delayed_refs->lock);
 724         head = btrfs_find_delayed_ref_head(trans, bytenr);
 725         if (head) {
 726                 if (!mutex_trylock(&head->mutex)) {
 727                         atomic_inc(&head->node.refs);
 728                         spin_unlock(&delayed_refs->lock);
 729
 730                         btrfs_release_path(root->fs_info->extent_root, path);
 731
 732                         mutex_lock(&head->mutex);
 733                         mutex_unlock(&head->mutex);
 734                         btrfs_put_delayed_ref(&head->node);
 735                         goto again;
 736                 }
 737                 if (head->extent_op && head->extent_op->update_flags)
 738                         extent_flags |= head->extent_op->flags_to_set;
 739                 else
 740                         BUG_ON(num_refs == 0);
 741
 742                 num_refs += head->node.ref_mod;
 743                 mutex_unlock(&head->mutex);
 744         }
 745         spin_unlock(&delayed_refs->lock);
 746 out:
 747         WARN_ON(num_refs == 0);
 748         if (refs)
 749                 *refs = num_refs;
 750         if (flags)
 751                 *flags = extent_flags;
 752 out_free:
 753         btrfs_free_path(path);
 754         return ret;
 755 }
 756
 757 /*
 758  * Back reference rules.  Back refs have three main goals:
 759  *
 760  * 1) differentiate between all holders of references to an extent so that
 761  *    when a reference is dropped we can make sure it was a valid reference
 762  *    before freeing the extent.
 763  *
 764  * 2) Provide enough information to quickly find the holders of an extent
 765  *    if we notice a given block is corrupted or bad.
 766  *
 767  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 768  *    maintenance.  This is actually the same as #2, but with a slightly
 769  *    different use case.
 770  *
 771  * There are two kinds of back refs. The implicit back refs is optimized
 772  * for pointers in non-shared tree blocks. For a given pointer in a block,
 773  * back refs of this kind provide information about the block's owner tree
 774  * and the pointer's key. These information allow us to find the block by
 775  * b-tree searching. The full back refs is for pointers in tree blocks not
 776  * referenced by their owner trees. The location of tree block is recorded
 777  * in the back refs. Actually the full back refs is generic, and can be
 778  * used in all cases the implicit back refs is used. The major shortcoming
 779  * of the full back refs is its overhead. Every time a tree block gets
 780  * COWed, we have to update back refs entry for all pointers in it.
 781  *
 782  * For a newly allocated tree block, we use implicit back refs for
 783  * pointers in it. This means most tree related operations only involve
 784  * implicit back refs. For a tree block created in old transaction, the
 785  * only way to drop a reference to it is COW it. So we can detect the
 786  * event that tree block loses its owner tree's reference and do the
 787  * back refs conversion.
 788  *
 789  * When a tree block is COW'd through a tree, there are four cases:
 790  *
 791  * The reference count of the block is one and the tree is the block's
 792  * owner tree. Nothing to do in this case.
 793  *
 794  * The reference count of the block is one and the tree is not the
 795  * block's owner tree. In this case, full back refs is used for pointers
 796  * in the block. Remove these full back refs, add implicit back refs for
 797  * every pointers in the new block.
 798  *
 799  * The reference count of the block is greater than one and the tree is
 800  * the block's owner tree. In this case, implicit back refs is used for
 801  * pointers in the block. Add full back refs for every pointers in the
 802  * block, increase lower level extents' reference counts. The original
 803  * implicit back refs are entailed to the new block.
 804  *
 805  * The reference count of the block is greater than one and the tree is
 806  * not the block's owner tree. Add implicit back refs for every pointer in
 807  * the new block, increase lower level extents' reference count.
 808  *
 809  * Back Reference Key composing:
 810  *
 811  * The key objectid corresponds to the first byte in the extent,
 812  * The key type is used to differentiate between types of back refs.
 813  * There are different meanings of the key offset for different types
 814  * of back refs.
 815  *
 816  * File extents can be referenced by:
 817  *
 818  * - multiple snapshots, subvolumes, or different generations in one subvol
 819  * - different files inside a single subvolume
 820  * - different offsets inside a file (bookend extents in file.c)
 821  *
 822  * The extent ref structure for the implicit back refs has fields for:
 823  *
 824  * - Objectid of the subvolume root
 825  * - objectid of the file holding the reference
 826  * - original offset in the file
 827  * - how many bookend extents
 828  *
 829  * The key offset for the implicit back refs is hash of the first
 830  * three fields.
 831  *
 832  * The extent ref structure for the full back refs has field for:
 833  *
 834  * - number of pointers in the tree leaf
 835  *
 836  * The key offset for the implicit back refs is the first byte of
 837  * the tree leaf
 838  *
 839  * When a file extent is allocated, The implicit back refs is used.
 840  * the fields are filled in:
 841  *
 842  *     (root_key.objectid, inode objectid, offset in file, 1)
 843  *
 844  * When a file extent is removed file truncation, we find the
 845  * corresponding implicit back refs and check the following fields:
 846  *
 847  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 848  *
 849  * Btree extents can be referenced by:
 850  *
 851  * - Different subvolumes
 852  *
 853  * Both the implicit back refs and the full back refs for tree blocks
 854  * only consist of key. The key offset for the implicit back refs is
 855  * objectid of block's owner tree. The key offset for the full back refs
 856  * is the first byte of parent block.
 857  *
 858  * When implicit back refs is used, information about the lowest key and
 859  * level of the tree block are required. These information are stored in
 860  * tree block info structure.
 861  */
 862
 863 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 864 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 865                                   struct btrfs_root *root,
 866                                   struct btrfs_path *path,
 867                                   u64 owner, u32 extra_size)
 868 {
 869         struct btrfs_extent_item *item;
 870         struct btrfs_extent_item_v0 *ei0;
 871         struct btrfs_extent_ref_v0 *ref0;
 872         struct btrfs_tree_block_info *bi;
 873         struct extent_buffer *leaf;
 874         struct btrfs_key key;
 875         struct btrfs_key found_key;
 876         u32 new_size = sizeof(*item);
 877         u64 refs;
 878         int ret;
 879
 880         leaf = path->nodes[0];
 881         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 882
 883         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 884         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 885                              struct btrfs_extent_item_v0);
 886         refs = btrfs_extent_refs_v0(leaf, ei0);
 887
 888         if (owner == (u64)-1) {
 889                 while (1) {
 890                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 891                                 ret = btrfs_next_leaf(root, path);
 892                                 if (ret < 0)
 893                                         return ret;
 894                                 BUG_ON(ret > 0);
 895                                 leaf = path->nodes[0];
 896                         }
 897                         btrfs_item_key_to_cpu(leaf, &found_key,
 898                                               path->slots[0]);
 899                         BUG_ON(key.objectid != found_key.objectid);
 900                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 901                                 path->slots[0]++;
 902                                 continue;
 903                         }
 904                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 905                                               struct btrfs_extent_ref_v0);
 906                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 907                         break;
 908                 }
 909         }
 910         btrfs_release_path(root, path);
 911
 912         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 913                 new_size += sizeof(*bi);
 914
 915         new_size -= sizeof(*ei0);
 916         ret = btrfs_search_slot(trans, root, &key, path,
 917                                 new_size + extra_size, 1);
 918         if (ret < 0)
 919                 return ret;
 920         BUG_ON(ret);
 921
 922         ret = btrfs_extend_item(trans, root, path, new_size);
 923         BUG_ON(ret);
 924
 925         leaf = path->nodes[0];
 926         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 927         btrfs_set_extent_refs(leaf, item, refs);
 928         /* FIXME: get real generation */
 929         btrfs_set_extent_generation(leaf, item, 0);
 930         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 931                 btrfs_set_extent_flags(leaf, item,
 932                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 933                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 934                 bi = (struct btrfs_tree_block_info *)(item + 1);
 935                 /* FIXME: get first key of the block */
 936                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 937                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 938         } else {
 939                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 940         }
 941         btrfs_mark_buffer_dirty(leaf);
 942         return 0;
 943 }
 944 #endif
 945
 946 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 947 {
 948         u32 high_crc = ~(u32)0;
 949         u32 low_crc = ~(u32)0;
 950         __le64 lenum;
 951
 952         lenum = cpu_to_le64(root_objectid);
 953         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 954         lenum = cpu_to_le64(owner);
 955         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 956         lenum = cpu_to_le64(offset);
 957         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 958
 959         return ((u64)high_crc << 31) ^ (u64)low_crc;
 960 }
 961
 962 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 963                                      struct btrfs_extent_data_ref *ref)
 964 {
 965         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 966                                     btrfs_extent_data_ref_objectid(leaf, ref),
 967                                     btrfs_extent_data_ref_offset(leaf, ref));
 968 }
 969
 970 static int match_extent_data_ref(struct extent_buffer *leaf,
 971                                  struct btrfs_extent_data_ref *ref,
 972                                  u64 root_objectid, u64 owner, u64 offset)
 973 {
 974         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 975             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 976             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 977                 return 0;
 978         return 1;
 979 }
 980
 981 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 982                                            struct btrfs_root *root,
 983                                            struct btrfs_path *path,
 984                                            u64 bytenr, u64 parent,
 985                                            u64 root_objectid,
 986                                            u64 owner, u64 offset)
 987 {
 988         struct btrfs_key key;
 989         struct btrfs_extent_data_ref *ref;
 990         struct extent_buffer *leaf;
 991         u32 nritems;
 992         int ret;
 993         int recow;
 994         int err = -ENOENT;
 995
 996         key.objectid = bytenr;
 997         if (parent) {
 998                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 999                 key.offset = parent;
1000         } else {
1001                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1002                 key.offset = hash_extent_data_ref(root_objectid,
1003                                                   owner, offset);
1004         }
1005 again:
1006         recow = 0;
1007         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1008         if (ret < 0) {
1009                 err = ret;
1010                 goto fail;
1011         }
1012
1013         if (parent) {
1014                 if (!ret)
1015                         return 0;
1016 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1017                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1018                 btrfs_release_path(root, path);
1019                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1020                 if (ret < 0) {
1021                         err = ret;
1022                         goto fail;
1023                 }
1024                 if (!ret)
1025                         return 0;
1026 #endif
1027                 goto fail;
1028         }
1029
1030         leaf = path->nodes[0];
1031         nritems = btrfs_header_nritems(leaf);
1032         while (1) {
1033                 if (path->slots[0] >= nritems) {
1034                         ret = btrfs_next_leaf(root, path);
1035                         if (ret < 0)
1036                                 err = ret;
1037                         if (ret)
1038                                 goto fail;
1039
1040                         leaf = path->nodes[0];
1041                         nritems = btrfs_header_nritems(leaf);
1042                         recow = 1;
1043                 }
1044
1045                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1046                 if (key.objectid != bytenr ||
1047                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1048                         goto fail;
1049
1050                 ref = btrfs_item_ptr(leaf, path->slots[0],
1051                                      struct btrfs_extent_data_ref);
1052
1053                 if (match_extent_data_ref(leaf, ref, root_objectid,
1054                                           owner, offset)) {
1055                         if (recow) {
1056                                 btrfs_release_path(root, path);
1057                                 goto again;
1058                         }
1059                         err = 0;
1060                         break;
1061                 }
1062                 path->slots[0]++;
1063         }
1064 fail:
1065         return err;
1066 }
1067
1068 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1069                                            struct btrfs_root *root,
1070                                            struct btrfs_path *path,
1071                                            u64 bytenr, u64 parent,
1072                                            u64 root_objectid, u64 owner,
1073                                            u64 offset, int refs_to_add)
1074 {
1075         struct btrfs_key key;
1076         struct extent_buffer *leaf;
1077         u32 size;
1078         u32 num_refs;
1079         int ret;
1080
1081         key.objectid = bytenr;
1082         if (parent) {
1083                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1084                 key.offset = parent;
1085                 size = sizeof(struct btrfs_shared_data_ref);
1086         } else {
1087                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1088                 key.offset = hash_extent_data_ref(root_objectid,
1089                                                   owner, offset);
1090                 size = sizeof(struct btrfs_extent_data_ref);
1091         }
1092
1093         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1094         if (ret && ret != -EEXIST)
1095                 goto fail;
1096
1097         leaf = path->nodes[0];
1098         if (parent) {
1099                 struct btrfs_shared_data_ref *ref;
1100                 ref = btrfs_item_ptr(leaf, path->slots[0],
1101                                      struct btrfs_shared_data_ref);
1102                 if (ret == 0) {
1103                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1104                 } else {
1105                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1106                         num_refs += refs_to_add;
1107                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1108                 }
1109         } else {
1110                 struct btrfs_extent_data_ref *ref;
1111                 while (ret == -EEXIST) {
1112                         ref = btrfs_item_ptr(leaf, path->slots[0],
1113                                              struct btrfs_extent_data_ref);
1114                         if (match_extent_data_ref(leaf, ref, root_objectid,
1115                                                   owner, offset))
1116                                 break;
1117                         btrfs_release_path(root, path);
1118                         key.offset++;
1119                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1120                                                       size);
1121                         if (ret && ret != -EEXIST)
1122                                 goto fail;
1123
1124                         leaf = path->nodes[0];
1125                 }
1126                 ref = btrfs_item_ptr(leaf, path->slots[0],
1127                                      struct btrfs_extent_data_ref);
1128                 if (ret == 0) {
1129                         btrfs_set_extent_data_ref_root(leaf, ref,
1130                                                        root_objectid);
1131                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1132                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1133                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1134                 } else {
1135                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1136                         num_refs += refs_to_add;
1137                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1138                 }
1139         }
1140         btrfs_mark_buffer_dirty(leaf);
1141         ret = 0;
1142 fail:
1143         btrfs_release_path(root, path);
1144         return ret;
1145 }
1146
1147 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1148                                            struct btrfs_root *root,
1149                                            struct btrfs_path *path,
1150                                            int refs_to_drop)
1151 {
1152         struct btrfs_key key;
1153         struct btrfs_extent_data_ref *ref1 = NULL;
1154         struct btrfs_shared_data_ref *ref2 = NULL;
1155         struct extent_buffer *leaf;
1156         u32 num_refs = 0;
1157         int ret = 0;
1158
1159         leaf = path->nodes[0];
1160         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1161
1162         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1163                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1164                                       struct btrfs_extent_data_ref);
1165                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1166         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1167                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1168                                       struct btrfs_shared_data_ref);
1169                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1170 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1171         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1172                 struct btrfs_extent_ref_v0 *ref0;
1173                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1174                                       struct btrfs_extent_ref_v0);
1175                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1176 #endif
1177         } else {
1178                 BUG();
1179         }
1180
1181         BUG_ON(num_refs < refs_to_drop);
1182         num_refs -= refs_to_drop;
1183
1184         if (num_refs == 0) {
1185                 ret = btrfs_del_item(trans, root, path);
1186         } else {
1187                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1188                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1189                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1190                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1191 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1192                 else {
1193                         struct btrfs_extent_ref_v0 *ref0;
1194                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1195                                         struct btrfs_extent_ref_v0);
1196                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1197                 }
1198 #endif
1199                 btrfs_mark_buffer_dirty(leaf);
1200         }
1201         return ret;
1202 }
1203
1204 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1205                                           struct btrfs_path *path,
1206                                           struct btrfs_extent_inline_ref *iref)
1207 {
1208         struct btrfs_key key;
1209         struct extent_buffer *leaf;
1210         struct btrfs_extent_data_ref *ref1;
1211         struct btrfs_shared_data_ref *ref2;
1212         u32 num_refs = 0;
1213
1214         leaf = path->nodes[0];
1215         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1216         if (iref) {
1217                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1218                     BTRFS_EXTENT_DATA_REF_KEY) {
1219                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1220                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1221                 } else {
1222                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1223                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1224                 }
1225         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1226                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1227                                       struct btrfs_extent_data_ref);
1228                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1229         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1230                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1231                                       struct btrfs_shared_data_ref);
1232                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1233 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1234         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1235                 struct btrfs_extent_ref_v0 *ref0;
1236                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1237                                       struct btrfs_extent_ref_v0);
1238                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1239 #endif
1240         } else {
1241                 WARN_ON(1);
1242         }
1243         return num_refs;
1244 }
1245
1246 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1247                                           struct btrfs_root *root,
1248                                           struct btrfs_path *path,
1249                                           u64 bytenr, u64 parent,
1250                                           u64 root_objectid)
1251 {
1252         struct btrfs_key key;
1253         int ret;
1254
1255         key.objectid = bytenr;
1256         if (parent) {
1257                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1258                 key.offset = parent;
1259         } else {
1260                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1261                 key.offset = root_objectid;
1262         }
1263
1264         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1265         if (ret > 0)
1266                 ret = -ENOENT;
1267 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1268         if (ret == -ENOENT && parent) {
1269                 btrfs_release_path(root, path);
1270                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1271                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1272                 if (ret > 0)
1273                         ret = -ENOENT;
1274         }
1275 #endif
1276         return ret;
1277 }
1278
1279 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1280                                           struct btrfs_root *root,
1281                                           struct btrfs_path *path,
1282                                           u64 bytenr, u64 parent,
1283                                           u64 root_objectid)
1284 {
1285         struct btrfs_key key;
1286         int ret;
1287
1288         key.objectid = bytenr;
1289         if (parent) {
1290                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1291                 key.offset = parent;
1292         } else {
1293                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1294                 key.offset = root_objectid;
1295         }
1296
1297         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1298         btrfs_release_path(root, path);
1299         return ret;
1300 }
1301
1302 static inline int extent_ref_type(u64 parent, u64 owner)
1303 {
1304         int type;
1305         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1306                 if (parent > 0)
1307                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1308                 else
1309                         type = BTRFS_TREE_BLOCK_REF_KEY;
1310         } else {
1311                 if (parent > 0)
1312                         type = BTRFS_SHARED_DATA_REF_KEY;
1313                 else
1314                         type = BTRFS_EXTENT_DATA_REF_KEY;
1315         }
1316         return type;
1317 }
1318
1319 static int find_next_key(struct btrfs_path *path, int level,
1320                          struct btrfs_key *key)
1321
1322 {
1323         for (; level < BTRFS_MAX_LEVEL; level++) {
1324                 if (!path->nodes[level])
1325                         break;
1326                 if (path->slots[level] + 1 >=
1327                     btrfs_header_nritems(path->nodes[level]))
1328                         continue;
1329                 if (level == 0)
1330                         btrfs_item_key_to_cpu(path->nodes[level], key,
1331                                               path->slots[level] + 1);
1332                 else
1333                         btrfs_node_key_to_cpu(path->nodes[level], key,
1334                                               path->slots[level] + 1);
1335                 return 0;
1336         }
1337         return 1;
1338 }
1339
1340 /*
1341  * look for inline back ref. if back ref is found, *ref_ret is set
1342  * to the address of inline back ref, and 0 is returned.
1343  *
1344  * if back ref isn't found, *ref_ret is set to the address where it
1345  * should be inserted, and -ENOENT is returned.
1346  *
1347  * if insert is true and there are too many inline back refs, the path
1348  * points to the extent item, and -EAGAIN is returned.
1349  *
1350  * NOTE: inline back refs are ordered in the same way that back ref
1351  *       items in the tree are ordered.
1352  */
1353 static noinline_for_stack
1354 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1355                                  struct btrfs_root *root,
1356                                  struct btrfs_path *path,
1357                                  struct btrfs_extent_inline_ref **ref_ret,
1358                                  u64 bytenr, u64 num_bytes,
1359                                  u64 parent, u64 root_objectid,
1360                                  u64 owner, u64 offset, int insert)
1361 {
1362         struct btrfs_key key;
1363         struct extent_buffer *leaf;
1364         struct btrfs_extent_item *ei;
1365         struct btrfs_extent_inline_ref *iref;
1366         u64 flags;
1367         u64 item_size;
1368         unsigned long ptr;
1369         unsigned long end;
1370         int extra_size;
1371         int type;
1372         int want;
1373         int ret;
1374         int err = 0;
1375
1376         key.objectid = bytenr;
1377         key.type = BTRFS_EXTENT_ITEM_KEY;
1378         key.offset = num_bytes;
1379
1380         want = extent_ref_type(parent, owner);
1381         if (insert) {
1382                 extra_size = btrfs_extent_inline_ref_size(want);
1383                 path->keep_locks = 1;
1384         } else
1385                 extra_size = -1;
1386         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1387         if (ret < 0) {
1388                 err = ret;
1389                 goto out;
1390         }
1391         BUG_ON(ret);
1392
1393         leaf = path->nodes[0];
1394         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1395 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1396         if (item_size < sizeof(*ei)) {
1397                 if (!insert) {
1398                         err = -ENOENT;
1399                         goto out;
1400                 }
1401                 ret = convert_extent_item_v0(trans, root, path, owner,
1402                                              extra_size);
1403                 if (ret < 0) {
1404                         err = ret;
1405                         goto out;
1406                 }
1407                 leaf = path->nodes[0];
1408                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1409         }
1410 #endif
1411         BUG_ON(item_size < sizeof(*ei));
1412
1413         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1414         flags = btrfs_extent_flags(leaf, ei);
1415
1416         ptr = (unsigned long)(ei + 1);
1417         end = (unsigned long)ei + item_size;
1418
1419         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1420                 ptr += sizeof(struct btrfs_tree_block_info);
1421                 BUG_ON(ptr > end);
1422         } else {
1423                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1424         }
1425
1426         err = -ENOENT;
1427         while (1) {
1428                 if (ptr >= end) {
1429                         WARN_ON(ptr > end);
1430                         break;
1431                 }
1432                 iref = (struct btrfs_extent_inline_ref *)ptr;
1433                 type = btrfs_extent_inline_ref_type(leaf, iref);
1434                 if (want < type)
1435                         break;
1436                 if (want > type) {
1437                         ptr += btrfs_extent_inline_ref_size(type);
1438                         continue;
1439                 }
1440
1441                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1442                         struct btrfs_extent_data_ref *dref;
1443                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1444                         if (match_extent_data_ref(leaf, dref, root_objectid,
1445                                                   owner, offset)) {
1446                                 err = 0;
1447                                 break;
1448                         }
1449                         if (hash_extent_data_ref_item(leaf, dref) <
1450                             hash_extent_data_ref(root_objectid, owner, offset))
1451                                 break;
1452                 } else {
1453                         u64 ref_offset;
1454                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1455                         if (parent > 0) {
1456                                 if (parent == ref_offset) {
1457                                         err = 0;
1458                                         break;
1459                                 }
1460                                 if (ref_offset < parent)
1461                                         break;
1462                         } else {
1463                                 if (root_objectid == ref_offset) {
1464                                         err = 0;
1465                                         break;
1466                                 }
1467                                 if (ref_offset < root_objectid)
1468                                         break;
1469                         }
1470                 }
1471                 ptr += btrfs_extent_inline_ref_size(type);
1472         }
1473         if (err == -ENOENT && insert) {
1474                 if (item_size + extra_size >=
1475                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1476                         err = -EAGAIN;
1477                         goto out;
1478                 }
1479                 /*
1480                  * To add new inline back ref, we have to make sure
1481                  * there is no corresponding back ref item.
1482                  * For simplicity, we just do not add new inline back
1483                  * ref if there is any kind of item for this block
1484                  */
1485                 if (find_next_key(path, 0, &key) == 0 &&
1486                     key.objectid == bytenr &&
1487                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1488                         err = -EAGAIN;
1489                         goto out;
1490                 }
1491         }
1492         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1493 out:
1494         if (insert) {
1495                 path->keep_locks = 0;
1496                 btrfs_unlock_up_safe(path, 1);
1497         }
1498         return err;
1499 }
1500
1501 /*
1502  * helper to add new inline back ref
1503  */
1504 static noinline_for_stack
1505 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1506                                 struct btrfs_root *root,
1507                                 struct btrfs_path *path,
1508                                 struct btrfs_extent_inline_ref *iref,
1509                                 u64 parent, u64 root_objectid,
1510                                 u64 owner, u64 offset, int refs_to_add,
1511                                 struct btrfs_delayed_extent_op *extent_op)
1512 {
1513         struct extent_buffer *leaf;
1514         struct btrfs_extent_item *ei;
1515         unsigned long ptr;
1516         unsigned long end;
1517         unsigned long item_offset;
1518         u64 refs;
1519         int size;
1520         int type;
1521         int ret;
1522
1523         leaf = path->nodes[0];
1524         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1525         item_offset = (unsigned long)iref - (unsigned long)ei;
1526
1527         type = extent_ref_type(parent, owner);
1528         size = btrfs_extent_inline_ref_size(type);
1529
1530         ret = btrfs_extend_item(trans, root, path, size);
1531         BUG_ON(ret);
1532
1533         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1534         refs = btrfs_extent_refs(leaf, ei);
1535         refs += refs_to_add;
1536         btrfs_set_extent_refs(leaf, ei, refs);
1537         if (extent_op)
1538                 __run_delayed_extent_op(extent_op, leaf, ei);
1539
1540         ptr = (unsigned long)ei + item_offset;
1541         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1542         if (ptr < end - size)
1543                 memmove_extent_buffer(leaf, ptr + size, ptr,
1544                                       end - size - ptr);
1545
1546         iref = (struct btrfs_extent_inline_ref *)ptr;
1547         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1548         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1549                 struct btrfs_extent_data_ref *dref;
1550                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1551                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1552                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1553                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1554                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1555         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1556                 struct btrfs_shared_data_ref *sref;
1557                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1558                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1559                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1560         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1561                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1562         } else {
1563                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1564         }
1565         btrfs_mark_buffer_dirty(leaf);
1566         return 0;
1567 }
1568
1569 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1570                                  struct btrfs_root *root,
1571                                  struct btrfs_path *path,
1572                                  struct btrfs_extent_inline_ref **ref_ret,
1573                                  u64 bytenr, u64 num_bytes, u64 parent,
1574                                  u64 root_objectid, u64 owner, u64 offset)
1575 {
1576         int ret;
1577
1578         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1579                                            bytenr, num_bytes, parent,
1580                                            root_objectid, owner, offset, 0);
1581         if (ret != -ENOENT)
1582                 return ret;
1583
1584         btrfs_release_path(root, path);
1585         *ref_ret = NULL;
1586
1587         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1588                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1589                                             root_objectid);
1590         } else {
1591                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1592                                              root_objectid, owner, offset);
1593         }
1594         return ret;
1595 }
1596
1597 /*
1598  * helper to update/remove inline back ref
1599  */
1600 static noinline_for_stack
1601 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1602                                  struct btrfs_root *root,
1603                                  struct btrfs_path *path,
1604                                  struct btrfs_extent_inline_ref *iref,
1605                                  int refs_to_mod,
1606                                  struct btrfs_delayed_extent_op *extent_op)
1607 {
1608         struct extent_buffer *leaf;
1609         struct btrfs_extent_item *ei;
1610         struct btrfs_extent_data_ref *dref = NULL;
1611         struct btrfs_shared_data_ref *sref = NULL;
1612         unsigned long ptr;
1613         unsigned long end;
1614         u32 item_size;
1615         int size;
1616         int type;
1617         int ret;
1618         u64 refs;
1619
1620         leaf = path->nodes[0];
1621         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1622         refs = btrfs_extent_refs(leaf, ei);
1623         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1624         refs += refs_to_mod;
1625         btrfs_set_extent_refs(leaf, ei, refs);
1626         if (extent_op)
1627                 __run_delayed_extent_op(extent_op, leaf, ei);
1628
1629         type = btrfs_extent_inline_ref_type(leaf, iref);
1630
1631         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1632                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1633                 refs = btrfs_extent_data_ref_count(leaf, dref);
1634         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1635                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1636                 refs = btrfs_shared_data_ref_count(leaf, sref);
1637         } else {
1638                 refs = 1;
1639                 BUG_ON(refs_to_mod != -1);
1640         }
1641
1642         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1643         refs += refs_to_mod;
1644
1645         if (refs > 0) {
1646                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1647                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1648                 else
1649                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1650         } else {
1651                 size =  btrfs_extent_inline_ref_size(type);
1652                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1653                 ptr = (unsigned long)iref;
1654                 end = (unsigned long)ei + item_size;
1655                 if (ptr + size < end)
1656                         memmove_extent_buffer(leaf, ptr, ptr + size,
1657                                               end - ptr - size);
1658                 item_size -= size;
1659                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1660                 BUG_ON(ret);
1661         }
1662         btrfs_mark_buffer_dirty(leaf);
1663         return 0;
1664 }
1665
1666 static noinline_for_stack
1667 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1668                                  struct btrfs_root *root,
1669                                  struct btrfs_path *path,
1670                                  u64 bytenr, u64 num_bytes, u64 parent,
1671                                  u64 root_objectid, u64 owner,
1672                                  u64 offset, int refs_to_add,
1673                                  struct btrfs_delayed_extent_op *extent_op)
1674 {
1675         struct btrfs_extent_inline_ref *iref;
1676         int ret;
1677
1678         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1679                                            bytenr, num_bytes, parent,
1680                                            root_objectid, owner, offset, 1);
1681         if (ret == 0) {
1682                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1683                 ret = update_inline_extent_backref(trans, root, path, iref,
1684                                                    refs_to_add, extent_op);
1685         } else if (ret == -ENOENT) {
1686                 ret = setup_inline_extent_backref(trans, root, path, iref,
1687                                                   parent, root_objectid,
1688                                                   owner, offset, refs_to_add,
1689                                                   extent_op);
1690         }
1691         return ret;
1692 }
1693
1694 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1695                                  struct btrfs_root *root,
1696                                  struct btrfs_path *path,
1697                                  u64 bytenr, u64 parent, u64 root_objectid,
1698                                  u64 owner, u64 offset, int refs_to_add)
1699 {
1700         int ret;
1701         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1702                 BUG_ON(refs_to_add != 1);
1703                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1704                                             parent, root_objectid);
1705         } else {
1706                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1707                                              parent, root_objectid,
1708                                              owner, offset, refs_to_add);
1709         }
1710         return ret;
1711 }
1712
1713 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1714                                  struct btrfs_root *root,
1715                                  struct btrfs_path *path,
1716                                  struct btrfs_extent_inline_ref *iref,
1717                                  int refs_to_drop, int is_data)
1718 {
1719         int ret;
1720
1721         BUG_ON(!is_data && refs_to_drop != 1);
1722         if (iref) {
1723                 ret = update_inline_extent_backref(trans, root, path, iref,
1724                                                    -refs_to_drop, NULL);
1725         } else if (is_data) {
1726                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1727         } else {
1728                 ret = btrfs_del_item(trans, root, path);
1729         }
1730         return ret;
1731 }
1732
1733 static void btrfs_issue_discard(struct block_device *bdev,
1734                                 u64 start, u64 len)
1735 {
1736         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1737                         BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
1738 }
1739
1740 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1741                                 u64 num_bytes)
1742 {
1743         int ret;
1744         u64 map_length = num_bytes;
1745         struct btrfs_multi_bio *multi = NULL;
1746
1747         if (!btrfs_test_opt(root, DISCARD))
1748                 return 0;
1749
1750         /* Tell the block device(s) that the sectors can be discarded */
1751         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1752                               bytenr, &map_length, &multi, 0);
1753         if (!ret) {
1754                 struct btrfs_bio_stripe *stripe = multi->stripes;
1755                 int i;
1756
1757                 if (map_length > num_bytes)
1758                         map_length = num_bytes;
1759
1760                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1761                         btrfs_issue_discard(stripe->dev->bdev,
1762                                             stripe->physical,
1763                                             map_length);
1764                 }
1765                 kfree(multi);
1766         }
1767
1768         return ret;
1769 }
1770
1771 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1772                          struct btrfs_root *root,
1773                          u64 bytenr, u64 num_bytes, u64 parent,
1774                          u64 root_objectid, u64 owner, u64 offset)
1775 {
1776         int ret;
1777         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1778                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1779
1780         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1781                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1782                                         parent, root_objectid, (int)owner,
1783                                         BTRFS_ADD_DELAYED_REF, NULL);
1784         } else {
1785                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1786                                         parent, root_objectid, owner, offset,
1787                                         BTRFS_ADD_DELAYED_REF, NULL);
1788         }
1789         return ret;
1790 }
1791
1792 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1793                                   struct btrfs_root *root,
1794                                   u64 bytenr, u64 num_bytes,
1795                                   u64 parent, u64 root_objectid,
1796                                   u64 owner, u64 offset, int refs_to_add,
1797                                   struct btrfs_delayed_extent_op *extent_op)
1798 {
1799         struct btrfs_path *path;
1800         struct extent_buffer *leaf;
1801         struct btrfs_extent_item *item;
1802         u64 refs;
1803         int ret;
1804         int err = 0;
1805
1806         path = btrfs_alloc_path();
1807         if (!path)
1808                 return -ENOMEM;
1809
1810         path->reada = 1;
1811         path->leave_spinning = 1;
1812         /* this will setup the path even if it fails to insert the back ref */
1813         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1814                                            path, bytenr, num_bytes, parent,
1815                                            root_objectid, owner, offset,
1816                                            refs_to_add, extent_op);
1817         if (ret == 0)
1818                 goto out;
1819
1820         if (ret != -EAGAIN) {
1821                 err = ret;
1822                 goto out;
1823         }
1824
1825         leaf = path->nodes[0];
1826         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1827         refs = btrfs_extent_refs(leaf, item);
1828         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1829         if (extent_op)
1830                 __run_delayed_extent_op(extent_op, leaf, item);
1831
1832         btrfs_mark_buffer_dirty(leaf);
1833         btrfs_release_path(root->fs_info->extent_root, path);
1834
1835         path->reada = 1;
1836         path->leave_spinning = 1;
1837
1838         /* now insert the actual backref */
1839         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1840                                     path, bytenr, parent, root_objectid,
1841                                     owner, offset, refs_to_add);
1842         BUG_ON(ret);
1843 out:
1844         btrfs_free_path(path);
1845         return err;
1846 }
1847
1848 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1849                                 struct btrfs_root *root,
1850                                 struct btrfs_delayed_ref_node *node,
1851                                 struct btrfs_delayed_extent_op *extent_op,
1852                                 int insert_reserved)
1853 {
1854         int ret = 0;
1855         struct btrfs_delayed_data_ref *ref;
1856         struct btrfs_key ins;
1857         u64 parent = 0;
1858         u64 ref_root = 0;
1859         u64 flags = 0;
1860
1861         ins.objectid = node->bytenr;
1862         ins.offset = node->num_bytes;
1863         ins.type = BTRFS_EXTENT_ITEM_KEY;
1864
1865         ref = btrfs_delayed_node_to_data_ref(node);
1866         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1867                 parent = ref->parent;
1868         else
1869                 ref_root = ref->root;
1870
1871         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1872                 if (extent_op) {
1873                         BUG_ON(extent_op->update_key);
1874                         flags |= extent_op->flags_to_set;
1875                 }
1876                 ret = alloc_reserved_file_extent(trans, root,
1877                                                  parent, ref_root, flags,
1878                                                  ref->objectid, ref->offset,
1879                                                  &ins, node->ref_mod);
1880         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1881                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1882                                              node->num_bytes, parent,
1883                                              ref_root, ref->objectid,
1884                                              ref->offset, node->ref_mod,
1885                                              extent_op);
1886         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1887                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1888                                           node->num_bytes, parent,
1889                                           ref_root, ref->objectid,
1890                                           ref->offset, node->ref_mod,
1891                                           extent_op);
1892         } else {
1893                 BUG();
1894         }
1895         return ret;
1896 }
1897
1898 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1899                                     struct extent_buffer *leaf,
1900                                     struct btrfs_extent_item *ei)
1901 {
1902         u64 flags = btrfs_extent_flags(leaf, ei);
1903         if (extent_op->update_flags) {
1904                 flags |= extent_op->flags_to_set;
1905                 btrfs_set_extent_flags(leaf, ei, flags);
1906         }
1907
1908         if (extent_op->update_key) {
1909                 struct btrfs_tree_block_info *bi;
1910                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1911                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1912                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1913         }
1914 }
1915
1916 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1917                                  struct btrfs_root *root,
1918                                  struct btrfs_delayed_ref_node *node,
1919                                  struct btrfs_delayed_extent_op *extent_op)
1920 {
1921         struct btrfs_key key;
1922         struct btrfs_path *path;
1923         struct btrfs_extent_item *ei;
1924         struct extent_buffer *leaf;
1925         u32 item_size;
1926         int ret;
1927         int err = 0;
1928
1929         path = btrfs_alloc_path();
1930         if (!path)
1931                 return -ENOMEM;
1932
1933         key.objectid = node->bytenr;
1934         key.type = BTRFS_EXTENT_ITEM_KEY;
1935         key.offset = node->num_bytes;
1936
1937         path->reada = 1;
1938         path->leave_spinning = 1;
1939         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1940                                 path, 0, 1);
1941         if (ret < 0) {
1942                 err = ret;
1943                 goto out;
1944         }
1945         if (ret > 0) {
1946                 err = -EIO;
1947                 goto out;
1948         }
1949
1950         leaf = path->nodes[0];
1951         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1952 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1953         if (item_size < sizeof(*ei)) {
1954                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1955                                              path, (u64)-1, 0);
1956                 if (ret < 0) {
1957                         err = ret;
1958                         goto out;
1959                 }
1960                 leaf = path->nodes[0];
1961                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1962         }
1963 #endif
1964         BUG_ON(item_size < sizeof(*ei));
1965         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1966         __run_delayed_extent_op(extent_op, leaf, ei);
1967
1968         btrfs_mark_buffer_dirty(leaf);
1969 out:
1970         btrfs_free_path(path);
1971         return err;
1972 }
1973
1974 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1975                                 struct btrfs_root *root,
1976                                 struct btrfs_delayed_ref_node *node,
1977                                 struct btrfs_delayed_extent_op *extent_op,
1978                                 int insert_reserved)
1979 {
1980         int ret = 0;
1981         struct btrfs_delayed_tree_ref *ref;
1982         struct btrfs_key ins;
1983         u64 parent = 0;
1984         u64 ref_root = 0;
1985
1986         ins.objectid = node->bytenr;
1987         ins.offset = node->num_bytes;
1988         ins.type = BTRFS_EXTENT_ITEM_KEY;
1989
1990         ref = btrfs_delayed_node_to_tree_ref(node);
1991         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1992                 parent = ref->parent;
1993         else
1994                 ref_root = ref->root;
1995
1996         BUG_ON(node->ref_mod != 1);
1997         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1998                 BUG_ON(!extent_op || !extent_op->update_flags ||
1999                        !extent_op->update_key);
2000                 ret = alloc_reserved_tree_block(trans, root,
2001                                                 parent, ref_root,
2002                                                 extent_op->flags_to_set,
2003                                                 &extent_op->key,
2004                                                 ref->level, &ins);
2005         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2006                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2007                                              node->num_bytes, parent, ref_root,
2008                                              ref->level, 0, 1, extent_op);
2009         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2010                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2011                                           node->num_bytes, parent, ref_root,
2012                                           ref->level, 0, 1, extent_op);
2013         } else {
2014                 BUG();
2015         }
2016         return ret;
2017 }
2018
2019 /* helper function to actually process a single delayed ref entry */
2020 static int run_one_delayed_ref(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                                int insert_reserved)
2025 {
2026         int ret;
2027         if (btrfs_delayed_ref_is_head(node)) {
2028                 struct btrfs_delayed_ref_head *head;
2029                 /*
2030                  * we've hit the end of the chain and we were supposed
2031                  * to insert this extent into the tree.  But, it got
2032                  * deleted before we ever needed to insert it, so all
2033                  * we have to do is clean up the accounting
2034                  */
2035                 BUG_ON(extent_op);
2036                 head = btrfs_delayed_node_to_head(node);
2037                 if (insert_reserved) {
2038                         btrfs_pin_extent(root, node->bytenr,
2039                                          node->num_bytes, 1);
2040                         if (head->is_data) {
2041                                 ret = btrfs_del_csums(trans, root,
2042                                                       node->bytenr,
2043                                                       node->num_bytes);
2044                                 BUG_ON(ret);
2045                         }
2046                 }
2047                 mutex_unlock(&head->mutex);
2048                 return 0;
2049         }
2050
2051         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2052             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2053                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2054                                            insert_reserved);
2055         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2056                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2057                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2058                                            insert_reserved);
2059         else
2060                 BUG();
2061         return ret;
2062 }
2063
2064 static noinline struct btrfs_delayed_ref_node *
2065 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2066 {
2067         struct rb_node *node;
2068         struct btrfs_delayed_ref_node *ref;
2069         int action = BTRFS_ADD_DELAYED_REF;
2070 again:
2071         /*
2072          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2073          * this prevents ref count from going down to zero when
2074          * there still are pending delayed ref.
2075          */
2076         node = rb_prev(&head->node.rb_node);
2077         while (1) {
2078                 if (!node)
2079                         break;
2080                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2081                                 rb_node);
2082                 if (ref->bytenr != head->node.bytenr)
2083                         break;
2084                 if (ref->action == action)
2085                         return ref;
2086                 node = rb_prev(node);
2087         }
2088         if (action == BTRFS_ADD_DELAYED_REF) {
2089                 action = BTRFS_DROP_DELAYED_REF;
2090                 goto again;
2091         }
2092         return NULL;
2093 }
2094
2095 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2096                                        struct btrfs_root *root,
2097                                        struct list_head *cluster)
2098 {
2099         struct btrfs_delayed_ref_root *delayed_refs;
2100         struct btrfs_delayed_ref_node *ref;
2101         struct btrfs_delayed_ref_head *locked_ref = NULL;
2102         struct btrfs_delayed_extent_op *extent_op;
2103         int ret;
2104         int count = 0;
2105         int must_insert_reserved = 0;
2106
2107         delayed_refs = &trans->transaction->delayed_refs;
2108         while (1) {
2109                 if (!locked_ref) {
2110                         /* pick a new head ref from the cluster list */
2111                         if (list_empty(cluster))
2112                                 break;
2113
2114                         locked_ref = list_entry(cluster->next,
2115                                      struct btrfs_delayed_ref_head, cluster);
2116
2117                         /* grab the lock that says we are going to process
2118                          * all the refs for this head */
2119                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2120
2121                         /*
2122                          * we may have dropped the spin lock to get the head
2123                          * mutex lock, and that might have given someone else
2124                          * time to free the head.  If that's true, it has been
2125                          * removed from our list and we can move on.
2126                          */
2127                         if (ret == -EAGAIN) {
2128                                 locked_ref = NULL;
2129                                 count++;
2130                                 continue;
2131                         }
2132                 }
2133
2134                 /*
2135                  * record the must insert reserved flag before we
2136                  * drop the spin lock.
2137                  */
2138                 must_insert_reserved = locked_ref->must_insert_reserved;
2139                 locked_ref->must_insert_reserved = 0;
2140
2141                 extent_op = locked_ref->extent_op;
2142                 locked_ref->extent_op = NULL;
2143
2144                 /*
2145                  * locked_ref is the head node, so we have to go one
2146                  * node back for any delayed ref updates
2147                  */
2148                 ref = select_delayed_ref(locked_ref);
2149                 if (!ref) {
2150                         /* All delayed refs have been processed, Go ahead
2151                          * and send the head node to run_one_delayed_ref,
2152                          * so that any accounting fixes can happen
2153                          */
2154                         ref = &locked_ref->node;
2155
2156                         if (extent_op && must_insert_reserved) {
2157                                 kfree(extent_op);
2158                                 extent_op = NULL;
2159                         }
2160
2161                         if (extent_op) {
2162                                 spin_unlock(&delayed_refs->lock);
2163
2164                                 ret = run_delayed_extent_op(trans, root,
2165                                                             ref, extent_op);
2166                                 BUG_ON(ret);
2167                                 kfree(extent_op);
2168
2169                                 cond_resched();
2170                                 spin_lock(&delayed_refs->lock);
2171                                 continue;
2172                         }
2173
2174                         list_del_init(&locked_ref->cluster);
2175                         locked_ref = NULL;
2176                 }
2177
2178                 ref->in_tree = 0;
2179                 rb_erase(&ref->rb_node, &delayed_refs->root);
2180                 delayed_refs->num_entries--;
2181
2182                 spin_unlock(&delayed_refs->lock);
2183
2184                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2185                                           must_insert_reserved);
2186                 BUG_ON(ret);
2187
2188                 btrfs_put_delayed_ref(ref);
2189                 kfree(extent_op);
2190                 count++;
2191
2192                 cond_resched();
2193                 spin_lock(&delayed_refs->lock);
2194         }
2195         return count;
2196 }
2197
2198 /*
2199  * this starts processing the delayed reference count updates and
2200  * extent insertions we have queued up so far.  count can be
2201  * 0, which means to process everything in the tree at the start
2202  * of the run (but not newly added entries), or it can be some target
2203  * number you'd like to process.
2204  */
2205 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2206                            struct btrfs_root *root, unsigned long count)
2207 {
2208         struct rb_node *node;
2209         struct btrfs_delayed_ref_root *delayed_refs;
2210         struct btrfs_delayed_ref_node *ref;
2211         struct list_head cluster;
2212         int ret;
2213         int run_all = count == (unsigned long)-1;
2214         int run_most = 0;
2215
2216         if (root == root->fs_info->extent_root)
2217                 root = root->fs_info->tree_root;
2218
2219         delayed_refs = &trans->transaction->delayed_refs;
2220         INIT_LIST_HEAD(&cluster);
2221 again:
2222         spin_lock(&delayed_refs->lock);
2223         if (count == 0) {
2224                 count = delayed_refs->num_entries * 2;
2225                 run_most = 1;
2226         }
2227         while (1) {
2228                 if (!(run_all || run_most) &&
2229                     delayed_refs->num_heads_ready < 64)
2230                         break;
2231
2232                 /*
2233                  * go find something we can process in the rbtree.  We start at
2234                  * the beginning of the tree, and then build a cluster
2235                  * of refs to process starting at the first one we are able to
2236                  * lock
2237                  */
2238                 ret = btrfs_find_ref_cluster(trans, &cluster,
2239                                              delayed_refs->run_delayed_start);
2240                 if (ret)
2241                         break;
2242
2243                 ret = run_clustered_refs(trans, root, &cluster);
2244                 BUG_ON(ret < 0);
2245
2246                 count -= min_t(unsigned long, ret, count);
2247
2248                 if (count == 0)
2249                         break;
2250         }
2251
2252         if (run_all) {
2253                 node = rb_first(&delayed_refs->root);
2254                 if (!node)
2255                         goto out;
2256                 count = (unsigned long)-1;
2257
2258                 while (node) {
2259                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2260                                        rb_node);
2261                         if (btrfs_delayed_ref_is_head(ref)) {
2262                                 struct btrfs_delayed_ref_head *head;
2263
2264                                 head = btrfs_delayed_node_to_head(ref);
2265                                 atomic_inc(&ref->refs);
2266
2267                                 spin_unlock(&delayed_refs->lock);
2268                                 mutex_lock(&head->mutex);
2269                                 mutex_unlock(&head->mutex);
2270
2271                                 btrfs_put_delayed_ref(ref);
2272                                 cond_resched();
2273                                 goto again;
2274                         }
2275                         node = rb_next(node);
2276                 }
2277                 spin_unlock(&delayed_refs->lock);
2278                 schedule_timeout(1);
2279                 goto again;
2280         }
2281 out:
2282         spin_unlock(&delayed_refs->lock);
2283         return 0;
2284 }
2285
2286 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2287                                 struct btrfs_root *root,
2288                                 u64 bytenr, u64 num_bytes, u64 flags,
2289                                 int is_data)
2290 {
2291         struct btrfs_delayed_extent_op *extent_op;
2292         int ret;
2293
2294         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2295         if (!extent_op)
2296                 return -ENOMEM;
2297
2298         extent_op->flags_to_set = flags;
2299         extent_op->update_flags = 1;
2300         extent_op->update_key = 0;
2301         extent_op->is_data = is_data ? 1 : 0;
2302
2303         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2304         if (ret)
2305                 kfree(extent_op);
2306         return ret;
2307 }
2308
2309 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2310                                       struct btrfs_root *root,
2311                                       struct btrfs_path *path,
2312                                       u64 objectid, u64 offset, u64 bytenr)
2313 {
2314         struct btrfs_delayed_ref_head *head;
2315         struct btrfs_delayed_ref_node *ref;
2316         struct btrfs_delayed_data_ref *data_ref;
2317         struct btrfs_delayed_ref_root *delayed_refs;
2318         struct rb_node *node;
2319         int ret = 0;
2320
2321         ret = -ENOENT;
2322         delayed_refs = &trans->transaction->delayed_refs;
2323         spin_lock(&delayed_refs->lock);
2324         head = btrfs_find_delayed_ref_head(trans, bytenr);
2325         if (!head)
2326                 goto out;
2327
2328         if (!mutex_trylock(&head->mutex)) {
2329                 atomic_inc(&head->node.refs);
2330                 spin_unlock(&delayed_refs->lock);
2331
2332                 btrfs_release_path(root->fs_info->extent_root, path);
2333
2334                 mutex_lock(&head->mutex);
2335                 mutex_unlock(&head->mutex);
2336                 btrfs_put_delayed_ref(&head->node);
2337                 return -EAGAIN;
2338         }
2339
2340         node = rb_prev(&head->node.rb_node);
2341         if (!node)
2342                 goto out_unlock;
2343
2344         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2345
2346         if (ref->bytenr != bytenr)
2347                 goto out_unlock;
2348
2349         ret = 1;
2350         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2351                 goto out_unlock;
2352
2353         data_ref = btrfs_delayed_node_to_data_ref(ref);
2354
2355         node = rb_prev(node);
2356         if (node) {
2357                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2358                 if (ref->bytenr == bytenr)
2359                         goto out_unlock;
2360         }
2361
2362         if (data_ref->root != root->root_key.objectid ||
2363             data_ref->objectid != objectid || data_ref->offset != offset)
2364                 goto out_unlock;
2365
2366         ret = 0;
2367 out_unlock:
2368         mutex_unlock(&head->mutex);
2369 out:
2370         spin_unlock(&delayed_refs->lock);
2371         return ret;
2372 }
2373
2374 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2375                                         struct btrfs_root *root,
2376                                         struct btrfs_path *path,
2377                                         u64 objectid, u64 offset, u64 bytenr)
2378 {
2379         struct btrfs_root *extent_root = root->fs_info->extent_root;
2380         struct extent_buffer *leaf;
2381         struct btrfs_extent_data_ref *ref;
2382         struct btrfs_extent_inline_ref *iref;
2383         struct btrfs_extent_item *ei;
2384         struct btrfs_key key;
2385         u32 item_size;
2386         int ret;
2387
2388         key.objectid = bytenr;
2389         key.offset = (u64)-1;
2390         key.type = BTRFS_EXTENT_ITEM_KEY;
2391
2392         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2393         if (ret < 0)
2394                 goto out;
2395         BUG_ON(ret == 0);
2396
2397         ret = -ENOENT;
2398         if (path->slots[0] == 0)
2399                 goto out;
2400
2401         path->slots[0]--;
2402         leaf = path->nodes[0];
2403         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2404
2405         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2406                 goto out;
2407
2408         ret = 1;
2409         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2410 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2411         if (item_size < sizeof(*ei)) {
2412                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2413                 goto out;
2414         }
2415 #endif
2416         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2417
2418         if (item_size != sizeof(*ei) +
2419             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2420                 goto out;
2421
2422         if (btrfs_extent_generation(leaf, ei) <=
2423             btrfs_root_last_snapshot(&root->root_item))
2424                 goto out;
2425
2426         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2427         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2428             BTRFS_EXTENT_DATA_REF_KEY)
2429                 goto out;
2430
2431         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2432         if (btrfs_extent_refs(leaf, ei) !=
2433             btrfs_extent_data_ref_count(leaf, ref) ||
2434             btrfs_extent_data_ref_root(leaf, ref) !=
2435             root->root_key.objectid ||
2436             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2437             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2438                 goto out;
2439
2440         ret = 0;
2441 out:
2442         return ret;
2443 }
2444
2445 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2446                           struct btrfs_root *root,
2447                           u64 objectid, u64 offset, u64 bytenr)
2448 {
2449         struct btrfs_path *path;
2450         int ret;
2451         int ret2;
2452
2453         path = btrfs_alloc_path();
2454         if (!path)
2455                 return -ENOENT;
2456
2457         do {
2458                 ret = check_committed_ref(trans, root, path, objectid,
2459                                           offset, bytenr);
2460                 if (ret && ret != -ENOENT)
2461                         goto out;
2462
2463                 ret2 = check_delayed_ref(trans, root, path, objectid,
2464                                          offset, bytenr);
2465         } while (ret2 == -EAGAIN);
2466
2467         if (ret2 && ret2 != -ENOENT) {
2468                 ret = ret2;
2469                 goto out;
2470         }
2471
2472         if (ret != -ENOENT || ret2 != -ENOENT)
2473                 ret = 0;
2474 out:
2475         btrfs_free_path(path);
2476         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2477                 WARN_ON(ret > 0);
2478         return ret;
2479 }
2480
2481 #if 0
2482 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2483                     struct extent_buffer *buf, u32 nr_extents)
2484 {
2485         struct btrfs_key key;
2486         struct btrfs_file_extent_item *fi;
2487         u64 root_gen;
2488         u32 nritems;
2489         int i;
2490         int level;
2491         int ret = 0;
2492         int shared = 0;
2493
2494         if (!root->ref_cows)
2495                 return 0;
2496
2497         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2498                 shared = 0;
2499                 root_gen = root->root_key.offset;
2500         } else {
2501                 shared = 1;
2502                 root_gen = trans->transid - 1;
2503         }
2504
2505         level = btrfs_header_level(buf);
2506         nritems = btrfs_header_nritems(buf);
2507
2508         if (level == 0) {
2509                 struct btrfs_leaf_ref *ref;
2510                 struct btrfs_extent_info *info;
2511
2512                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2513                 if (!ref) {
2514                         ret = -ENOMEM;
2515                         goto out;
2516                 }
2517
2518                 ref->root_gen = root_gen;
2519                 ref->bytenr = buf->start;
2520                 ref->owner = btrfs_header_owner(buf);
2521                 ref->generation = btrfs_header_generation(buf);
2522                 ref->nritems = nr_extents;
2523                 info = ref->extents;
2524
2525                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2526                         u64 disk_bytenr;
2527                         btrfs_item_key_to_cpu(buf, &key, i);
2528                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2529                                 continue;
2530                         fi = btrfs_item_ptr(buf, i,
2531                                             struct btrfs_file_extent_item);
2532                         if (btrfs_file_extent_type(buf, fi) ==
2533                             BTRFS_FILE_EXTENT_INLINE)
2534                                 continue;
2535                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2536                         if (disk_bytenr == 0)
2537                                 continue;
2538
2539                         info->bytenr = disk_bytenr;
2540                         info->num_bytes =
2541                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2542                         info->objectid = key.objectid;
2543                         info->offset = key.offset;
2544                         info++;
2545                 }
2546
2547                 ret = btrfs_add_leaf_ref(root, ref, shared);
2548                 if (ret == -EEXIST && shared) {
2549                         struct btrfs_leaf_ref *old;
2550                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2551                         BUG_ON(!old);
2552                         btrfs_remove_leaf_ref(root, old);
2553                         btrfs_free_leaf_ref(root, old);
2554                         ret = btrfs_add_leaf_ref(root, ref, shared);
2555                 }
2556                 WARN_ON(ret);
2557                 btrfs_free_leaf_ref(root, ref);
2558         }
2559 out:
2560         return ret;
2561 }
2562
2563 /* when a block goes through cow, we update the reference counts of
2564  * everything that block points to.  The internal pointers of the block
2565  * can be in just about any order, and it is likely to have clusters of
2566  * things that are close together and clusters of things that are not.
2567  *
2568  * To help reduce the seeks that come with updating all of these reference
2569  * counts, sort them by byte number before actual updates are done.
2570  *
2571  * struct refsort is used to match byte number to slot in the btree block.
2572  * we sort based on the byte number and then use the slot to actually
2573  * find the item.
2574  *
2575  * struct refsort is smaller than strcut btrfs_item and smaller than
2576  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2577  * for a btree block, there's no way for a kmalloc of refsorts for a
2578  * single node to be bigger than a page.
2579  */
2580 struct refsort {
2581         u64 bytenr;
2582         u32 slot;
2583 };
2584
2585 /*
2586  * for passing into sort()
2587  */
2588 static int refsort_cmp(const void *a_void, const void *b_void)
2589 {
2590         const struct refsort *a = a_void;
2591         const struct refsort *b = b_void;
2592
2593         if (a->bytenr < b->bytenr)
2594                 return -1;
2595         if (a->bytenr > b->bytenr)
2596                 return 1;
2597         return 0;
2598 }
2599 #endif
2600
2601 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2602                            struct btrfs_root *root,
2603                            struct extent_buffer *buf,
2604                            int full_backref, int inc)
2605 {
2606         u64 bytenr;
2607         u64 num_bytes;
2608         u64 parent;
2609         u64 ref_root;
2610         u32 nritems;
2611         struct btrfs_key key;
2612         struct btrfs_file_extent_item *fi;
2613         int i;
2614         int level;
2615         int ret = 0;
2616         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2617                             u64, u64, u64, u64, u64, u64);
2618
2619         ref_root = btrfs_header_owner(buf);
2620         nritems = btrfs_header_nritems(buf);
2621         level = btrfs_header_level(buf);
2622
2623         if (!root->ref_cows && level == 0)
2624                 return 0;
2625
2626         if (inc)
2627                 process_func = btrfs_inc_extent_ref;
2628         else
2629                 process_func = btrfs_free_extent;
2630
2631         if (full_backref)
2632                 parent = buf->start;
2633         else
2634                 parent = 0;
2635
2636         for (i = 0; i < nritems; i++) {
2637                 if (level == 0) {
2638                         btrfs_item_key_to_cpu(buf, &key, i);
2639                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2640                                 continue;
2641                         fi = btrfs_item_ptr(buf, i,
2642                                             struct btrfs_file_extent_item);
2643                         if (btrfs_file_extent_type(buf, fi) ==
2644                             BTRFS_FILE_EXTENT_INLINE)
2645                                 continue;
2646                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2647                         if (bytenr == 0)
2648                                 continue;
2649
2650                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2651                         key.offset -= btrfs_file_extent_offset(buf, fi);
2652                         ret = process_func(trans, root, bytenr, num_bytes,
2653                                            parent, ref_root, key.objectid,
2654                                            key.offset);
2655                         if (ret)
2656                                 goto fail;
2657                 } else {
2658                         bytenr = btrfs_node_blockptr(buf, i);
2659                         num_bytes = btrfs_level_size(root, level - 1);
2660                         ret = process_func(trans, root, bytenr, num_bytes,
2661                                            parent, ref_root, level - 1, 0);
2662                         if (ret)
2663                                 goto fail;
2664                 }
2665         }
2666         return 0;
2667 fail:
2668         BUG();
2669         return ret;
2670 }
2671
2672 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2673                   struct extent_buffer *buf, int full_backref)
2674 {
2675         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2676 }
2677
2678 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2679                   struct extent_buffer *buf, int full_backref)
2680 {
2681         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2682 }
2683
2684 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2685                                  struct btrfs_root *root,
2686                                  struct btrfs_path *path,
2687                                  struct btrfs_block_group_cache *cache)
2688 {
2689         int ret;
2690         struct btrfs_root *extent_root = root->fs_info->extent_root;
2691         unsigned long bi;
2692         struct extent_buffer *leaf;
2693
2694         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2695         if (ret < 0)
2696                 goto fail;
2697         BUG_ON(ret);
2698
2699         leaf = path->nodes[0];
2700         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2701         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2702         btrfs_mark_buffer_dirty(leaf);
2703         btrfs_release_path(extent_root, path);
2704 fail:
2705         if (ret)
2706                 return ret;
2707         return 0;
2708
2709 }
2710
2711 static struct btrfs_block_group_cache *
2712 next_block_group(struct btrfs_root *root,
2713                  struct btrfs_block_group_cache *cache)
2714 {
2715         struct rb_node *node;
2716         spin_lock(&root->fs_info->block_group_cache_lock);
2717         node = rb_next(&cache->cache_node);
2718         btrfs_put_block_group(cache);
2719         if (node) {
2720                 cache = rb_entry(node, struct btrfs_block_group_cache,
2721                                  cache_node);
2722                 btrfs_get_block_group(cache);
2723         } else
2724                 cache = NULL;
2725         spin_unlock(&root->fs_info->block_group_cache_lock);
2726         return cache;
2727 }
2728
2729 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2730                             struct btrfs_trans_handle *trans,
2731                             struct btrfs_path *path)
2732 {
2733         struct btrfs_root *root = block_group->fs_info->tree_root;
2734         struct inode *inode = NULL;
2735         u64 alloc_hint = 0;
2736         int num_pages = 0;
2737         int retries = 0;
2738         int ret = 0;
2739
2740         /*
2741          * If this block group is smaller than 100 megs don't bother caching the
2742          * block group.
2743          */
2744         if (block_group->key.offset < (100 * 1024 * 1024)) {
2745                 spin_lock(&block_group->lock);
2746                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2747                 spin_unlock(&block_group->lock);
2748                 return 0;
2749         }
2750
2751 again:
2752         inode = lookup_free_space_inode(root, block_group, path);
2753         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2754                 ret = PTR_ERR(inode);
2755                 btrfs_release_path(root, path);
2756                 goto out;
2757         }
2758
2759         if (IS_ERR(inode)) {
2760                 BUG_ON(retries);
2761                 retries++;
2762
2763                 if (block_group->ro)
2764                         goto out_free;
2765
2766                 ret = create_free_space_inode(root, trans, block_group, path);
2767                 if (ret)
2768                         goto out_free;
2769                 goto again;
2770         }
2771
2772         /*
2773          * We want to set the generation to 0, that way if anything goes wrong
2774          * from here on out we know not to trust this cache when we load up next
2775          * time.
2776          */
2777         BTRFS_I(inode)->generation = 0;
2778         ret = btrfs_update_inode(trans, root, inode);
2779         WARN_ON(ret);
2780
2781         if (i_size_read(inode) > 0) {
2782                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2783                                                       inode);
2784                 if (ret)
2785                         goto out_put;
2786         }
2787
2788         spin_lock(&block_group->lock);
2789         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2790                 spin_unlock(&block_group->lock);
2791                 goto out_put;
2792         }
2793         spin_unlock(&block_group->lock);
2794
2795         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2796         if (!num_pages)
2797                 num_pages = 1;
2798
2799         /*
2800          * Just to make absolutely sure we have enough space, we're going to
2801          * preallocate 12 pages worth of space for each block group.  In
2802          * practice we ought to use at most 8, but we need extra space so we can
2803          * add our header and have a terminator between the extents and the
2804          * bitmaps.
2805          */
2806         num_pages *= 16;
2807         num_pages *= PAGE_CACHE_SIZE;
2808
2809         ret = btrfs_check_data_free_space(inode, num_pages);
2810         if (ret)
2811                 goto out_put;
2812
2813         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2814                                               num_pages, num_pages,
2815                                               &alloc_hint);
2816         btrfs_free_reserved_data_space(inode, num_pages);
2817 out_put:
2818         iput(inode);
2819 out_free:
2820         btrfs_release_path(root, path);
2821 out:
2822         spin_lock(&block_group->lock);
2823         if (ret)
2824                 block_group->disk_cache_state = BTRFS_DC_ERROR;
2825         else
2826                 block_group->disk_cache_state = BTRFS_DC_SETUP;
2827         spin_unlock(&block_group->lock);
2828
2829         return ret;
2830 }
2831
2832 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2833                                    struct btrfs_root *root)
2834 {
2835         struct btrfs_block_group_cache *cache;
2836         int err = 0;
2837         struct btrfs_path *path;
2838         u64 last = 0;
2839
2840         path = btrfs_alloc_path();
2841         if (!path)
2842                 return -ENOMEM;
2843
2844 again:
2845         while (1) {
2846                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2847                 while (cache) {
2848                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2849                                 break;
2850                         cache = next_block_group(root, cache);
2851                 }
2852                 if (!cache) {
2853                         if (last == 0)
2854                                 break;
2855                         last = 0;
2856                         continue;
2857                 }
2858                 err = cache_save_setup(cache, trans, path);
2859                 last = cache->key.objectid + cache->key.offset;
2860                 btrfs_put_block_group(cache);
2861         }
2862
2863         while (1) {
2864                 if (last == 0) {
2865                         err = btrfs_run_delayed_refs(trans, root,
2866                                                      (unsigned long)-1);
2867                         BUG_ON(err);
2868                 }
2869
2870                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2871                 while (cache) {
2872                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2873                                 btrfs_put_block_group(cache);
2874                                 goto again;
2875                         }
2876
2877                         if (cache->dirty)
2878                                 break;
2879                         cache = next_block_group(root, cache);
2880                 }
2881                 if (!cache) {
2882                         if (last == 0)
2883                                 break;
2884                         last = 0;
2885                         continue;
2886                 }
2887
2888                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2889                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2890                 cache->dirty = 0;
2891                 last = cache->key.objectid + cache->key.offset;
2892
2893                 err = write_one_cache_group(trans, root, path, cache);
2894                 BUG_ON(err);
2895                 btrfs_put_block_group(cache);
2896         }
2897
2898         while (1) {
2899                 /*
2900                  * I don't think this is needed since we're just marking our
2901                  * preallocated extent as written, but just in case it can't
2902                  * hurt.
2903                  */
2904                 if (last == 0) {
2905                         err = btrfs_run_delayed_refs(trans, root,
2906                                                      (unsigned long)-1);
2907                         BUG_ON(err);
2908                 }
2909
2910                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2911                 while (cache) {
2912                         /*
2913                          * Really this shouldn't happen, but it could if we
2914                          * couldn't write the entire preallocated extent and
2915                          * splitting the extent resulted in a new block.
2916                          */
2917                         if (cache->dirty) {
2918                                 btrfs_put_block_group(cache);
2919                                 goto again;
2920                         }
2921                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2922                                 break;
2923                         cache = next_block_group(root, cache);
2924                 }
2925                 if (!cache) {
2926                         if (last == 0)
2927                                 break;
2928                         last = 0;
2929                         continue;
2930                 }
2931
2932                 btrfs_write_out_cache(root, trans, cache, path);
2933
2934                 /*
2935                  * If we didn't have an error then the cache state is still
2936                  * NEED_WRITE, so we can set it to WRITTEN.
2937                  */
2938                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2939                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2940                 last = cache->key.objectid + cache->key.offset;
2941                 btrfs_put_block_group(cache);
2942         }
2943
2944         btrfs_free_path(path);
2945         return 0;
2946 }
2947
2948 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2949 {
2950         struct btrfs_block_group_cache *block_group;
2951         int readonly = 0;
2952
2953         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2954         if (!block_group || block_group->ro)
2955                 readonly = 1;
2956         if (block_group)
2957                 btrfs_put_block_group(block_group);
2958         return readonly;
2959 }
2960
2961 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2962                              u64 total_bytes, u64 bytes_used,
2963                              struct btrfs_space_info **space_info)
2964 {
2965         struct btrfs_space_info *found;
2966         int i;
2967         int factor;
2968
2969         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2970                      BTRFS_BLOCK_GROUP_RAID10))
2971                 factor = 2;
2972         else
2973                 factor = 1;
2974
2975         found = __find_space_info(info, flags);
2976         if (found) {
2977                 spin_lock(&found->lock);
2978                 found->total_bytes += total_bytes;
2979                 found->disk_total += total_bytes * factor;
2980                 found->bytes_used += bytes_used;
2981                 found->disk_used += bytes_used * factor;
2982                 found->full = 0;
2983                 spin_unlock(&found->lock);
2984                 *space_info = found;
2985                 return 0;
2986         }
2987         found = kzalloc(sizeof(*found), GFP_NOFS);
2988         if (!found)
2989                 return -ENOMEM;
2990
2991         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2992                 INIT_LIST_HEAD(&found->block_groups[i]);
2993         init_rwsem(&found->groups_sem);
2994         spin_lock_init(&found->lock);
2995         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2996                                 BTRFS_BLOCK_GROUP_SYSTEM |
2997                                 BTRFS_BLOCK_GROUP_METADATA);
2998         found->total_bytes = total_bytes;
2999         found->disk_total = total_bytes * factor;
3000         found->bytes_used = bytes_used;
3001         found->disk_used = bytes_used * factor;
3002         found->bytes_pinned = 0;
3003         found->bytes_reserved = 0;
3004         found->bytes_readonly = 0;
3005         found->bytes_may_use = 0;
3006         found->full = 0;
3007         found->force_alloc = 0;
3008         *space_info = found;
3009         list_add_rcu(&found->list, &info->space_info);
3010         atomic_set(&found->caching_threads, 0);
3011         return 0;
3012 }
3013
3014 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3015 {
3016         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3017                                    BTRFS_BLOCK_GROUP_RAID1 |
3018                                    BTRFS_BLOCK_GROUP_RAID10 |
3019                                    BTRFS_BLOCK_GROUP_DUP);
3020         if (extra_flags) {
3021                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3022                         fs_info->avail_data_alloc_bits |= extra_flags;
3023                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3024                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3025                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3026                         fs_info->avail_system_alloc_bits |= extra_flags;
3027         }
3028 }
3029
3030 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3031 {
3032         u64 num_devices = root->fs_info->fs_devices->rw_devices;
3033
3034         if (num_devices == 1)
3035                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3036         if (num_devices < 4)
3037                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3038
3039         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3040             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3041                       BTRFS_BLOCK_GROUP_RAID10))) {
3042                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3043         }
3044
3045         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3046             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3047                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3048         }
3049
3050         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3051             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3052              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3053              (flags & BTRFS_BLOCK_GROUP_DUP)))
3054                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3055         return flags;
3056 }
3057
3058 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3059 {
3060         if (flags & BTRFS_BLOCK_GROUP_DATA)
3061                 flags |= root->fs_info->avail_data_alloc_bits &
3062                          root->fs_info->data_alloc_profile;
3063         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3064                 flags |= root->fs_info->avail_system_alloc_bits &
3065                          root->fs_info->system_alloc_profile;
3066         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3067                 flags |= root->fs_info->avail_metadata_alloc_bits &
3068                          root->fs_info->metadata_alloc_profile;
3069         return btrfs_reduce_alloc_profile(root, flags);
3070 }
3071
3072 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3073 {
3074         u64 flags;
3075
3076         if (data)
3077                 flags = BTRFS_BLOCK_GROUP_DATA;
3078         else if (root == root->fs_info->chunk_root)
3079                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3080         else
3081                 flags = BTRFS_BLOCK_GROUP_METADATA;
3082
3083         return get_alloc_profile(root, flags);
3084 }
3085
3086 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3087 {
3088         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3089                                                        BTRFS_BLOCK_GROUP_DATA);
3090 }
3091
3092 /*
3093  * This will check the space that the inode allocates from to make sure we have
3094  * enough space for bytes.
3095  */
3096 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3097 {
3098         struct btrfs_space_info *data_sinfo;
3099         struct btrfs_root *root = BTRFS_I(inode)->root;
3100         u64 used;
3101         int ret = 0, committed = 0, alloc_chunk = 1;
3102
3103         /* make sure bytes are sectorsize aligned */
3104         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3105
3106         if (root == root->fs_info->tree_root) {
3107                 alloc_chunk = 0;
3108                 committed = 1;
3109         }
3110
3111         data_sinfo = BTRFS_I(inode)->space_info;
3112         if (!data_sinfo)
3113                 goto alloc;
3114
3115 again:
3116         /* make sure we have enough space to handle the data first */
3117         spin_lock(&data_sinfo->lock);
3118         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3119                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3120                 data_sinfo->bytes_may_use;
3121
3122         if (used + bytes > data_sinfo->total_bytes) {
3123                 struct btrfs_trans_handle *trans;
3124
3125                 /*
3126                  * if we don't have enough free bytes in this space then we need
3127                  * to alloc a new chunk.
3128                  */
3129                 if (!data_sinfo->full && alloc_chunk) {
3130                         u64 alloc_target;
3131
3132                         data_sinfo->force_alloc = 1;
3133                         spin_unlock(&data_sinfo->lock);
3134 alloc:
3135                         alloc_target = btrfs_get_alloc_profile(root, 1);
3136                         trans = btrfs_join_transaction(root, 1);
3137                         if (IS_ERR(trans))
3138                                 return PTR_ERR(trans);
3139
3140                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3141                                              bytes + 2 * 1024 * 1024,
3142                                              alloc_target, 0);
3143                         btrfs_end_transaction(trans, root);
3144                         if (ret < 0)
3145                                 return ret;
3146
3147                         if (!data_sinfo) {
3148                                 btrfs_set_inode_space_info(root, inode);
3149                                 data_sinfo = BTRFS_I(inode)->space_info;
3150                         }
3151                         goto again;
3152                 }
3153                 spin_unlock(&data_sinfo->lock);
3154
3155                 /* commit the current transaction and try again */
3156                 if (!committed && !root->fs_info->open_ioctl_trans) {
3157                         committed = 1;
3158                         trans = btrfs_join_transaction(root, 1);
3159                         if (IS_ERR(trans))
3160                                 return PTR_ERR(trans);
3161                         ret = btrfs_commit_transaction(trans, root);
3162                         if (ret)
3163                                 return ret;
3164                         goto again;
3165                 }
3166
3167 #if 0 /* I hope we never need this code again, just in case */
3168                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3169                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3170                        "%llu bytes_readonly, %llu may use %llu total\n",
3171                        (unsigned long long)bytes,
3172                        (unsigned long long)data_sinfo->bytes_used,
3173                        (unsigned long long)data_sinfo->bytes_reserved,
3174                        (unsigned long long)data_sinfo->bytes_pinned,
3175                        (unsigned long long)data_sinfo->bytes_readonly,
3176                        (unsigned long long)data_sinfo->bytes_may_use,
3177                        (unsigned long long)data_sinfo->total_bytes);
3178 #endif
3179                 return -ENOSPC;
3180         }
3181         data_sinfo->bytes_may_use += bytes;
3182         BTRFS_I(inode)->reserved_bytes += bytes;
3183         spin_unlock(&data_sinfo->lock);
3184
3185         return 0;
3186 }
3187
3188 /*
3189  * called when we are clearing an delalloc extent from the
3190  * inode's io_tree or there was an error for whatever reason
3191  * after calling btrfs_check_data_free_space
3192  */
3193 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3194 {
3195         struct btrfs_root *root = BTRFS_I(inode)->root;
3196         struct btrfs_space_info *data_sinfo;
3197
3198         /* make sure bytes are sectorsize aligned */
3199         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3200
3201         data_sinfo = BTRFS_I(inode)->space_info;
3202         spin_lock(&data_sinfo->lock);
3203         data_sinfo->bytes_may_use -= bytes;
3204         BTRFS_I(inode)->reserved_bytes -= bytes;
3205         spin_unlock(&data_sinfo->lock);
3206 }
3207
3208 static void force_metadata_allocation(struct btrfs_fs_info *info)
3209 {
3210         struct list_head *head = &info->space_info;
3211         struct btrfs_space_info *found;
3212
3213         rcu_read_lock();
3214         list_for_each_entry_rcu(found, head, list) {
3215                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3216                         found->force_alloc = 1;
3217         }
3218         rcu_read_unlock();
3219 }
3220
3221 static int should_alloc_chunk(struct btrfs_space_info *sinfo, u64 alloc_bytes)
3222 {
3223         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3224
3225         if (sinfo->bytes_used + sinfo->bytes_reserved +
3226             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3227                 return 0;
3228
3229         if (sinfo->bytes_used + sinfo->bytes_reserved +
3230             alloc_bytes < div_factor(num_bytes, 8))
3231                 return 0;
3232
3233         if (num_bytes > 256 * 1024 * 1024 &&
3234             sinfo->bytes_used < div_factor(num_bytes, 3))
3235                 return 0;
3236
3237         return 1;
3238 }
3239
3240 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3241                           struct btrfs_root *extent_root, u64 alloc_bytes,
3242                           u64 flags, int force)
3243 {
3244         struct btrfs_space_info *space_info;
3245         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3246         int ret = 0;
3247
3248         mutex_lock(&fs_info->chunk_mutex);
3249
3250         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3251
3252         space_info = __find_space_info(extent_root->fs_info, flags);
3253         if (!space_info) {
3254                 ret = update_space_info(extent_root->fs_info, flags,
3255                                         0, 0, &space_info);
3256                 BUG_ON(ret);
3257         }
3258         BUG_ON(!space_info);
3259
3260         spin_lock(&space_info->lock);
3261         if (space_info->force_alloc)
3262                 force = 1;
3263         if (space_info->full) {
3264                 spin_unlock(&space_info->lock);
3265                 goto out;
3266         }
3267
3268         if (!force && !should_alloc_chunk(space_info, alloc_bytes)) {
3269                 spin_unlock(&space_info->lock);
3270                 goto out;
3271         }
3272         spin_unlock(&space_info->lock);
3273
3274         /*
3275          * If we have mixed data/metadata chunks we want to make sure we keep
3276          * allocating mixed chunks instead of individual chunks.
3277          */
3278         if (btrfs_mixed_space_info(space_info))
3279                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3280
3281         /*
3282          * if we're doing a data chunk, go ahead and make sure that
3283          * we keep a reasonable number of metadata chunks allocated in the
3284          * FS as well.
3285          */
3286         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3287                 fs_info->data_chunk_allocations++;
3288                 if (!(fs_info->data_chunk_allocations %
3289                       fs_info->metadata_ratio))
3290                         force_metadata_allocation(fs_info);
3291         }
3292
3293         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3294         spin_lock(&space_info->lock);
3295         if (ret)
3296                 space_info->full = 1;
3297         else
3298                 ret = 1;
3299         space_info->force_alloc = 0;
3300         spin_unlock(&space_info->lock);
3301 out:
3302         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3303         return ret;
3304 }
3305
3306 /*
3307  * shrink metadata reservation for delalloc
3308  */
3309 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3310                            struct btrfs_root *root, u64 to_reclaim, int sync)
3311 {
3312         struct btrfs_block_rsv *block_rsv;
3313         struct btrfs_space_info *space_info;
3314         u64 reserved;
3315         u64 max_reclaim;
3316         u64 reclaimed = 0;
3317         int no_reclaim = 0;
3318         int pause = 1;
3319         int ret;
3320
3321         block_rsv = &root->fs_info->delalloc_block_rsv;
3322         space_info = block_rsv->space_info;
3323         spin_lock(&space_info->lock);
3324         reserved = space_info->bytes_reserved;
3325         spin_unlock(&space_info->lock);
3326
3327         if (reserved == 0)
3328                 return 0;
3329
3330         max_reclaim = min(reserved, to_reclaim);
3331
3332         while (1) {
3333                 ret = btrfs_start_one_delalloc_inode(root, trans ? 1 : 0, sync);
3334                 if (!ret) {
3335                         if (no_reclaim > 2)
3336                                 break;
3337                         no_reclaim++;
3338                         __set_current_state(TASK_INTERRUPTIBLE);
3339                         schedule_timeout(pause);
3340                         pause <<= 1;
3341                         if (pause > HZ / 10)
3342                                 pause = HZ / 10;
3343                 } else {
3344                         no_reclaim = 0;
3345                         pause = 1;
3346                 }
3347
3348                 spin_lock(&space_info->lock);
3349                 if (reserved > space_info->bytes_reserved)
3350                         reclaimed += reserved - space_info->bytes_reserved;
3351                 reserved = space_info->bytes_reserved;
3352                 spin_unlock(&space_info->lock);
3353
3354                 if (reserved == 0 || reclaimed >= max_reclaim)
3355                         break;
3356
3357                 if (trans && trans->transaction->blocked)
3358                         return -EAGAIN;
3359         }
3360         return reclaimed >= to_reclaim;
3361 }
3362
3363 /*
3364  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3365  * idea is if this is the first call (retries == 0) then we will add to our
3366  * reserved count if we can't make the allocation in order to hold our place
3367  * while we go and try and free up space.  That way for retries > 1 we don't try
3368  * and add space, we just check to see if the amount of unused space is >= the
3369  * total space, meaning that our reservation is valid.
3370  *
3371  * However if we don't intend to retry this reservation, pass -1 as retries so
3372  * that it short circuits this logic.
3373  */
3374 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3375                                   struct btrfs_root *root,
3376                                   struct btrfs_block_rsv *block_rsv,
3377                                   u64 orig_bytes, int flush)
3378 {
3379         struct btrfs_space_info *space_info = block_rsv->space_info;
3380         u64 unused;
3381         u64 num_bytes = orig_bytes;
3382         int retries = 0;
3383         int ret = 0;
3384         bool reserved = false;
3385         bool committed = false;
3386
3387 again:
3388         ret = -ENOSPC;
3389         if (reserved)
3390                 num_bytes = 0;
3391
3392         spin_lock(&space_info->lock);
3393         unused = space_info->bytes_used + space_info->bytes_reserved +
3394                  space_info->bytes_pinned + space_info->bytes_readonly +
3395                  space_info->bytes_may_use;
3396
3397         /*
3398          * The idea here is that we've not already over-reserved the block group
3399          * then we can go ahead and save our reservation first and then start
3400          * flushing if we need to.  Otherwise if we've already overcommitted
3401          * lets start flushing stuff first and then come back and try to make
3402          * our reservation.
3403          */
3404         if (unused <= space_info->total_bytes) {
3405                 unused -= space_info->total_bytes;
3406                 if (unused >= num_bytes) {
3407                         if (!reserved)
3408                                 space_info->bytes_reserved += orig_bytes;
3409                         ret = 0;
3410                 } else {
3411                         /*
3412                          * Ok set num_bytes to orig_bytes since we aren't
3413                          * overocmmitted, this way we only try and reclaim what
3414                          * we need.
3415                          */
3416                         num_bytes = orig_bytes;
3417                 }
3418         } else {
3419                 /*
3420                  * Ok we're over committed, set num_bytes to the overcommitted
3421                  * amount plus the amount of bytes that we need for this
3422                  * reservation.
3423                  */
3424                 num_bytes = unused - space_info->total_bytes +
3425                         (orig_bytes * (retries + 1));
3426         }
3427
3428         /*
3429          * Couldn't make our reservation, save our place so while we're trying
3430          * to reclaim space we can actually use it instead of somebody else
3431          * stealing it from us.
3432          */
3433         if (ret && !reserved) {
3434                 space_info->bytes_reserved += orig_bytes;
3435                 reserved = true;
3436         }
3437
3438         spin_unlock(&space_info->lock);
3439
3440         if (!ret)
3441                 return 0;
3442
3443         if (!flush)
3444                 goto out;
3445
3446         /*
3447          * We do synchronous shrinking since we don't actually unreserve
3448          * metadata until after the IO is completed.
3449          */
3450         ret = shrink_delalloc(trans, root, num_bytes, 1);
3451         if (ret > 0)
3452                 return 0;
3453         else if (ret < 0)
3454                 goto out;
3455
3456         /*
3457          * So if we were overcommitted it's possible that somebody else flushed
3458          * out enough space and we simply didn't have enough space to reclaim,
3459          * so go back around and try again.
3460          */
3461         if (retries < 2) {
3462                 retries++;
3463                 goto again;
3464         }
3465
3466         spin_lock(&space_info->lock);
3467         /*
3468          * Not enough space to be reclaimed, don't bother committing the
3469          * transaction.
3470          */
3471         if (space_info->bytes_pinned < orig_bytes)
3472                 ret = -ENOSPC;
3473         spin_unlock(&space_info->lock);
3474         if (ret)
3475                 goto out;
3476
3477         ret = -EAGAIN;
3478         if (trans || committed)
3479                 goto out;
3480
3481         ret = -ENOSPC;
3482         trans = btrfs_join_transaction(root, 1);
3483         if (IS_ERR(trans))
3484                 goto out;
3485         ret = btrfs_commit_transaction(trans, root);
3486         if (!ret) {
3487                 trans = NULL;
3488                 committed = true;
3489                 goto again;
3490         }
3491
3492 out:
3493         if (reserved) {
3494                 spin_lock(&space_info->lock);
3495                 space_info->bytes_reserved -= orig_bytes;
3496                 spin_unlock(&space_info->lock);
3497         }
3498
3499         return ret;
3500 }
3501
3502 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3503                                              struct btrfs_root *root)
3504 {
3505         struct btrfs_block_rsv *block_rsv;
3506         if (root->ref_cows)
3507                 block_rsv = trans->block_rsv;
3508         else
3509                 block_rsv = root->block_rsv;
3510
3511         if (!block_rsv)
3512                 block_rsv = &root->fs_info->empty_block_rsv;
3513
3514         return block_rsv;
3515 }
3516
3517 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3518                                u64 num_bytes)
3519 {
3520         int ret = -ENOSPC;
3521         spin_lock(&block_rsv->lock);
3522         if (block_rsv->reserved >= num_bytes) {
3523                 block_rsv->reserved -= num_bytes;
3524                 if (block_rsv->reserved < block_rsv->size)
3525                         block_rsv->full = 0;
3526                 ret = 0;
3527         }
3528         spin_unlock(&block_rsv->lock);
3529         return ret;
3530 }
3531
3532 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3533                                 u64 num_bytes, int update_size)
3534 {
3535         spin_lock(&block_rsv->lock);
3536         block_rsv->reserved += num_bytes;
3537         if (update_size)
3538                 block_rsv->size += num_bytes;
3539         else if (block_rsv->reserved >= block_rsv->size)
3540                 block_rsv->full = 1;
3541         spin_unlock(&block_rsv->lock);
3542 }
3543
3544 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3545                              struct btrfs_block_rsv *dest, u64 num_bytes)
3546 {
3547         struct btrfs_space_info *space_info = block_rsv->space_info;
3548
3549         spin_lock(&block_rsv->lock);
3550         if (num_bytes == (u64)-1)
3551                 num_bytes = block_rsv->size;
3552         block_rsv->size -= num_bytes;
3553         if (block_rsv->reserved >= block_rsv->size) {
3554                 num_bytes = block_rsv->reserved - block_rsv->size;
3555                 block_rsv->reserved = block_rsv->size;
3556                 block_rsv->full = 1;
3557         } else {
3558                 num_bytes = 0;
3559         }
3560         spin_unlock(&block_rsv->lock);
3561
3562         if (num_bytes > 0) {
3563                 if (dest) {
3564                         block_rsv_add_bytes(dest, num_bytes, 0);
3565                 } else {
3566                         spin_lock(&space_info->lock);
3567                         space_info->bytes_reserved -= num_bytes;
3568                         spin_unlock(&space_info->lock);
3569                 }
3570         }
3571 }
3572
3573 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3574                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3575 {
3576         int ret;
3577
3578         ret = block_rsv_use_bytes(src, num_bytes);
3579         if (ret)
3580                 return ret;
3581
3582         block_rsv_add_bytes(dst, num_bytes, 1);
3583         return 0;
3584 }
3585
3586 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3587 {
3588         memset(rsv, 0, sizeof(*rsv));
3589         spin_lock_init(&rsv->lock);
3590         atomic_set(&rsv->usage, 1);
3591         rsv->priority = 6;
3592         INIT_LIST_HEAD(&rsv->list);
3593 }
3594
3595 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3596 {
3597         struct btrfs_block_rsv *block_rsv;
3598         struct btrfs_fs_info *fs_info = root->fs_info;
3599         u64 alloc_target;
3600
3601         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3602         if (!block_rsv)
3603                 return NULL;
3604
3605         btrfs_init_block_rsv(block_rsv);
3606
3607         alloc_target = btrfs_get_alloc_profile(root, 0);
3608         block_rsv->space_info = __find_space_info(fs_info,
3609                                                   BTRFS_BLOCK_GROUP_METADATA);
3610
3611         return block_rsv;
3612 }
3613
3614 void btrfs_free_block_rsv(struct btrfs_root *root,
3615                           struct btrfs_block_rsv *rsv)
3616 {
3617         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3618                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3619                 if (!rsv->durable)
3620                         kfree(rsv);
3621         }
3622 }
3623
3624 /*
3625  * make the block_rsv struct be able to capture freed space.
3626  * the captured space will re-add to the the block_rsv struct
3627  * after transaction commit
3628  */
3629 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3630                                  struct btrfs_block_rsv *block_rsv)
3631 {
3632         block_rsv->durable = 1;
3633         mutex_lock(&fs_info->durable_block_rsv_mutex);
3634         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3635         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3636 }
3637
3638 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3639                         struct btrfs_root *root,
3640                         struct btrfs_block_rsv *block_rsv,
3641                         u64 num_bytes)
3642 {
3643         int ret;
3644
3645         if (num_bytes == 0)
3646                 return 0;
3647
3648         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3649         if (!ret) {
3650                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3651                 return 0;
3652         }
3653
3654         return ret;
3655 }
3656
3657 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3658                           struct btrfs_root *root,
3659                           struct btrfs_block_rsv *block_rsv,
3660                           u64 min_reserved, int min_factor)
3661 {
3662         u64 num_bytes = 0;
3663         int commit_trans = 0;
3664         int ret = -ENOSPC;
3665
3666         if (!block_rsv)
3667                 return 0;
3668
3669         spin_lock(&block_rsv->lock);
3670         if (min_factor > 0)
3671                 num_bytes = div_factor(block_rsv->size, min_factor);
3672         if (min_reserved > num_bytes)
3673                 num_bytes = min_reserved;
3674
3675         if (block_rsv->reserved >= num_bytes) {
3676                 ret = 0;
3677         } else {
3678                 num_bytes -= block_rsv->reserved;
3679                 if (block_rsv->durable &&
3680                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3681                         commit_trans = 1;
3682         }
3683         spin_unlock(&block_rsv->lock);
3684         if (!ret)
3685                 return 0;
3686
3687         if (block_rsv->refill_used) {
3688                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3689                                              num_bytes, 0);
3690                 if (!ret) {
3691                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3692                         return 0;
3693                 }
3694         }
3695
3696         if (commit_trans) {
3697                 if (trans)
3698                         return -EAGAIN;
3699
3700                 trans = btrfs_join_transaction(root, 1);
3701                 BUG_ON(IS_ERR(trans));
3702                 ret = btrfs_commit_transaction(trans, root);
3703                 return 0;
3704         }
3705
3706         WARN_ON(1);
3707         printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
3708                 block_rsv->size, block_rsv->reserved,
3709                 block_rsv->freed[0], block_rsv->freed[1]);
3710
3711         return -ENOSPC;
3712 }
3713
3714 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3715                             struct btrfs_block_rsv *dst_rsv,
3716                             u64 num_bytes)
3717 {
3718         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3719 }
3720
3721 void btrfs_block_rsv_release(struct btrfs_root *root,
3722                              struct btrfs_block_rsv *block_rsv,
3723                              u64 num_bytes)
3724 {
3725         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3726         if (global_rsv->full || global_rsv == block_rsv ||
3727             block_rsv->space_info != global_rsv->space_info)
3728                 global_rsv = NULL;
3729         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3730 }
3731
3732 /*
3733  * helper to calculate size of global block reservation.
3734  * the desired value is sum of space used by extent tree,
3735  * checksum tree and root tree
3736  */
3737 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3738 {
3739         struct btrfs_space_info *sinfo;
3740         u64 num_bytes;
3741         u64 meta_used;
3742         u64 data_used;
3743         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3744 #if 0
3745         /*
3746          * per tree used space accounting can be inaccuracy, so we
3747          * can't rely on it.
3748          */
3749         spin_lock(&fs_info->extent_root->accounting_lock);
3750         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3751         spin_unlock(&fs_info->extent_root->accounting_lock);
3752
3753         spin_lock(&fs_info->csum_root->accounting_lock);
3754         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3755         spin_unlock(&fs_info->csum_root->accounting_lock);
3756
3757         spin_lock(&fs_info->tree_root->accounting_lock);
3758         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3759         spin_unlock(&fs_info->tree_root->accounting_lock);
3760 #endif
3761         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3762         spin_lock(&sinfo->lock);
3763         data_used = sinfo->bytes_used;
3764         spin_unlock(&sinfo->lock);
3765
3766         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3767         spin_lock(&sinfo->lock);
3768         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3769                 data_used = 0;
3770         meta_used = sinfo->bytes_used;
3771         spin_unlock(&sinfo->lock);
3772
3773         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3774                     csum_size * 2;
3775         num_bytes += div64_u64(data_used + meta_used, 50);
3776
3777         if (num_bytes * 3 > meta_used)
3778                 num_bytes = div64_u64(meta_used, 3);
3779
3780         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3781 }
3782
3783 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3784 {
3785         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3786         struct btrfs_space_info *sinfo = block_rsv->space_info;
3787         u64 num_bytes;
3788
3789         num_bytes = calc_global_metadata_size(fs_info);
3790
3791         spin_lock(&block_rsv->lock);
3792         spin_lock(&sinfo->lock);
3793
3794         block_rsv->size = num_bytes;
3795
3796         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3797                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3798                     sinfo->bytes_may_use;
3799
3800         if (sinfo->total_bytes > num_bytes) {
3801                 num_bytes = sinfo->total_bytes - num_bytes;
3802                 block_rsv->reserved += num_bytes;
3803                 sinfo->bytes_reserved += num_bytes;
3804         }
3805
3806         if (block_rsv->reserved >= block_rsv->size) {
3807                 num_bytes = block_rsv->reserved - block_rsv->size;
3808                 sinfo->bytes_reserved -= num_bytes;
3809                 block_rsv->reserved = block_rsv->size;
3810                 block_rsv->full = 1;
3811         }
3812 #if 0
3813         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3814                 block_rsv->size, block_rsv->reserved);
3815 #endif
3816         spin_unlock(&sinfo->lock);
3817         spin_unlock(&block_rsv->lock);
3818 }
3819
3820 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3821 {
3822         struct btrfs_space_info *space_info;
3823
3824         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3825         fs_info->chunk_block_rsv.space_info = space_info;
3826         fs_info->chunk_block_rsv.priority = 10;
3827
3828         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3829         fs_info->global_block_rsv.space_info = space_info;
3830         fs_info->global_block_rsv.priority = 10;
3831         fs_info->global_block_rsv.refill_used = 1;
3832         fs_info->delalloc_block_rsv.space_info = space_info;
3833         fs_info->trans_block_rsv.space_info = space_info;
3834         fs_info->empty_block_rsv.space_info = space_info;
3835         fs_info->empty_block_rsv.priority = 10;
3836
3837         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3838         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3839         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3840         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3841         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3842
3843         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3844
3845         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3846
3847         update_global_block_rsv(fs_info);
3848 }
3849
3850 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3851 {
3852         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3853         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3854         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3855         WARN_ON(fs_info->trans_block_rsv.size > 0);
3856         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3857         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3858         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3859 }
3860
3861 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3862 {
3863         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3864                 3 * num_items;
3865 }
3866
3867 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3868                                  struct btrfs_root *root,
3869                                  int num_items)
3870 {
3871         u64 num_bytes;
3872         int ret;
3873
3874         if (num_items == 0 || root->fs_info->chunk_root == root)
3875                 return 0;
3876
3877         num_bytes = calc_trans_metadata_size(root, num_items);
3878         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3879                                   num_bytes);
3880         if (!ret) {
3881                 trans->bytes_reserved += num_bytes;
3882                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3883         }
3884         return ret;
3885 }
3886
3887 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3888                                   struct btrfs_root *root)
3889 {
3890         if (!trans->bytes_reserved)
3891                 return;
3892
3893         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3894         btrfs_block_rsv_release(root, trans->block_rsv,
3895                                 trans->bytes_reserved);
3896         trans->bytes_reserved = 0;
3897 }
3898
3899 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3900                                   struct inode *inode)
3901 {
3902         struct btrfs_root *root = BTRFS_I(inode)->root;
3903         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3904         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3905
3906         /*
3907          * one for deleting orphan item, one for updating inode and
3908          * two for calling btrfs_truncate_inode_items.
3909          *
3910          * btrfs_truncate_inode_items is a delete operation, it frees
3911          * more space than it uses in most cases. So two units of
3912          * metadata space should be enough for calling it many times.
3913          * If all of the metadata space is used, we can commit
3914          * transaction and use space it freed.
3915          */
3916         u64 num_bytes = calc_trans_metadata_size(root, 4);
3917         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3918 }
3919
3920 void btrfs_orphan_release_metadata(struct inode *inode)
3921 {
3922         struct btrfs_root *root = BTRFS_I(inode)->root;
3923         u64 num_bytes = calc_trans_metadata_size(root, 4);
3924         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3925 }
3926
3927 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3928                                 struct btrfs_pending_snapshot *pending)
3929 {
3930         struct btrfs_root *root = pending->root;
3931         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3932         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3933         /*
3934          * two for root back/forward refs, two for directory entries
3935          * and one for root of the snapshot.
3936          */
3937         u64 num_bytes = calc_trans_metadata_size(root, 5);
3938         dst_rsv->space_info = src_rsv->space_info;
3939         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3940 }
3941
3942 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3943 {
3944         return num_bytes >>= 3;
3945 }
3946
3947 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3948 {
3949         struct btrfs_root *root = BTRFS_I(inode)->root;
3950         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3951         u64 to_reserve;
3952         int nr_extents;
3953         int ret;
3954
3955         if (btrfs_transaction_in_commit(root->fs_info))
3956                 schedule_timeout(1);
3957
3958         num_bytes = ALIGN(num_bytes, root->sectorsize);
3959
3960         spin_lock(&BTRFS_I(inode)->accounting_lock);
3961         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
3962         if (nr_extents > BTRFS_I(inode)->reserved_extents) {
3963                 nr_extents -= BTRFS_I(inode)->reserved_extents;
3964                 to_reserve = calc_trans_metadata_size(root, nr_extents);
3965         } else {
3966                 nr_extents = 0;
3967                 to_reserve = 0;
3968         }
3969         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3970
3971         to_reserve += calc_csum_metadata_size(inode, num_bytes);
3972         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
3973         if (ret)
3974                 return ret;
3975
3976         spin_lock(&BTRFS_I(inode)->accounting_lock);
3977         BTRFS_I(inode)->reserved_extents += nr_extents;
3978         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
3979         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3980
3981         block_rsv_add_bytes(block_rsv, to_reserve, 1);
3982
3983         if (block_rsv->size > 512 * 1024 * 1024)
3984                 shrink_delalloc(NULL, root, to_reserve, 0);
3985
3986         return 0;
3987 }
3988
3989 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
3990 {
3991         struct btrfs_root *root = BTRFS_I(inode)->root;
3992         u64 to_free;
3993         int nr_extents;
3994
3995         num_bytes = ALIGN(num_bytes, root->sectorsize);
3996         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
3997
3998         spin_lock(&BTRFS_I(inode)->accounting_lock);
3999         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4000         if (nr_extents < BTRFS_I(inode)->reserved_extents) {
4001                 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
4002                 BTRFS_I(inode)->reserved_extents -= nr_extents;
4003         } else {
4004                 nr_extents = 0;
4005         }
4006         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4007
4008         to_free = calc_csum_metadata_size(inode, num_bytes);
4009         if (nr_extents > 0)
4010                 to_free += calc_trans_metadata_size(root, nr_extents);
4011
4012         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4013                                 to_free);
4014 }
4015
4016 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4017 {
4018         int ret;
4019
4020         ret = btrfs_check_data_free_space(inode, num_bytes);
4021         if (ret)
4022                 return ret;
4023
4024         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4025         if (ret) {
4026                 btrfs_free_reserved_data_space(inode, num_bytes);
4027                 return ret;
4028         }
4029
4030         return 0;
4031 }
4032
4033 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4034 {
4035         btrfs_delalloc_release_metadata(inode, num_bytes);
4036         btrfs_free_reserved_data_space(inode, num_bytes);
4037 }
4038
4039 static int update_block_group(struct btrfs_trans_handle *trans,
4040                               struct btrfs_root *root,
4041                               u64 bytenr, u64 num_bytes, int alloc)
4042 {
4043         struct btrfs_block_group_cache *cache = NULL;
4044         struct btrfs_fs_info *info = root->fs_info;
4045         u64 total = num_bytes;
4046         u64 old_val;
4047         u64 byte_in_group;
4048         int factor;
4049
4050         /* block accounting for super block */
4051         spin_lock(&info->delalloc_lock);
4052         old_val = btrfs_super_bytes_used(&info->super_copy);
4053         if (alloc)
4054                 old_val += num_bytes;
4055         else
4056                 old_val -= num_bytes;
4057         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4058         spin_unlock(&info->delalloc_lock);
4059
4060         while (total) {
4061                 cache = btrfs_lookup_block_group(info, bytenr);
4062                 if (!cache)
4063                         return -1;
4064                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4065                                     BTRFS_BLOCK_GROUP_RAID1 |
4066                                     BTRFS_BLOCK_GROUP_RAID10))
4067                         factor = 2;
4068                 else
4069                         factor = 1;
4070                 /*
4071                  * If this block group has free space cache written out, we
4072                  * need to make sure to load it if we are removing space.  This
4073                  * is because we need the unpinning stage to actually add the
4074                  * space back to the block group, otherwise we will leak space.
4075                  */
4076                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4077                         cache_block_group(cache, trans, 1);
4078
4079                 byte_in_group = bytenr - cache->key.objectid;
4080                 WARN_ON(byte_in_group > cache->key.offset);
4081
4082                 spin_lock(&cache->space_info->lock);
4083                 spin_lock(&cache->lock);
4084
4085                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4086                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4087                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4088
4089                 cache->dirty = 1;
4090                 old_val = btrfs_block_group_used(&cache->item);
4091                 num_bytes = min(total, cache->key.offset - byte_in_group);
4092                 if (alloc) {
4093                         old_val += num_bytes;
4094                         btrfs_set_block_group_used(&cache->item, old_val);
4095                         cache->reserved -= num_bytes;
4096                         cache->space_info->bytes_reserved -= num_bytes;
4097                         cache->space_info->bytes_used += num_bytes;
4098                         cache->space_info->disk_used += num_bytes * factor;
4099                         spin_unlock(&cache->lock);
4100                         spin_unlock(&cache->space_info->lock);
4101                 } else {
4102                         old_val -= num_bytes;
4103                         btrfs_set_block_group_used(&cache->item, old_val);
4104                         cache->pinned += num_bytes;
4105                         cache->space_info->bytes_pinned += num_bytes;
4106                         cache->space_info->bytes_used -= num_bytes;
4107                         cache->space_info->disk_used -= num_bytes * factor;
4108                         spin_unlock(&cache->lock);
4109                         spin_unlock(&cache->space_info->lock);
4110
4111                         set_extent_dirty(info->pinned_extents,
4112                                          bytenr, bytenr + num_bytes - 1,
4113                                          GFP_NOFS | __GFP_NOFAIL);
4114                 }
4115                 btrfs_put_block_group(cache);
4116                 total -= num_bytes;
4117                 bytenr += num_bytes;
4118         }
4119         return 0;
4120 }
4121
4122 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4123 {
4124         struct btrfs_block_group_cache *cache;
4125         u64 bytenr;
4126
4127         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4128         if (!cache)
4129                 return 0;
4130
4131         bytenr = cache->key.objectid;
4132         btrfs_put_block_group(cache);
4133
4134         return bytenr;
4135 }
4136
4137 static int pin_down_extent(struct btrfs_root *root,
4138                            struct btrfs_block_group_cache *cache,
4139                            u64 bytenr, u64 num_bytes, int reserved)
4140 {
4141         spin_lock(&cache->space_info->lock);
4142         spin_lock(&cache->lock);
4143         cache->pinned += num_bytes;
4144         cache->space_info->bytes_pinned += num_bytes;
4145         if (reserved) {
4146                 cache->reserved -= num_bytes;
4147                 cache->space_info->bytes_reserved -= num_bytes;
4148         }
4149         spin_unlock(&cache->lock);
4150         spin_unlock(&cache->space_info->lock);
4151
4152         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4153                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4154         return 0;
4155 }
4156
4157 /*
4158  * this function must be called within transaction
4159  */
4160 int btrfs_pin_extent(struct btrfs_root *root,
4161                      u64 bytenr, u64 num_bytes, int reserved)
4162 {
4163         struct btrfs_block_group_cache *cache;
4164
4165         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4166         BUG_ON(!cache);
4167
4168         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4169
4170         btrfs_put_block_group(cache);
4171         return 0;
4172 }
4173
4174 /*
4175  * update size of reserved extents. this function may return -EAGAIN
4176  * if 'reserve' is true or 'sinfo' is false.
4177  */
4178 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4179                                  u64 num_bytes, int reserve, int sinfo)
4180 {
4181         int ret = 0;
4182         if (sinfo) {
4183                 struct btrfs_space_info *space_info = cache->space_info;
4184                 spin_lock(&space_info->lock);
4185                 spin_lock(&cache->lock);
4186                 if (reserve) {
4187                         if (cache->ro) {
4188                                 ret = -EAGAIN;
4189                         } else {
4190                                 cache->reserved += num_bytes;
4191                                 space_info->bytes_reserved += num_bytes;
4192                         }
4193                 } else {
4194                         if (cache->ro)
4195                                 space_info->bytes_readonly += num_bytes;
4196                         cache->reserved -= num_bytes;
4197                         space_info->bytes_reserved -= num_bytes;
4198                 }
4199                 spin_unlock(&cache->lock);
4200                 spin_unlock(&space_info->lock);
4201         } else {
4202                 spin_lock(&cache->lock);
4203                 if (cache->ro) {
4204                         ret = -EAGAIN;
4205                 } else {
4206                         if (reserve)
4207                                 cache->reserved += num_bytes;
4208                         else
4209                                 cache->reserved -= num_bytes;
4210                 }
4211                 spin_unlock(&cache->lock);
4212         }
4213         return ret;
4214 }
4215
4216 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4217                                 struct btrfs_root *root)
4218 {
4219         struct btrfs_fs_info *fs_info = root->fs_info;
4220         struct btrfs_caching_control *next;
4221         struct btrfs_caching_control *caching_ctl;
4222         struct btrfs_block_group_cache *cache;
4223
4224         down_write(&fs_info->extent_commit_sem);
4225
4226         list_for_each_entry_safe(caching_ctl, next,
4227                                  &fs_info->caching_block_groups, list) {
4228                 cache = caching_ctl->block_group;
4229                 if (block_group_cache_done(cache)) {
4230                         cache->last_byte_to_unpin = (u64)-1;
4231                         list_del_init(&caching_ctl->list);
4232                         put_caching_control(caching_ctl);
4233                 } else {
4234                         cache->last_byte_to_unpin = caching_ctl->progress;
4235                 }
4236         }
4237
4238         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4239                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4240         else
4241                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4242
4243         up_write(&fs_info->extent_commit_sem);
4244
4245         update_global_block_rsv(fs_info);
4246         return 0;
4247 }
4248
4249 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4250 {
4251         struct btrfs_fs_info *fs_info = root->fs_info;
4252         struct btrfs_block_group_cache *cache = NULL;
4253         u64 len;
4254
4255         while (start <= end) {
4256                 if (!cache ||
4257                     start >= cache->key.objectid + cache->key.offset) {
4258                         if (cache)
4259                                 btrfs_put_block_group(cache);
4260                         cache = btrfs_lookup_block_group(fs_info, start);
4261                         BUG_ON(!cache);
4262                 }
4263
4264                 len = cache->key.objectid + cache->key.offset - start;
4265                 len = min(len, end + 1 - start);
4266
4267                 if (start < cache->last_byte_to_unpin) {
4268                         len = min(len, cache->last_byte_to_unpin - start);
4269                         btrfs_add_free_space(cache, start, len);
4270                 }
4271
4272                 start += len;
4273
4274                 spin_lock(&cache->space_info->lock);
4275                 spin_lock(&cache->lock);
4276                 cache->pinned -= len;
4277                 cache->space_info->bytes_pinned -= len;
4278                 if (cache->ro) {
4279                         cache->space_info->bytes_readonly += len;
4280                 } else if (cache->reserved_pinned > 0) {
4281                         len = min(len, cache->reserved_pinned);
4282                         cache->reserved_pinned -= len;
4283                         cache->space_info->bytes_reserved += len;
4284                 }
4285                 spin_unlock(&cache->lock);
4286                 spin_unlock(&cache->space_info->lock);
4287         }
4288
4289         if (cache)
4290                 btrfs_put_block_group(cache);
4291         return 0;
4292 }
4293
4294 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4295                                struct btrfs_root *root)
4296 {
4297         struct btrfs_fs_info *fs_info = root->fs_info;
4298         struct extent_io_tree *unpin;
4299         struct btrfs_block_rsv *block_rsv;
4300         struct btrfs_block_rsv *next_rsv;
4301         u64 start;
4302         u64 end;
4303         int idx;
4304         int ret;
4305
4306         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4307                 unpin = &fs_info->freed_extents[1];
4308         else
4309                 unpin = &fs_info->freed_extents[0];
4310
4311         while (1) {
4312                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4313                                             EXTENT_DIRTY);
4314                 if (ret)
4315                         break;
4316
4317                 ret = btrfs_discard_extent(root, start, end + 1 - start);
4318
4319                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4320                 unpin_extent_range(root, start, end);
4321                 cond_resched();
4322         }
4323
4324         mutex_lock(&fs_info->durable_block_rsv_mutex);
4325         list_for_each_entry_safe(block_rsv, next_rsv,
4326                                  &fs_info->durable_block_rsv_list, list) {
4327
4328                 idx = trans->transid & 0x1;
4329                 if (block_rsv->freed[idx] > 0) {
4330                         block_rsv_add_bytes(block_rsv,
4331                                             block_rsv->freed[idx], 0);
4332                         block_rsv->freed[idx] = 0;
4333                 }
4334                 if (atomic_read(&block_rsv->usage) == 0) {
4335                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4336
4337                         if (block_rsv->freed[0] == 0 &&
4338                             block_rsv->freed[1] == 0) {
4339                                 list_del_init(&block_rsv->list);
4340                                 kfree(block_rsv);
4341                         }
4342                 } else {
4343                         btrfs_block_rsv_release(root, block_rsv, 0);
4344                 }
4345         }
4346         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4347
4348         return 0;
4349 }
4350
4351 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4352                                 struct btrfs_root *root,
4353                                 u64 bytenr, u64 num_bytes, u64 parent,
4354                                 u64 root_objectid, u64 owner_objectid,
4355                                 u64 owner_offset, int refs_to_drop,
4356                                 struct btrfs_delayed_extent_op *extent_op)
4357 {
4358         struct btrfs_key key;
4359         struct btrfs_path *path;
4360         struct btrfs_fs_info *info = root->fs_info;
4361         struct btrfs_root *extent_root = info->extent_root;
4362         struct extent_buffer *leaf;
4363         struct btrfs_extent_item *ei;
4364         struct btrfs_extent_inline_ref *iref;
4365         int ret;
4366         int is_data;
4367         int extent_slot = 0;
4368         int found_extent = 0;
4369         int num_to_del = 1;
4370         u32 item_size;
4371         u64 refs;
4372
4373         path = btrfs_alloc_path();
4374         if (!path)
4375                 return -ENOMEM;
4376
4377         path->reada = 1;
4378         path->leave_spinning = 1;
4379
4380         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4381         BUG_ON(!is_data && refs_to_drop != 1);
4382
4383         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4384                                     bytenr, num_bytes, parent,
4385                                     root_objectid, owner_objectid,
4386                                     owner_offset);
4387         if (ret == 0) {
4388                 extent_slot = path->slots[0];
4389                 while (extent_slot >= 0) {
4390                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4391                                               extent_slot);
4392                         if (key.objectid != bytenr)
4393                                 break;
4394                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4395                             key.offset == num_bytes) {
4396                                 found_extent = 1;
4397                                 break;
4398                         }
4399                         if (path->slots[0] - extent_slot > 5)
4400                                 break;
4401                         extent_slot--;
4402                 }
4403 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4404                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4405                 if (found_extent && item_size < sizeof(*ei))
4406                         found_extent = 0;
4407 #endif
4408                 if (!found_extent) {
4409                         BUG_ON(iref);
4410                         ret = remove_extent_backref(trans, extent_root, path,
4411                                                     NULL, refs_to_drop,
4412                                                     is_data);
4413                         BUG_ON(ret);
4414                         btrfs_release_path(extent_root, path);
4415                         path->leave_spinning = 1;
4416
4417                         key.objectid = bytenr;
4418                         key.type = BTRFS_EXTENT_ITEM_KEY;
4419                         key.offset = num_bytes;
4420
4421                         ret = btrfs_search_slot(trans, extent_root,
4422                                                 &key, path, -1, 1);
4423                         if (ret) {
4424                                 printk(KERN_ERR "umm, got %d back from search"
4425                                        ", was looking for %llu\n", ret,
4426                                        (unsigned long long)bytenr);
4427                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4428                         }
4429                         BUG_ON(ret);
4430                         extent_slot = path->slots[0];
4431                 }
4432         } else {
4433                 btrfs_print_leaf(extent_root, path->nodes[0]);
4434                 WARN_ON(1);
4435                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4436                        "parent %llu root %llu  owner %llu offset %llu\n",
4437                        (unsigned long long)bytenr,
4438                        (unsigned long long)parent,
4439                        (unsigned long long)root_objectid,
4440                        (unsigned long long)owner_objectid,
4441                        (unsigned long long)owner_offset);
4442         }
4443
4444         leaf = path->nodes[0];
4445         item_size = btrfs_item_size_nr(leaf, extent_slot);
4446 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4447         if (item_size < sizeof(*ei)) {
4448                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4449                 ret = convert_extent_item_v0(trans, extent_root, path,
4450                                              owner_objectid, 0);
4451                 BUG_ON(ret < 0);
4452
4453                 btrfs_release_path(extent_root, path);
4454                 path->leave_spinning = 1;
4455
4456                 key.objectid = bytenr;
4457                 key.type = BTRFS_EXTENT_ITEM_KEY;
4458                 key.offset = num_bytes;
4459
4460                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4461                                         -1, 1);
4462                 if (ret) {
4463                         printk(KERN_ERR "umm, got %d back from search"
4464                                ", was looking for %llu\n", ret,
4465                                (unsigned long long)bytenr);
4466                         btrfs_print_leaf(extent_root, path->nodes[0]);
4467                 }
4468                 BUG_ON(ret);
4469                 extent_slot = path->slots[0];
4470                 leaf = path->nodes[0];
4471                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4472         }
4473 #endif
4474         BUG_ON(item_size < sizeof(*ei));
4475         ei = btrfs_item_ptr(leaf, extent_slot,
4476                             struct btrfs_extent_item);
4477         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4478                 struct btrfs_tree_block_info *bi;
4479                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4480                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4481                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4482         }
4483
4484         refs = btrfs_extent_refs(leaf, ei);
4485         BUG_ON(refs < refs_to_drop);
4486         refs -= refs_to_drop;
4487
4488         if (refs > 0) {
4489                 if (extent_op)
4490                         __run_delayed_extent_op(extent_op, leaf, ei);
4491                 /*
4492                  * In the case of inline back ref, reference count will
4493                  * be updated by remove_extent_backref
4494                  */
4495                 if (iref) {
4496                         BUG_ON(!found_extent);
4497                 } else {
4498                         btrfs_set_extent_refs(leaf, ei, refs);
4499                         btrfs_mark_buffer_dirty(leaf);
4500                 }
4501                 if (found_extent) {
4502                         ret = remove_extent_backref(trans, extent_root, path,
4503                                                     iref, refs_to_drop,
4504                                                     is_data);
4505                         BUG_ON(ret);
4506                 }
4507         } else {
4508                 if (found_extent) {
4509                         BUG_ON(is_data && refs_to_drop !=
4510                                extent_data_ref_count(root, path, iref));
4511                         if (iref) {
4512                                 BUG_ON(path->slots[0] != extent_slot);
4513                         } else {
4514                                 BUG_ON(path->slots[0] != extent_slot + 1);
4515                                 path->slots[0] = extent_slot;
4516                                 num_to_del = 2;
4517                         }
4518                 }
4519
4520                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4521                                       num_to_del);
4522                 BUG_ON(ret);
4523                 btrfs_release_path(extent_root, path);
4524
4525                 if (is_data) {
4526                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4527                         BUG_ON(ret);
4528                 } else {
4529                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4530                              bytenr >> PAGE_CACHE_SHIFT,
4531                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4532                 }
4533
4534                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4535                 BUG_ON(ret);
4536         }
4537         btrfs_free_path(path);
4538         return ret;
4539 }
4540
4541 /*
4542  * when we free an block, it is possible (and likely) that we free the last
4543  * delayed ref for that extent as well.  This searches the delayed ref tree for
4544  * a given extent, and if there are no other delayed refs to be processed, it
4545  * removes it from the tree.
4546  */
4547 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4548                                       struct btrfs_root *root, u64 bytenr)
4549 {
4550         struct btrfs_delayed_ref_head *head;
4551         struct btrfs_delayed_ref_root *delayed_refs;
4552         struct btrfs_delayed_ref_node *ref;
4553         struct rb_node *node;
4554         int ret = 0;
4555
4556         delayed_refs = &trans->transaction->delayed_refs;
4557         spin_lock(&delayed_refs->lock);
4558         head = btrfs_find_delayed_ref_head(trans, bytenr);
4559         if (!head)
4560                 goto out;
4561
4562         node = rb_prev(&head->node.rb_node);
4563         if (!node)
4564                 goto out;
4565
4566         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4567
4568         /* there are still entries for this ref, we can't drop it */
4569         if (ref->bytenr == bytenr)
4570                 goto out;
4571
4572         if (head->extent_op) {
4573                 if (!head->must_insert_reserved)
4574                         goto out;
4575                 kfree(head->extent_op);
4576                 head->extent_op = NULL;
4577         }
4578
4579         /*
4580          * waiting for the lock here would deadlock.  If someone else has it
4581          * locked they are already in the process of dropping it anyway
4582          */
4583         if (!mutex_trylock(&head->mutex))
4584                 goto out;
4585
4586         /*
4587          * at this point we have a head with no other entries.  Go
4588          * ahead and process it.
4589          */
4590         head->node.in_tree = 0;
4591         rb_erase(&head->node.rb_node, &delayed_refs->root);
4592
4593         delayed_refs->num_entries--;
4594
4595         /*
4596          * we don't take a ref on the node because we're removing it from the
4597          * tree, so we just steal the ref the tree was holding.
4598          */
4599         delayed_refs->num_heads--;
4600         if (list_empty(&head->cluster))
4601                 delayed_refs->num_heads_ready--;
4602
4603         list_del_init(&head->cluster);
4604         spin_unlock(&delayed_refs->lock);
4605
4606         BUG_ON(head->extent_op);
4607         if (head->must_insert_reserved)
4608                 ret = 1;
4609
4610         mutex_unlock(&head->mutex);
4611         btrfs_put_delayed_ref(&head->node);
4612         return ret;
4613 out:
4614         spin_unlock(&delayed_refs->lock);
4615         return 0;
4616 }
4617
4618 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4619                            struct btrfs_root *root,
4620                            struct extent_buffer *buf,
4621                            u64 parent, int last_ref)
4622 {
4623         struct btrfs_block_rsv *block_rsv;
4624         struct btrfs_block_group_cache *cache = NULL;
4625         int ret;
4626
4627         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4628                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4629                                                 parent, root->root_key.objectid,
4630                                                 btrfs_header_level(buf),
4631                                                 BTRFS_DROP_DELAYED_REF, NULL);
4632                 BUG_ON(ret);
4633         }
4634
4635         if (!last_ref)
4636                 return;
4637
4638         block_rsv = get_block_rsv(trans, root);
4639         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4640         if (block_rsv->space_info != cache->space_info)
4641                 goto out;
4642
4643         if (btrfs_header_generation(buf) == trans->transid) {
4644                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4645                         ret = check_ref_cleanup(trans, root, buf->start);
4646                         if (!ret)
4647                                 goto pin;
4648                 }
4649
4650                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4651                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4652                         goto pin;
4653                 }
4654
4655                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4656
4657                 btrfs_add_free_space(cache, buf->start, buf->len);
4658                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4659                 if (ret == -EAGAIN) {
4660                         /* block group became read-only */
4661                         update_reserved_bytes(cache, buf->len, 0, 1);
4662                         goto out;
4663                 }
4664
4665                 ret = 1;
4666                 spin_lock(&block_rsv->lock);
4667                 if (block_rsv->reserved < block_rsv->size) {
4668                         block_rsv->reserved += buf->len;
4669                         ret = 0;
4670                 }
4671                 spin_unlock(&block_rsv->lock);
4672
4673                 if (ret) {
4674                         spin_lock(&cache->space_info->lock);
4675                         cache->space_info->bytes_reserved -= buf->len;
4676                         spin_unlock(&cache->space_info->lock);
4677                 }
4678                 goto out;
4679         }
4680 pin:
4681         if (block_rsv->durable && !cache->ro) {
4682                 ret = 0;
4683                 spin_lock(&cache->lock);
4684                 if (!cache->ro) {
4685                         cache->reserved_pinned += buf->len;
4686                         ret = 1;
4687                 }
4688                 spin_unlock(&cache->lock);
4689
4690                 if (ret) {
4691                         spin_lock(&block_rsv->lock);
4692                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4693                         spin_unlock(&block_rsv->lock);
4694                 }
4695         }
4696 out:
4697         btrfs_put_block_group(cache);
4698 }
4699
4700 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4701                       struct btrfs_root *root,
4702                       u64 bytenr, u64 num_bytes, u64 parent,
4703                       u64 root_objectid, u64 owner, u64 offset)
4704 {
4705         int ret;
4706
4707         /*
4708          * tree log blocks never actually go into the extent allocation
4709          * tree, just update pinning info and exit early.
4710          */
4711         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4712                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4713                 /* unlocks the pinned mutex */
4714                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4715                 ret = 0;
4716         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4717                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4718                                         parent, root_objectid, (int)owner,
4719                                         BTRFS_DROP_DELAYED_REF, NULL);
4720                 BUG_ON(ret);
4721         } else {
4722                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4723                                         parent, root_objectid, owner,
4724                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4725                 BUG_ON(ret);
4726         }
4727         return ret;
4728 }
4729
4730 static u64 stripe_align(struct btrfs_root *root, u64 val)
4731 {
4732         u64 mask = ((u64)root->stripesize - 1);
4733         u64 ret = (val + mask) & ~mask;
4734         return ret;
4735 }
4736
4737 /*
4738  * when we wait for progress in the block group caching, its because
4739  * our allocation attempt failed at least once.  So, we must sleep
4740  * and let some progress happen before we try again.
4741  *
4742  * This function will sleep at least once waiting for new free space to
4743  * show up, and then it will check the block group free space numbers
4744  * for our min num_bytes.  Another option is to have it go ahead
4745  * and look in the rbtree for a free extent of a given size, but this
4746  * is a good start.
4747  */
4748 static noinline int
4749 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4750                                 u64 num_bytes)
4751 {
4752         struct btrfs_caching_control *caching_ctl;
4753         DEFINE_WAIT(wait);
4754
4755         caching_ctl = get_caching_control(cache);
4756         if (!caching_ctl)
4757                 return 0;
4758
4759         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4760                    (cache->free_space >= num_bytes));
4761
4762         put_caching_control(caching_ctl);
4763         return 0;
4764 }
4765
4766 static noinline int
4767 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4768 {
4769         struct btrfs_caching_control *caching_ctl;
4770         DEFINE_WAIT(wait);
4771
4772         caching_ctl = get_caching_control(cache);
4773         if (!caching_ctl)
4774                 return 0;
4775
4776         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4777
4778         put_caching_control(caching_ctl);
4779         return 0;
4780 }
4781
4782 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4783 {
4784         int index;
4785         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4786                 index = 0;
4787         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4788                 index = 1;
4789         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4790                 index = 2;
4791         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4792                 index = 3;
4793         else
4794                 index = 4;
4795         return index;
4796 }
4797
4798 enum btrfs_loop_type {
4799         LOOP_FIND_IDEAL = 0,
4800         LOOP_CACHING_NOWAIT = 1,
4801         LOOP_CACHING_WAIT = 2,
4802         LOOP_ALLOC_CHUNK = 3,
4803         LOOP_NO_EMPTY_SIZE = 4,
4804 };
4805
4806 /*
4807  * walks the btree of allocated extents and find a hole of a given size.
4808  * The key ins is changed to record the hole:
4809  * ins->objectid == block start
4810  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4811  * ins->offset == number of blocks
4812  * Any available blocks before search_start are skipped.
4813  */
4814 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4815                                      struct btrfs_root *orig_root,
4816                                      u64 num_bytes, u64 empty_size,
4817                                      u64 search_start, u64 search_end,
4818                                      u64 hint_byte, struct btrfs_key *ins,
4819                                      int data)
4820 {
4821         int ret = 0;
4822         struct btrfs_root *root = orig_root->fs_info->extent_root;
4823         struct btrfs_free_cluster *last_ptr = NULL;
4824         struct btrfs_block_group_cache *block_group = NULL;
4825         int empty_cluster = 2 * 1024 * 1024;
4826         int allowed_chunk_alloc = 0;
4827         int done_chunk_alloc = 0;
4828         struct btrfs_space_info *space_info;
4829         int last_ptr_loop = 0;
4830         int loop = 0;
4831         int index = 0;
4832         bool found_uncached_bg = false;
4833         bool failed_cluster_refill = false;
4834         bool failed_alloc = false;
4835         bool use_cluster = true;
4836         u64 ideal_cache_percent = 0;
4837         u64 ideal_cache_offset = 0;
4838
4839         WARN_ON(num_bytes < root->sectorsize);
4840         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4841         ins->objectid = 0;
4842         ins->offset = 0;
4843
4844         space_info = __find_space_info(root->fs_info, data);
4845         if (!space_info) {
4846                 printk(KERN_ERR "No space info for %d\n", data);
4847                 return -ENOSPC;
4848         }
4849
4850         /*
4851          * If the space info is for both data and metadata it means we have a
4852          * small filesystem and we can't use the clustering stuff.
4853          */
4854         if (btrfs_mixed_space_info(space_info))
4855                 use_cluster = false;
4856
4857         if (orig_root->ref_cows || empty_size)
4858                 allowed_chunk_alloc = 1;
4859
4860         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4861                 last_ptr = &root->fs_info->meta_alloc_cluster;
4862                 if (!btrfs_test_opt(root, SSD))
4863                         empty_cluster = 64 * 1024;
4864         }
4865
4866         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4867             btrfs_test_opt(root, SSD)) {
4868                 last_ptr = &root->fs_info->data_alloc_cluster;
4869         }
4870
4871         if (last_ptr) {
4872                 spin_lock(&last_ptr->lock);
4873                 if (last_ptr->block_group)
4874                         hint_byte = last_ptr->window_start;
4875                 spin_unlock(&last_ptr->lock);
4876         }
4877
4878         search_start = max(search_start, first_logical_byte(root, 0));
4879         search_start = max(search_start, hint_byte);
4880
4881         if (!last_ptr)
4882                 empty_cluster = 0;
4883
4884         if (search_start == hint_byte) {
4885 ideal_cache:
4886                 block_group = btrfs_lookup_block_group(root->fs_info,
4887                                                        search_start);
4888                 /*
4889                  * we don't want to use the block group if it doesn't match our
4890                  * allocation bits, or if its not cached.
4891                  *
4892                  * However if we are re-searching with an ideal block group
4893                  * picked out then we don't care that the block group is cached.
4894                  */
4895                 if (block_group && block_group_bits(block_group, data) &&
4896                     (block_group->cached != BTRFS_CACHE_NO ||
4897                      search_start == ideal_cache_offset)) {
4898                         down_read(&space_info->groups_sem);
4899                         if (list_empty(&block_group->list) ||
4900                             block_group->ro) {
4901                                 /*
4902                                  * someone is removing this block group,
4903                                  * we can't jump into the have_block_group
4904                                  * target because our list pointers are not
4905                                  * valid
4906                                  */
4907                                 btrfs_put_block_group(block_group);
4908                                 up_read(&space_info->groups_sem);
4909                         } else {
4910                                 index = get_block_group_index(block_group);
4911                                 goto have_block_group;
4912                         }
4913                 } else if (block_group) {
4914                         btrfs_put_block_group(block_group);
4915                 }
4916         }
4917 search:
4918         down_read(&space_info->groups_sem);
4919         list_for_each_entry(block_group, &space_info->block_groups[index],
4920                             list) {
4921                 u64 offset;
4922                 int cached;
4923
4924                 btrfs_get_block_group(block_group);
4925                 search_start = block_group->key.objectid;
4926
4927 have_block_group:
4928                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4929                         u64 free_percent;
4930
4931                         ret = cache_block_group(block_group, trans, 1);
4932                         if (block_group->cached == BTRFS_CACHE_FINISHED)
4933                                 goto have_block_group;
4934
4935                         free_percent = btrfs_block_group_used(&block_group->item);
4936                         free_percent *= 100;
4937                         free_percent = div64_u64(free_percent,
4938                                                  block_group->key.offset);
4939                         free_percent = 100 - free_percent;
4940                         if (free_percent > ideal_cache_percent &&
4941                             likely(!block_group->ro)) {
4942                                 ideal_cache_offset = block_group->key.objectid;
4943                                 ideal_cache_percent = free_percent;
4944                         }
4945
4946                         /*
4947                          * We only want to start kthread caching if we are at
4948                          * the point where we will wait for caching to make
4949                          * progress, or if our ideal search is over and we've
4950                          * found somebody to start caching.
4951                          */
4952                         if (loop > LOOP_CACHING_NOWAIT ||
4953                             (loop > LOOP_FIND_IDEAL &&
4954                              atomic_read(&space_info->caching_threads) < 2)) {
4955                                 ret = cache_block_group(block_group, trans, 0);
4956                                 BUG_ON(ret);
4957                         }
4958                         found_uncached_bg = true;
4959
4960                         /*
4961                          * If loop is set for cached only, try the next block
4962                          * group.
4963                          */
4964                         if (loop == LOOP_FIND_IDEAL)
4965                                 goto loop;
4966                 }
4967
4968                 cached = block_group_cache_done(block_group);
4969                 if (unlikely(!cached))
4970                         found_uncached_bg = true;
4971
4972                 if (unlikely(block_group->ro))
4973                         goto loop;
4974
4975                 /*
4976                  * Ok we want to try and use the cluster allocator, so lets look
4977                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4978                  * have tried the cluster allocator plenty of times at this
4979                  * point and not have found anything, so we are likely way too
4980                  * fragmented for the clustering stuff to find anything, so lets
4981                  * just skip it and let the allocator find whatever block it can
4982                  * find
4983                  */
4984                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4985                         /*
4986                          * the refill lock keeps out other
4987                          * people trying to start a new cluster
4988                          */
4989                         spin_lock(&last_ptr->refill_lock);
4990                         if (last_ptr->block_group &&
4991                             (last_ptr->block_group->ro ||
4992                             !block_group_bits(last_ptr->block_group, data))) {
4993                                 offset = 0;
4994                                 goto refill_cluster;
4995                         }
4996
4997                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4998                                                  num_bytes, search_start);
4999                         if (offset) {
5000                                 /* we have a block, we're done */
5001                                 spin_unlock(&last_ptr->refill_lock);
5002                                 goto checks;
5003                         }
5004
5005                         spin_lock(&last_ptr->lock);
5006                         /*
5007                          * whoops, this cluster doesn't actually point to
5008                          * this block group.  Get a ref on the block
5009                          * group is does point to and try again
5010                          */
5011                         if (!last_ptr_loop && last_ptr->block_group &&
5012                             last_ptr->block_group != block_group) {
5013
5014                                 btrfs_put_block_group(block_group);
5015                                 block_group = last_ptr->block_group;
5016                                 btrfs_get_block_group(block_group);
5017                                 spin_unlock(&last_ptr->lock);
5018                                 spin_unlock(&last_ptr->refill_lock);
5019
5020                                 last_ptr_loop = 1;
5021                                 search_start = block_group->key.objectid;
5022                                 /*
5023                                  * we know this block group is properly
5024                                  * in the list because
5025                                  * btrfs_remove_block_group, drops the
5026                                  * cluster before it removes the block
5027                                  * group from the list
5028                                  */
5029                                 goto have_block_group;
5030                         }
5031                         spin_unlock(&last_ptr->lock);
5032 refill_cluster:
5033                         /*
5034                          * this cluster didn't work out, free it and
5035                          * start over
5036                          */
5037                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5038
5039                         last_ptr_loop = 0;
5040
5041                         /* allocate a cluster in this block group */
5042                         ret = btrfs_find_space_cluster(trans, root,
5043                                                block_group, last_ptr,
5044                                                offset, num_bytes,
5045                                                empty_cluster + empty_size);
5046                         if (ret == 0) {
5047                                 /*
5048                                  * now pull our allocation out of this
5049                                  * cluster
5050                                  */
5051                                 offset = btrfs_alloc_from_cluster(block_group,
5052                                                   last_ptr, num_bytes,
5053                                                   search_start);
5054                                 if (offset) {
5055                                         /* we found one, proceed */
5056                                         spin_unlock(&last_ptr->refill_lock);
5057                                         goto checks;
5058                                 }
5059                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5060                                    && !failed_cluster_refill) {
5061                                 spin_unlock(&last_ptr->refill_lock);
5062
5063                                 failed_cluster_refill = true;
5064                                 wait_block_group_cache_progress(block_group,
5065                                        num_bytes + empty_cluster + empty_size);
5066                                 goto have_block_group;
5067                         }
5068
5069                         /*
5070                          * at this point we either didn't find a cluster
5071                          * or we weren't able to allocate a block from our
5072                          * cluster.  Free the cluster we've been trying
5073                          * to use, and go to the next block group
5074                          */
5075                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5076                         spin_unlock(&last_ptr->refill_lock);
5077                         goto loop;
5078                 }
5079
5080                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5081                                                     num_bytes, empty_size);
5082                 /*
5083                  * If we didn't find a chunk, and we haven't failed on this
5084                  * block group before, and this block group is in the middle of
5085                  * caching and we are ok with waiting, then go ahead and wait
5086                  * for progress to be made, and set failed_alloc to true.
5087                  *
5088                  * If failed_alloc is true then we've already waited on this
5089                  * block group once and should move on to the next block group.
5090                  */
5091                 if (!offset && !failed_alloc && !cached &&
5092                     loop > LOOP_CACHING_NOWAIT) {
5093                         wait_block_group_cache_progress(block_group,
5094                                                 num_bytes + empty_size);
5095                         failed_alloc = true;
5096                         goto have_block_group;
5097                 } else if (!offset) {
5098                         goto loop;
5099                 }
5100 checks:
5101                 search_start = stripe_align(root, offset);
5102                 /* move on to the next group */
5103                 if (search_start + num_bytes >= search_end) {
5104                         btrfs_add_free_space(block_group, offset, num_bytes);
5105                         goto loop;
5106                 }
5107
5108                 /* move on to the next group */
5109                 if (search_start + num_bytes >
5110                     block_group->key.objectid + block_group->key.offset) {
5111                         btrfs_add_free_space(block_group, offset, num_bytes);
5112                         goto loop;
5113                 }
5114
5115                 ins->objectid = search_start;
5116                 ins->offset = num_bytes;
5117
5118                 if (offset < search_start)
5119                         btrfs_add_free_space(block_group, offset,
5120                                              search_start - offset);
5121                 BUG_ON(offset > search_start);
5122
5123                 ret = update_reserved_bytes(block_group, num_bytes, 1,
5124                                             (data & BTRFS_BLOCK_GROUP_DATA));
5125                 if (ret == -EAGAIN) {
5126                         btrfs_add_free_space(block_group, offset, num_bytes);
5127                         goto loop;
5128                 }
5129
5130                 /* we are all good, lets return */
5131                 ins->objectid = search_start;
5132                 ins->offset = num_bytes;
5133
5134                 if (offset < search_start)
5135                         btrfs_add_free_space(block_group, offset,
5136                                              search_start - offset);
5137                 BUG_ON(offset > search_start);
5138                 break;
5139 loop:
5140                 failed_cluster_refill = false;
5141                 failed_alloc = false;
5142                 BUG_ON(index != get_block_group_index(block_group));
5143                 btrfs_put_block_group(block_group);
5144         }
5145         up_read(&space_info->groups_sem);
5146
5147         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5148                 goto search;
5149
5150         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5151          *                      for them to make caching progress.  Also
5152          *                      determine the best possible bg to cache
5153          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5154          *                      caching kthreads as we move along
5155          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5156          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5157          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5158          *                      again
5159          */
5160         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5161             (found_uncached_bg || empty_size || empty_cluster ||
5162              allowed_chunk_alloc)) {
5163                 index = 0;
5164                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5165                         found_uncached_bg = false;
5166                         loop++;
5167                         if (!ideal_cache_percent &&
5168                             atomic_read(&space_info->caching_threads))
5169                                 goto search;
5170
5171                         /*
5172                          * 1 of the following 2 things have happened so far
5173                          *
5174                          * 1) We found an ideal block group for caching that
5175                          * is mostly full and will cache quickly, so we might
5176                          * as well wait for it.
5177                          *
5178                          * 2) We searched for cached only and we didn't find
5179                          * anything, and we didn't start any caching kthreads
5180                          * either, so chances are we will loop through and
5181                          * start a couple caching kthreads, and then come back
5182                          * around and just wait for them.  This will be slower
5183                          * because we will have 2 caching kthreads reading at
5184                          * the same time when we could have just started one
5185                          * and waited for it to get far enough to give us an
5186                          * allocation, so go ahead and go to the wait caching
5187                          * loop.
5188                          */
5189                         loop = LOOP_CACHING_WAIT;
5190                         search_start = ideal_cache_offset;
5191                         ideal_cache_percent = 0;
5192                         goto ideal_cache;
5193                 } else if (loop == LOOP_FIND_IDEAL) {
5194                         /*
5195                          * Didn't find a uncached bg, wait on anything we find
5196                          * next.
5197                          */
5198                         loop = LOOP_CACHING_WAIT;
5199                         goto search;
5200                 }
5201
5202                 if (loop < LOOP_CACHING_WAIT) {
5203                         loop++;
5204                         goto search;
5205                 }
5206
5207                 if (loop == LOOP_ALLOC_CHUNK) {
5208                         empty_size = 0;
5209                         empty_cluster = 0;
5210                 }
5211
5212                 if (allowed_chunk_alloc) {
5213                         ret = do_chunk_alloc(trans, root, num_bytes +
5214                                              2 * 1024 * 1024, data, 1);
5215                         allowed_chunk_alloc = 0;
5216                         done_chunk_alloc = 1;
5217                 } else if (!done_chunk_alloc) {
5218                         space_info->force_alloc = 1;
5219                 }
5220
5221                 if (loop < LOOP_NO_EMPTY_SIZE) {
5222                         loop++;
5223                         goto search;
5224                 }
5225                 ret = -ENOSPC;
5226         } else if (!ins->objectid) {
5227                 ret = -ENOSPC;
5228         }
5229
5230         /* we found what we needed */
5231         if (ins->objectid) {
5232                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5233                         trans->block_group = block_group->key.objectid;
5234
5235                 btrfs_put_block_group(block_group);
5236                 ret = 0;
5237         }
5238
5239         return ret;
5240 }
5241
5242 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5243                             int dump_block_groups)
5244 {
5245         struct btrfs_block_group_cache *cache;
5246         int index = 0;
5247
5248         spin_lock(&info->lock);
5249         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5250                (unsigned long long)(info->total_bytes - info->bytes_used -
5251                                     info->bytes_pinned - info->bytes_reserved -
5252                                     info->bytes_readonly),
5253                (info->full) ? "" : "not ");
5254         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5255                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5256                (unsigned long long)info->total_bytes,
5257                (unsigned long long)info->bytes_used,
5258                (unsigned long long)info->bytes_pinned,
5259                (unsigned long long)info->bytes_reserved,
5260                (unsigned long long)info->bytes_may_use,
5261                (unsigned long long)info->bytes_readonly);
5262         spin_unlock(&info->lock);
5263
5264         if (!dump_block_groups)
5265                 return;
5266
5267         down_read(&info->groups_sem);
5268 again:
5269         list_for_each_entry(cache, &info->block_groups[index], list) {
5270                 spin_lock(&cache->lock);
5271                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5272                        "%llu pinned %llu reserved\n",
5273                        (unsigned long long)cache->key.objectid,
5274                        (unsigned long long)cache->key.offset,
5275                        (unsigned long long)btrfs_block_group_used(&cache->item),
5276                        (unsigned long long)cache->pinned,
5277                        (unsigned long long)cache->reserved);
5278                 btrfs_dump_free_space(cache, bytes);
5279                 spin_unlock(&cache->lock);
5280         }
5281         if (++index < BTRFS_NR_RAID_TYPES)
5282                 goto again;
5283         up_read(&info->groups_sem);
5284 }
5285
5286 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5287                          struct btrfs_root *root,
5288                          u64 num_bytes, u64 min_alloc_size,
5289                          u64 empty_size, u64 hint_byte,
5290                          u64 search_end, struct btrfs_key *ins,
5291                          u64 data)
5292 {
5293         int ret;
5294         u64 search_start = 0;
5295
5296         data = btrfs_get_alloc_profile(root, data);
5297 again:
5298         /*
5299          * the only place that sets empty_size is btrfs_realloc_node, which
5300          * is not called recursively on allocations
5301          */
5302         if (empty_size || root->ref_cows)
5303                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5304                                      num_bytes + 2 * 1024 * 1024, data, 0);
5305
5306         WARN_ON(num_bytes < root->sectorsize);
5307         ret = find_free_extent(trans, root, num_bytes, empty_size,
5308                                search_start, search_end, hint_byte,
5309                                ins, data);
5310
5311         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5312                 num_bytes = num_bytes >> 1;
5313                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5314                 num_bytes = max(num_bytes, min_alloc_size);
5315                 do_chunk_alloc(trans, root->fs_info->extent_root,
5316                                num_bytes, data, 1);
5317                 goto again;
5318         }
5319         if (ret == -ENOSPC) {
5320                 struct btrfs_space_info *sinfo;
5321
5322                 sinfo = __find_space_info(root->fs_info, data);
5323                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5324                        "wanted %llu\n", (unsigned long long)data,
5325                        (unsigned long long)num_bytes);
5326                 dump_space_info(sinfo, num_bytes, 1);
5327         }
5328
5329         return ret;
5330 }
5331
5332 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5333 {
5334         struct btrfs_block_group_cache *cache;
5335         int ret = 0;
5336
5337         cache = btrfs_lookup_block_group(root->fs_info, start);
5338         if (!cache) {
5339                 printk(KERN_ERR "Unable to find block group for %llu\n",
5340                        (unsigned long long)start);
5341                 return -ENOSPC;
5342         }
5343
5344         ret = btrfs_discard_extent(root, start, len);
5345
5346         btrfs_add_free_space(cache, start, len);
5347         update_reserved_bytes(cache, len, 0, 1);
5348         btrfs_put_block_group(cache);
5349
5350         return ret;
5351 }
5352
5353 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5354                                       struct btrfs_root *root,
5355                                       u64 parent, u64 root_objectid,
5356                                       u64 flags, u64 owner, u64 offset,
5357                                       struct btrfs_key *ins, int ref_mod)
5358 {
5359         int ret;
5360         struct btrfs_fs_info *fs_info = root->fs_info;
5361         struct btrfs_extent_item *extent_item;
5362         struct btrfs_extent_inline_ref *iref;
5363         struct btrfs_path *path;
5364         struct extent_buffer *leaf;
5365         int type;
5366         u32 size;
5367
5368         if (parent > 0)
5369                 type = BTRFS_SHARED_DATA_REF_KEY;
5370         else
5371                 type = BTRFS_EXTENT_DATA_REF_KEY;
5372
5373         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5374
5375         path = btrfs_alloc_path();
5376         BUG_ON(!path);
5377
5378         path->leave_spinning = 1;
5379         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5380                                       ins, size);
5381         BUG_ON(ret);
5382
5383         leaf = path->nodes[0];
5384         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5385                                      struct btrfs_extent_item);
5386         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5387         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5388         btrfs_set_extent_flags(leaf, extent_item,
5389                                flags | BTRFS_EXTENT_FLAG_DATA);
5390
5391         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5392         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5393         if (parent > 0) {
5394                 struct btrfs_shared_data_ref *ref;
5395                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5396                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5397                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5398         } else {
5399                 struct btrfs_extent_data_ref *ref;
5400                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5401                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5402                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5403                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5404                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5405         }
5406
5407         btrfs_mark_buffer_dirty(path->nodes[0]);
5408         btrfs_free_path(path);
5409
5410         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5411         if (ret) {
5412                 printk(KERN_ERR "btrfs update block group failed for %llu "
5413                        "%llu\n", (unsigned long long)ins->objectid,
5414                        (unsigned long long)ins->offset);
5415                 BUG();
5416         }
5417         return ret;
5418 }
5419
5420 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5421                                      struct btrfs_root *root,
5422                                      u64 parent, u64 root_objectid,
5423                                      u64 flags, struct btrfs_disk_key *key,
5424                                      int level, struct btrfs_key *ins)
5425 {
5426         int ret;
5427         struct btrfs_fs_info *fs_info = root->fs_info;
5428         struct btrfs_extent_item *extent_item;
5429         struct btrfs_tree_block_info *block_info;
5430         struct btrfs_extent_inline_ref *iref;
5431         struct btrfs_path *path;
5432         struct extent_buffer *leaf;
5433         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5434
5435         path = btrfs_alloc_path();
5436         BUG_ON(!path);
5437
5438         path->leave_spinning = 1;
5439         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5440                                       ins, size);
5441         BUG_ON(ret);
5442
5443         leaf = path->nodes[0];
5444         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5445                                      struct btrfs_extent_item);
5446         btrfs_set_extent_refs(leaf, extent_item, 1);
5447         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5448         btrfs_set_extent_flags(leaf, extent_item,
5449                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5450         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5451
5452         btrfs_set_tree_block_key(leaf, block_info, key);
5453         btrfs_set_tree_block_level(leaf, block_info, level);
5454
5455         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5456         if (parent > 0) {
5457                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5458                 btrfs_set_extent_inline_ref_type(leaf, iref,
5459                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5460                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5461         } else {
5462                 btrfs_set_extent_inline_ref_type(leaf, iref,
5463                                                  BTRFS_TREE_BLOCK_REF_KEY);
5464                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5465         }
5466
5467         btrfs_mark_buffer_dirty(leaf);
5468         btrfs_free_path(path);
5469
5470         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5471         if (ret) {
5472                 printk(KERN_ERR "btrfs update block group failed for %llu "
5473                        "%llu\n", (unsigned long long)ins->objectid,
5474                        (unsigned long long)ins->offset);
5475                 BUG();
5476         }
5477         return ret;
5478 }
5479
5480 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5481                                      struct btrfs_root *root,
5482                                      u64 root_objectid, u64 owner,
5483                                      u64 offset, struct btrfs_key *ins)
5484 {
5485         int ret;
5486
5487         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5488
5489         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5490                                          0, root_objectid, owner, offset,
5491                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5492         return ret;
5493 }
5494
5495 /*
5496  * this is used by the tree logging recovery code.  It records that
5497  * an extent has been allocated and makes sure to clear the free
5498  * space cache bits as well
5499  */
5500 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5501                                    struct btrfs_root *root,
5502                                    u64 root_objectid, u64 owner, u64 offset,
5503                                    struct btrfs_key *ins)
5504 {
5505         int ret;
5506         struct btrfs_block_group_cache *block_group;
5507         struct btrfs_caching_control *caching_ctl;
5508         u64 start = ins->objectid;
5509         u64 num_bytes = ins->offset;
5510
5511         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5512         cache_block_group(block_group, trans, 0);
5513         caching_ctl = get_caching_control(block_group);
5514
5515         if (!caching_ctl) {
5516                 BUG_ON(!block_group_cache_done(block_group));
5517                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5518                 BUG_ON(ret);
5519         } else {
5520                 mutex_lock(&caching_ctl->mutex);
5521
5522                 if (start >= caching_ctl->progress) {
5523                         ret = add_excluded_extent(root, start, num_bytes);
5524                         BUG_ON(ret);
5525                 } else if (start + num_bytes <= caching_ctl->progress) {
5526                         ret = btrfs_remove_free_space(block_group,
5527                                                       start, num_bytes);
5528                         BUG_ON(ret);
5529                 } else {
5530                         num_bytes = caching_ctl->progress - start;
5531                         ret = btrfs_remove_free_space(block_group,
5532                                                       start, num_bytes);
5533                         BUG_ON(ret);
5534
5535                         start = caching_ctl->progress;
5536                         num_bytes = ins->objectid + ins->offset -
5537                                     caching_ctl->progress;
5538                         ret = add_excluded_extent(root, start, num_bytes);
5539                         BUG_ON(ret);
5540                 }
5541
5542                 mutex_unlock(&caching_ctl->mutex);
5543                 put_caching_control(caching_ctl);
5544         }
5545
5546         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5547         BUG_ON(ret);
5548         btrfs_put_block_group(block_group);
5549         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5550                                          0, owner, offset, ins, 1);
5551         return ret;
5552 }
5553
5554 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5555                                             struct btrfs_root *root,
5556                                             u64 bytenr, u32 blocksize,
5557                                             int level)
5558 {
5559         struct extent_buffer *buf;
5560
5561         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5562         if (!buf)
5563                 return ERR_PTR(-ENOMEM);
5564         btrfs_set_header_generation(buf, trans->transid);
5565         btrfs_set_buffer_lockdep_class(buf, level);
5566         btrfs_tree_lock(buf);
5567         clean_tree_block(trans, root, buf);
5568
5569         btrfs_set_lock_blocking(buf);
5570         btrfs_set_buffer_uptodate(buf);
5571
5572         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5573                 /*
5574                  * we allow two log transactions at a time, use different
5575                  * EXENT bit to differentiate dirty pages.
5576                  */
5577                 if (root->log_transid % 2 == 0)
5578                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5579                                         buf->start + buf->len - 1, GFP_NOFS);
5580                 else
5581                         set_extent_new(&root->dirty_log_pages, buf->start,
5582                                         buf->start + buf->len - 1, GFP_NOFS);
5583         } else {
5584                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5585                          buf->start + buf->len - 1, GFP_NOFS);
5586         }
5587         trans->blocks_used++;
5588         /* this returns a buffer locked for blocking */
5589         return buf;
5590 }
5591
5592 static struct btrfs_block_rsv *
5593 use_block_rsv(struct btrfs_trans_handle *trans,
5594               struct btrfs_root *root, u32 blocksize)
5595 {
5596         struct btrfs_block_rsv *block_rsv;
5597         int ret;
5598
5599         block_rsv = get_block_rsv(trans, root);
5600
5601         if (block_rsv->size == 0) {
5602                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5603                                              blocksize, 0);
5604                 if (ret)
5605                         return ERR_PTR(ret);
5606                 return block_rsv;
5607         }
5608
5609         ret = block_rsv_use_bytes(block_rsv, blocksize);
5610         if (!ret)
5611                 return block_rsv;
5612
5613         return ERR_PTR(-ENOSPC);
5614 }
5615
5616 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5617 {
5618         block_rsv_add_bytes(block_rsv, blocksize, 0);
5619         block_rsv_release_bytes(block_rsv, NULL, 0);
5620 }
5621
5622 /*
5623  * finds a free extent and does all the dirty work required for allocation
5624  * returns the key for the extent through ins, and a tree buffer for
5625  * the first block of the extent through buf.
5626  *
5627  * returns the tree buffer or NULL.
5628  */
5629 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5630                                         struct btrfs_root *root, u32 blocksize,
5631                                         u64 parent, u64 root_objectid,
5632                                         struct btrfs_disk_key *key, int level,
5633                                         u64 hint, u64 empty_size)
5634 {
5635         struct btrfs_key ins;
5636         struct btrfs_block_rsv *block_rsv;
5637         struct extent_buffer *buf;
5638         u64 flags = 0;
5639         int ret;
5640
5641
5642         block_rsv = use_block_rsv(trans, root, blocksize);
5643         if (IS_ERR(block_rsv))
5644                 return ERR_CAST(block_rsv);
5645
5646         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5647                                    empty_size, hint, (u64)-1, &ins, 0);
5648         if (ret) {
5649                 unuse_block_rsv(block_rsv, blocksize);
5650                 return ERR_PTR(ret);
5651         }
5652
5653         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5654                                     blocksize, level);
5655         BUG_ON(IS_ERR(buf));
5656
5657         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5658                 if (parent == 0)
5659                         parent = ins.objectid;
5660                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5661         } else
5662                 BUG_ON(parent > 0);
5663
5664         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5665                 struct btrfs_delayed_extent_op *extent_op;
5666                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5667                 BUG_ON(!extent_op);
5668                 if (key)
5669                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5670                 else
5671                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5672                 extent_op->flags_to_set = flags;
5673                 extent_op->update_key = 1;
5674                 extent_op->update_flags = 1;
5675                 extent_op->is_data = 0;
5676
5677                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5678                                         ins.offset, parent, root_objectid,
5679                                         level, BTRFS_ADD_DELAYED_EXTENT,
5680                                         extent_op);
5681                 BUG_ON(ret);
5682         }
5683         return buf;
5684 }
5685
5686 struct walk_control {
5687         u64 refs[BTRFS_MAX_LEVEL];
5688         u64 flags[BTRFS_MAX_LEVEL];
5689         struct btrfs_key update_progress;
5690         int stage;
5691         int level;
5692         int shared_level;
5693         int update_ref;
5694         int keep_locks;
5695         int reada_slot;
5696         int reada_count;
5697 };
5698
5699 #define DROP_REFERENCE  1
5700 #define UPDATE_BACKREF  2
5701
5702 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5703                                      struct btrfs_root *root,
5704                                      struct walk_control *wc,
5705                                      struct btrfs_path *path)
5706 {
5707         u64 bytenr;
5708         u64 generation;
5709         u64 refs;
5710         u64 flags;
5711         u64 last = 0;
5712         u32 nritems;
5713         u32 blocksize;
5714         struct btrfs_key key;
5715         struct extent_buffer *eb;
5716         int ret;
5717         int slot;
5718         int nread = 0;
5719
5720         if (path->slots[wc->level] < wc->reada_slot) {
5721                 wc->reada_count = wc->reada_count * 2 / 3;
5722                 wc->reada_count = max(wc->reada_count, 2);
5723         } else {
5724                 wc->reada_count = wc->reada_count * 3 / 2;
5725                 wc->reada_count = min_t(int, wc->reada_count,
5726                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5727         }
5728
5729         eb = path->nodes[wc->level];
5730         nritems = btrfs_header_nritems(eb);
5731         blocksize = btrfs_level_size(root, wc->level - 1);
5732
5733         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5734                 if (nread >= wc->reada_count)
5735                         break;
5736
5737                 cond_resched();
5738                 bytenr = btrfs_node_blockptr(eb, slot);
5739                 generation = btrfs_node_ptr_generation(eb, slot);
5740
5741                 if (slot == path->slots[wc->level])
5742                         goto reada;
5743
5744                 if (wc->stage == UPDATE_BACKREF &&
5745                     generation <= root->root_key.offset)
5746                         continue;
5747
5748                 /* We don't lock the tree block, it's OK to be racy here */
5749                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5750                                                &refs, &flags);
5751                 BUG_ON(ret);
5752                 BUG_ON(refs == 0);
5753
5754                 if (wc->stage == DROP_REFERENCE) {
5755                         if (refs == 1)
5756                                 goto reada;
5757
5758                         if (wc->level == 1 &&
5759                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5760                                 continue;
5761                         if (!wc->update_ref ||
5762                             generation <= root->root_key.offset)
5763                                 continue;
5764                         btrfs_node_key_to_cpu(eb, &key, slot);
5765                         ret = btrfs_comp_cpu_keys(&key,
5766                                                   &wc->update_progress);
5767                         if (ret < 0)
5768                                 continue;
5769                 } else {
5770                         if (wc->level == 1 &&
5771                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5772                                 continue;
5773                 }
5774 reada:
5775                 ret = readahead_tree_block(root, bytenr, blocksize,
5776                                            generation);
5777                 if (ret)
5778                         break;
5779                 last = bytenr + blocksize;
5780                 nread++;
5781         }
5782         wc->reada_slot = slot;
5783 }
5784
5785 /*
5786  * hepler to process tree block while walking down the tree.
5787  *
5788  * when wc->stage == UPDATE_BACKREF, this function updates
5789  * back refs for pointers in the block.
5790  *
5791  * NOTE: return value 1 means we should stop walking down.
5792  */
5793 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5794                                    struct btrfs_root *root,
5795                                    struct btrfs_path *path,
5796                                    struct walk_control *wc, int lookup_info)
5797 {
5798         int level = wc->level;
5799         struct extent_buffer *eb = path->nodes[level];
5800         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5801         int ret;
5802
5803         if (wc->stage == UPDATE_BACKREF &&
5804             btrfs_header_owner(eb) != root->root_key.objectid)
5805                 return 1;
5806
5807         /*
5808          * when reference count of tree block is 1, it won't increase
5809          * again. once full backref flag is set, we never clear it.
5810          */
5811         if (lookup_info &&
5812             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5813              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5814                 BUG_ON(!path->locks[level]);
5815                 ret = btrfs_lookup_extent_info(trans, root,
5816                                                eb->start, eb->len,
5817                                                &wc->refs[level],
5818                                                &wc->flags[level]);
5819                 BUG_ON(ret);
5820                 BUG_ON(wc->refs[level] == 0);
5821         }
5822
5823         if (wc->stage == DROP_REFERENCE) {
5824                 if (wc->refs[level] > 1)
5825                         return 1;
5826
5827                 if (path->locks[level] && !wc->keep_locks) {
5828                         btrfs_tree_unlock(eb);
5829                         path->locks[level] = 0;
5830                 }
5831                 return 0;
5832         }
5833
5834         /* wc->stage == UPDATE_BACKREF */
5835         if (!(wc->flags[level] & flag)) {
5836                 BUG_ON(!path->locks[level]);
5837                 ret = btrfs_inc_ref(trans, root, eb, 1);
5838                 BUG_ON(ret);
5839                 ret = btrfs_dec_ref(trans, root, eb, 0);
5840                 BUG_ON(ret);
5841                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5842                                                   eb->len, flag, 0);
5843                 BUG_ON(ret);
5844                 wc->flags[level] |= flag;
5845         }
5846
5847         /*
5848          * the block is shared by multiple trees, so it's not good to
5849          * keep the tree lock
5850          */
5851         if (path->locks[level] && level > 0) {
5852                 btrfs_tree_unlock(eb);
5853                 path->locks[level] = 0;
5854         }
5855         return 0;
5856 }
5857
5858 /*
5859  * hepler to process tree block pointer.
5860  *
5861  * when wc->stage == DROP_REFERENCE, this function checks
5862  * reference count of the block pointed to. if the block
5863  * is shared and we need update back refs for the subtree
5864  * rooted at the block, this function changes wc->stage to
5865  * UPDATE_BACKREF. if the block is shared and there is no
5866  * need to update back, this function drops the reference
5867  * to the block.
5868  *
5869  * NOTE: return value 1 means we should stop walking down.
5870  */
5871 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5872                                  struct btrfs_root *root,
5873                                  struct btrfs_path *path,
5874                                  struct walk_control *wc, int *lookup_info)
5875 {
5876         u64 bytenr;
5877         u64 generation;
5878         u64 parent;
5879         u32 blocksize;
5880         struct btrfs_key key;
5881         struct extent_buffer *next;
5882         int level = wc->level;
5883         int reada = 0;
5884         int ret = 0;
5885
5886         generation = btrfs_node_ptr_generation(path->nodes[level],
5887                                                path->slots[level]);
5888         /*
5889          * if the lower level block was created before the snapshot
5890          * was created, we know there is no need to update back refs
5891          * for the subtree
5892          */
5893         if (wc->stage == UPDATE_BACKREF &&
5894             generation <= root->root_key.offset) {
5895                 *lookup_info = 1;
5896                 return 1;
5897         }
5898
5899         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5900         blocksize = btrfs_level_size(root, level - 1);
5901
5902         next = btrfs_find_tree_block(root, bytenr, blocksize);
5903         if (!next) {
5904                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5905                 if (!next)
5906                         return -ENOMEM;
5907                 reada = 1;
5908         }
5909         btrfs_tree_lock(next);
5910         btrfs_set_lock_blocking(next);
5911
5912         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5913                                        &wc->refs[level - 1],
5914                                        &wc->flags[level - 1]);
5915         BUG_ON(ret);
5916         BUG_ON(wc->refs[level - 1] == 0);
5917         *lookup_info = 0;
5918
5919         if (wc->stage == DROP_REFERENCE) {
5920                 if (wc->refs[level - 1] > 1) {
5921                         if (level == 1 &&
5922                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5923                                 goto skip;
5924
5925                         if (!wc->update_ref ||
5926                             generation <= root->root_key.offset)
5927                                 goto skip;
5928
5929                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5930                                               path->slots[level]);
5931                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5932                         if (ret < 0)
5933                                 goto skip;
5934
5935                         wc->stage = UPDATE_BACKREF;
5936                         wc->shared_level = level - 1;
5937                 }
5938         } else {
5939                 if (level == 1 &&
5940                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5941                         goto skip;
5942         }
5943
5944         if (!btrfs_buffer_uptodate(next, generation)) {
5945                 btrfs_tree_unlock(next);
5946                 free_extent_buffer(next);
5947                 next = NULL;
5948                 *lookup_info = 1;
5949         }
5950
5951         if (!next) {
5952                 if (reada && level == 1)
5953                         reada_walk_down(trans, root, wc, path);
5954                 next = read_tree_block(root, bytenr, blocksize, generation);
5955                 btrfs_tree_lock(next);
5956                 btrfs_set_lock_blocking(next);
5957         }
5958
5959         level--;
5960         BUG_ON(level != btrfs_header_level(next));
5961         path->nodes[level] = next;
5962         path->slots[level] = 0;
5963         path->locks[level] = 1;
5964         wc->level = level;
5965         if (wc->level == 1)
5966                 wc->reada_slot = 0;
5967         return 0;
5968 skip:
5969         wc->refs[level - 1] = 0;
5970         wc->flags[level - 1] = 0;
5971         if (wc->stage == DROP_REFERENCE) {
5972                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5973                         parent = path->nodes[level]->start;
5974                 } else {
5975                         BUG_ON(root->root_key.objectid !=
5976                                btrfs_header_owner(path->nodes[level]));
5977                         parent = 0;
5978                 }
5979
5980                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5981                                         root->root_key.objectid, level - 1, 0);
5982                 BUG_ON(ret);
5983         }
5984         btrfs_tree_unlock(next);
5985         free_extent_buffer(next);
5986         *lookup_info = 1;
5987         return 1;
5988 }
5989
5990 /*
5991  * hepler to process tree block while walking up the tree.
5992  *
5993  * when wc->stage == DROP_REFERENCE, this function drops
5994  * reference count on the block.
5995  *
5996  * when wc->stage == UPDATE_BACKREF, this function changes
5997  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5998  * to UPDATE_BACKREF previously while processing the block.
5999  *
6000  * NOTE: return value 1 means we should stop walking up.
6001  */
6002 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6003                                  struct btrfs_root *root,
6004                                  struct btrfs_path *path,
6005                                  struct walk_control *wc)
6006 {
6007         int ret;
6008         int level = wc->level;
6009         struct extent_buffer *eb = path->nodes[level];
6010         u64 parent = 0;
6011
6012         if (wc->stage == UPDATE_BACKREF) {
6013                 BUG_ON(wc->shared_level < level);
6014                 if (level < wc->shared_level)
6015                         goto out;
6016
6017                 ret = find_next_key(path, level + 1, &wc->update_progress);
6018                 if (ret > 0)
6019                         wc->update_ref = 0;
6020
6021                 wc->stage = DROP_REFERENCE;
6022                 wc->shared_level = -1;
6023                 path->slots[level] = 0;
6024
6025                 /*
6026                  * check reference count again if the block isn't locked.
6027                  * we should start walking down the tree again if reference
6028                  * count is one.
6029                  */
6030                 if (!path->locks[level]) {
6031                         BUG_ON(level == 0);
6032                         btrfs_tree_lock(eb);
6033                         btrfs_set_lock_blocking(eb);
6034                         path->locks[level] = 1;
6035
6036                         ret = btrfs_lookup_extent_info(trans, root,
6037                                                        eb->start, eb->len,
6038                                                        &wc->refs[level],
6039                                                        &wc->flags[level]);
6040                         BUG_ON(ret);
6041                         BUG_ON(wc->refs[level] == 0);
6042                         if (wc->refs[level] == 1) {
6043                                 btrfs_tree_unlock(eb);
6044                                 path->locks[level] = 0;
6045                                 return 1;
6046                         }
6047                 }
6048         }
6049
6050         /* wc->stage == DROP_REFERENCE */
6051         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6052
6053         if (wc->refs[level] == 1) {
6054                 if (level == 0) {
6055                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6056                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6057                         else
6058                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6059                         BUG_ON(ret);
6060                 }
6061                 /* make block locked assertion in clean_tree_block happy */
6062                 if (!path->locks[level] &&
6063                     btrfs_header_generation(eb) == trans->transid) {
6064                         btrfs_tree_lock(eb);
6065                         btrfs_set_lock_blocking(eb);
6066                         path->locks[level] = 1;
6067                 }
6068                 clean_tree_block(trans, root, eb);
6069         }
6070
6071         if (eb == root->node) {
6072                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6073                         parent = eb->start;
6074                 else
6075                         BUG_ON(root->root_key.objectid !=
6076                                btrfs_header_owner(eb));
6077         } else {
6078                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6079                         parent = path->nodes[level + 1]->start;
6080                 else
6081                         BUG_ON(root->root_key.objectid !=
6082                                btrfs_header_owner(path->nodes[level + 1]));
6083         }
6084
6085         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6086 out:
6087         wc->refs[level] = 0;
6088         wc->flags[level] = 0;
6089         return 0;
6090 }
6091
6092 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6093                                    struct btrfs_root *root,
6094                                    struct btrfs_path *path,
6095                                    struct walk_control *wc)
6096 {
6097         int level = wc->level;
6098         int lookup_info = 1;
6099         int ret;
6100
6101         while (level >= 0) {
6102                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6103                 if (ret > 0)
6104                         break;
6105
6106                 if (level == 0)
6107                         break;
6108
6109                 if (path->slots[level] >=
6110                     btrfs_header_nritems(path->nodes[level]))
6111                         break;
6112
6113                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6114                 if (ret > 0) {
6115                         path->slots[level]++;
6116                         continue;
6117                 } else if (ret < 0)
6118                         return ret;
6119                 level = wc->level;
6120         }
6121         return 0;
6122 }
6123
6124 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6125                                  struct btrfs_root *root,
6126                                  struct btrfs_path *path,
6127                                  struct walk_control *wc, int max_level)
6128 {
6129         int level = wc->level;
6130         int ret;
6131
6132         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6133         while (level < max_level && path->nodes[level]) {
6134                 wc->level = level;
6135                 if (path->slots[level] + 1 <
6136                     btrfs_header_nritems(path->nodes[level])) {
6137                         path->slots[level]++;
6138                         return 0;
6139                 } else {
6140                         ret = walk_up_proc(trans, root, path, wc);
6141                         if (ret > 0)
6142                                 return 0;
6143
6144                         if (path->locks[level]) {
6145                                 btrfs_tree_unlock(path->nodes[level]);
6146                                 path->locks[level] = 0;
6147                         }
6148                         free_extent_buffer(path->nodes[level]);
6149                         path->nodes[level] = NULL;
6150                         level++;
6151                 }
6152         }
6153         return 1;
6154 }
6155
6156 /*
6157  * drop a subvolume tree.
6158  *
6159  * this function traverses the tree freeing any blocks that only
6160  * referenced by the tree.
6161  *
6162  * when a shared tree block is found. this function decreases its
6163  * reference count by one. if update_ref is true, this function
6164  * also make sure backrefs for the shared block and all lower level
6165  * blocks are properly updated.
6166  */
6167 int btrfs_drop_snapshot(struct btrfs_root *root,
6168                         struct btrfs_block_rsv *block_rsv, int update_ref)
6169 {
6170         struct btrfs_path *path;
6171         struct btrfs_trans_handle *trans;
6172         struct btrfs_root *tree_root = root->fs_info->tree_root;
6173         struct btrfs_root_item *root_item = &root->root_item;
6174         struct walk_control *wc;
6175         struct btrfs_key key;
6176         int err = 0;
6177         int ret;
6178         int level;
6179
6180         path = btrfs_alloc_path();
6181         BUG_ON(!path);
6182
6183         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6184         BUG_ON(!wc);
6185
6186         trans = btrfs_start_transaction(tree_root, 0);
6187         if (block_rsv)
6188                 trans->block_rsv = block_rsv;
6189
6190         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6191                 level = btrfs_header_level(root->node);
6192                 path->nodes[level] = btrfs_lock_root_node(root);
6193                 btrfs_set_lock_blocking(path->nodes[level]);
6194                 path->slots[level] = 0;
6195                 path->locks[level] = 1;
6196                 memset(&wc->update_progress, 0,
6197                        sizeof(wc->update_progress));
6198         } else {
6199                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6200                 memcpy(&wc->update_progress, &key,
6201                        sizeof(wc->update_progress));
6202
6203                 level = root_item->drop_level;
6204                 BUG_ON(level == 0);
6205                 path->lowest_level = level;
6206                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6207                 path->lowest_level = 0;
6208                 if (ret < 0) {
6209                         err = ret;
6210                         goto out;
6211                 }
6212                 WARN_ON(ret > 0);
6213
6214                 /*
6215                  * unlock our path, this is safe because only this
6216                  * function is allowed to delete this snapshot
6217                  */
6218                 btrfs_unlock_up_safe(path, 0);
6219
6220                 level = btrfs_header_level(root->node);
6221                 while (1) {
6222                         btrfs_tree_lock(path->nodes[level]);
6223                         btrfs_set_lock_blocking(path->nodes[level]);
6224
6225                         ret = btrfs_lookup_extent_info(trans, root,
6226                                                 path->nodes[level]->start,
6227                                                 path->nodes[level]->len,
6228                                                 &wc->refs[level],
6229                                                 &wc->flags[level]);
6230                         BUG_ON(ret);
6231                         BUG_ON(wc->refs[level] == 0);
6232
6233                         if (level == root_item->drop_level)
6234                                 break;
6235
6236                         btrfs_tree_unlock(path->nodes[level]);
6237                         WARN_ON(wc->refs[level] != 1);
6238                         level--;
6239                 }
6240         }
6241
6242         wc->level = level;
6243         wc->shared_level = -1;
6244         wc->stage = DROP_REFERENCE;
6245         wc->update_ref = update_ref;
6246         wc->keep_locks = 0;
6247         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6248
6249         while (1) {
6250                 ret = walk_down_tree(trans, root, path, wc);
6251                 if (ret < 0) {
6252                         err = ret;
6253                         break;
6254                 }
6255
6256                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6257                 if (ret < 0) {
6258                         err = ret;
6259                         break;
6260                 }
6261
6262                 if (ret > 0) {
6263                         BUG_ON(wc->stage != DROP_REFERENCE);
6264                         break;
6265                 }
6266
6267                 if (wc->stage == DROP_REFERENCE) {
6268                         level = wc->level;
6269                         btrfs_node_key(path->nodes[level],
6270                                        &root_item->drop_progress,
6271                                        path->slots[level]);
6272                         root_item->drop_level = level;
6273                 }
6274
6275                 BUG_ON(wc->level == 0);
6276                 if (btrfs_should_end_transaction(trans, tree_root)) {
6277                         ret = btrfs_update_root(trans, tree_root,
6278                                                 &root->root_key,
6279                                                 root_item);
6280                         BUG_ON(ret);
6281
6282                         btrfs_end_transaction_throttle(trans, tree_root);
6283                         trans = btrfs_start_transaction(tree_root, 0);
6284                         if (block_rsv)
6285                                 trans->block_rsv = block_rsv;
6286                 }
6287         }
6288         btrfs_release_path(root, path);
6289         BUG_ON(err);
6290
6291         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6292         BUG_ON(ret);
6293
6294         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6295                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6296                                            NULL, NULL);
6297                 BUG_ON(ret < 0);
6298                 if (ret > 0) {
6299                         ret = btrfs_del_orphan_item(trans, tree_root,
6300                                                     root->root_key.objectid);
6301                         BUG_ON(ret);
6302                 }
6303         }
6304
6305         if (root->in_radix) {
6306                 btrfs_free_fs_root(tree_root->fs_info, root);
6307         } else {
6308                 free_extent_buffer(root->node);
6309                 free_extent_buffer(root->commit_root);
6310                 kfree(root);
6311         }
6312 out:
6313         btrfs_end_transaction_throttle(trans, tree_root);
6314         kfree(wc);
6315         btrfs_free_path(path);
6316         return err;
6317 }
6318
6319 /*
6320  * drop subtree rooted at tree block 'node'.
6321  *
6322  * NOTE: this function will unlock and release tree block 'node'
6323  */
6324 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6325                         struct btrfs_root *root,
6326                         struct extent_buffer *node,
6327                         struct extent_buffer *parent)
6328 {
6329         struct btrfs_path *path;
6330         struct walk_control *wc;
6331         int level;
6332         int parent_level;
6333         int ret = 0;
6334         int wret;
6335
6336         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6337
6338         path = btrfs_alloc_path();
6339         BUG_ON(!path);
6340
6341         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6342         BUG_ON(!wc);
6343
6344         btrfs_assert_tree_locked(parent);
6345         parent_level = btrfs_header_level(parent);
6346         extent_buffer_get(parent);
6347         path->nodes[parent_level] = parent;
6348         path->slots[parent_level] = btrfs_header_nritems(parent);
6349
6350         btrfs_assert_tree_locked(node);
6351         level = btrfs_header_level(node);
6352         path->nodes[level] = node;
6353         path->slots[level] = 0;
6354         path->locks[level] = 1;
6355
6356         wc->refs[parent_level] = 1;
6357         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6358         wc->level = level;
6359         wc->shared_level = -1;
6360         wc->stage = DROP_REFERENCE;
6361         wc->update_ref = 0;
6362         wc->keep_locks = 1;
6363         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6364
6365         while (1) {
6366                 wret = walk_down_tree(trans, root, path, wc);
6367                 if (wret < 0) {
6368                         ret = wret;
6369                         break;
6370                 }
6371
6372                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6373                 if (wret < 0)
6374                         ret = wret;
6375                 if (wret != 0)
6376                         break;
6377         }
6378
6379         kfree(wc);
6380         btrfs_free_path(path);
6381         return ret;
6382 }
6383
6384 #if 0
6385 static unsigned long calc_ra(unsigned long start, unsigned long last,
6386                              unsigned long nr)
6387 {
6388         return min(last, start + nr - 1);
6389 }
6390
6391 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6392                                          u64 len)
6393 {
6394         u64 page_start;
6395         u64 page_end;
6396         unsigned long first_index;
6397         unsigned long last_index;
6398         unsigned long i;
6399         struct page *page;
6400         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6401         struct file_ra_state *ra;
6402         struct btrfs_ordered_extent *ordered;
6403         unsigned int total_read = 0;
6404         unsigned int total_dirty = 0;
6405         int ret = 0;
6406
6407         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6408
6409         mutex_lock(&inode->i_mutex);
6410         first_index = start >> PAGE_CACHE_SHIFT;
6411         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6412
6413         /* make sure the dirty trick played by the caller work */
6414         ret = invalidate_inode_pages2_range(inode->i_mapping,
6415                                             first_index, last_index);
6416         if (ret)
6417                 goto out_unlock;
6418
6419         file_ra_state_init(ra, inode->i_mapping);
6420
6421         for (i = first_index ; i <= last_index; i++) {
6422                 if (total_read % ra->ra_pages == 0) {
6423                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6424                                        calc_ra(i, last_index, ra->ra_pages));
6425                 }
6426                 total_read++;
6427 again:
6428                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6429                         BUG_ON(1);
6430                 page = grab_cache_page(inode->i_mapping, i);
6431                 if (!page) {
6432                         ret = -ENOMEM;
6433                         goto out_unlock;
6434                 }
6435                 if (!PageUptodate(page)) {
6436                         btrfs_readpage(NULL, page);
6437                         lock_page(page);
6438                         if (!PageUptodate(page)) {
6439                                 unlock_page(page);
6440                                 page_cache_release(page);
6441                                 ret = -EIO;
6442                                 goto out_unlock;
6443                         }
6444                 }
6445                 wait_on_page_writeback(page);
6446
6447                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6448                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6449                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6450
6451                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6452                 if (ordered) {
6453                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6454                         unlock_page(page);
6455                         page_cache_release(page);
6456                         btrfs_start_ordered_extent(inode, ordered, 1);
6457                         btrfs_put_ordered_extent(ordered);
6458                         goto again;
6459                 }
6460                 set_page_extent_mapped(page);
6461
6462                 if (i == first_index)
6463                         set_extent_bits(io_tree, page_start, page_end,
6464                                         EXTENT_BOUNDARY, GFP_NOFS);
6465                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6466
6467                 set_page_dirty(page);
6468                 total_dirty++;
6469
6470                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6471                 unlock_page(page);
6472                 page_cache_release(page);
6473         }
6474
6475 out_unlock:
6476         kfree(ra);
6477         mutex_unlock(&inode->i_mutex);
6478         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6479         return ret;
6480 }
6481
6482 static noinline int relocate_data_extent(struct inode *reloc_inode,
6483                                          struct btrfs_key *extent_key,
6484                                          u64 offset)
6485 {
6486         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6487         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6488         struct extent_map *em;
6489         u64 start = extent_key->objectid - offset;
6490         u64 end = start + extent_key->offset - 1;
6491
6492         em = alloc_extent_map(GFP_NOFS);
6493         BUG_ON(!em || IS_ERR(em));
6494
6495         em->start = start;
6496         em->len = extent_key->offset;
6497         em->block_len = extent_key->offset;
6498         em->block_start = extent_key->objectid;
6499         em->bdev = root->fs_info->fs_devices->latest_bdev;
6500         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6501
6502         /* setup extent map to cheat btrfs_readpage */
6503         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6504         while (1) {
6505                 int ret;
6506                 write_lock(&em_tree->lock);
6507                 ret = add_extent_mapping(em_tree, em);
6508                 write_unlock(&em_tree->lock);
6509                 if (ret != -EEXIST) {
6510                         free_extent_map(em);
6511                         break;
6512                 }
6513                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6514         }
6515         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6516
6517         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6518 }
6519
6520 struct btrfs_ref_path {
6521         u64 extent_start;
6522         u64 nodes[BTRFS_MAX_LEVEL];
6523         u64 root_objectid;
6524         u64 root_generation;
6525         u64 owner_objectid;
6526         u32 num_refs;
6527         int lowest_level;
6528         int current_level;
6529         int shared_level;
6530
6531         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6532         u64 new_nodes[BTRFS_MAX_LEVEL];
6533 };
6534
6535 struct disk_extent {
6536         u64 ram_bytes;
6537         u64 disk_bytenr;
6538         u64 disk_num_bytes;
6539         u64 offset;
6540         u64 num_bytes;
6541         u8 compression;
6542         u8 encryption;
6543         u16 other_encoding;
6544 };
6545
6546 static int is_cowonly_root(u64 root_objectid)
6547 {
6548         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6549             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6550             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6551             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6552             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6553             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6554                 return 1;
6555         return 0;
6556 }
6557
6558 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6559                                     struct btrfs_root *extent_root,
6560                                     struct btrfs_ref_path *ref_path,
6561                                     int first_time)
6562 {
6563         struct extent_buffer *leaf;
6564         struct btrfs_path *path;
6565         struct btrfs_extent_ref *ref;
6566         struct btrfs_key key;
6567         struct btrfs_key found_key;
6568         u64 bytenr;
6569         u32 nritems;
6570         int level;
6571         int ret = 1;
6572
6573         path = btrfs_alloc_path();
6574         if (!path)
6575                 return -ENOMEM;
6576
6577         if (first_time) {
6578                 ref_path->lowest_level = -1;
6579                 ref_path->current_level = -1;
6580                 ref_path->shared_level = -1;
6581                 goto walk_up;
6582         }
6583 walk_down:
6584         level = ref_path->current_level - 1;
6585         while (level >= -1) {
6586                 u64 parent;
6587                 if (level < ref_path->lowest_level)
6588                         break;
6589
6590                 if (level >= 0)
6591                         bytenr = ref_path->nodes[level];
6592                 else
6593                         bytenr = ref_path->extent_start;
6594                 BUG_ON(bytenr == 0);
6595
6596                 parent = ref_path->nodes[level + 1];
6597                 ref_path->nodes[level + 1] = 0;
6598                 ref_path->current_level = level;
6599                 BUG_ON(parent == 0);
6600
6601                 key.objectid = bytenr;
6602                 key.offset = parent + 1;
6603                 key.type = BTRFS_EXTENT_REF_KEY;
6604
6605                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6606                 if (ret < 0)
6607                         goto out;
6608                 BUG_ON(ret == 0);
6609
6610                 leaf = path->nodes[0];
6611                 nritems = btrfs_header_nritems(leaf);
6612                 if (path->slots[0] >= nritems) {
6613                         ret = btrfs_next_leaf(extent_root, path);
6614                         if (ret < 0)
6615                                 goto out;
6616                         if (ret > 0)
6617                                 goto next;
6618                         leaf = path->nodes[0];
6619                 }
6620
6621                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6622                 if (found_key.objectid == bytenr &&
6623                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6624                         if (level < ref_path->shared_level)
6625                                 ref_path->shared_level = level;
6626                         goto found;
6627                 }
6628 next:
6629                 level--;
6630                 btrfs_release_path(extent_root, path);
6631                 cond_resched();
6632         }
6633         /* reached lowest level */
6634         ret = 1;
6635         goto out;
6636 walk_up:
6637         level = ref_path->current_level;
6638         while (level < BTRFS_MAX_LEVEL - 1) {
6639                 u64 ref_objectid;
6640
6641                 if (level >= 0)
6642                         bytenr = ref_path->nodes[level];
6643                 else
6644                         bytenr = ref_path->extent_start;
6645
6646                 BUG_ON(bytenr == 0);
6647
6648                 key.objectid = bytenr;
6649                 key.offset = 0;
6650                 key.type = BTRFS_EXTENT_REF_KEY;
6651
6652                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6653                 if (ret < 0)
6654                         goto out;
6655
6656                 leaf = path->nodes[0];
6657                 nritems = btrfs_header_nritems(leaf);
6658                 if (path->slots[0] >= nritems) {
6659                         ret = btrfs_next_leaf(extent_root, path);
6660                         if (ret < 0)
6661                                 goto out;
6662                         if (ret > 0) {
6663                                 /* the extent was freed by someone */
6664                                 if (ref_path->lowest_level == level)
6665                                         goto out;
6666                                 btrfs_release_path(extent_root, path);
6667                                 goto walk_down;
6668                         }
6669                         leaf = path->nodes[0];
6670                 }
6671
6672                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6673                 if (found_key.objectid != bytenr ||
6674                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6675                         /* the extent was freed by someone */
6676                         if (ref_path->lowest_level == level) {
6677                                 ret = 1;
6678                                 goto out;
6679                         }
6680                         btrfs_release_path(extent_root, path);
6681                         goto walk_down;
6682                 }
6683 found:
6684                 ref = btrfs_item_ptr(leaf, path->slots[0],
6685                                 struct btrfs_extent_ref);
6686                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6687                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6688                         if (first_time) {
6689                                 level = (int)ref_objectid;
6690                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6691                                 ref_path->lowest_level = level;
6692                                 ref_path->current_level = level;
6693                                 ref_path->nodes[level] = bytenr;
6694                         } else {
6695                                 WARN_ON(ref_objectid != level);
6696                         }
6697                 } else {
6698                         WARN_ON(level != -1);
6699                 }
6700                 first_time = 0;
6701
6702                 if (ref_path->lowest_level == level) {
6703                         ref_path->owner_objectid = ref_objectid;
6704                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6705                 }
6706
6707                 /*
6708                  * the block is tree root or the block isn't in reference
6709                  * counted tree.
6710                  */
6711                 if (found_key.objectid == found_key.offset ||
6712                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6713                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6714                         ref_path->root_generation =
6715                                 btrfs_ref_generation(leaf, ref);
6716                         if (level < 0) {
6717                                 /* special reference from the tree log */
6718                                 ref_path->nodes[0] = found_key.offset;
6719                                 ref_path->current_level = 0;
6720                         }
6721                         ret = 0;
6722                         goto out;
6723                 }
6724
6725                 level++;
6726                 BUG_ON(ref_path->nodes[level] != 0);
6727                 ref_path->nodes[level] = found_key.offset;
6728                 ref_path->current_level = level;
6729
6730                 /*
6731                  * the reference was created in the running transaction,
6732                  * no need to continue walking up.
6733                  */
6734                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6735                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6736                         ref_path->root_generation =
6737                                 btrfs_ref_generation(leaf, ref);
6738                         ret = 0;
6739                         goto out;
6740                 }
6741
6742                 btrfs_release_path(extent_root, path);
6743                 cond_resched();
6744         }
6745         /* reached max tree level, but no tree root found. */
6746         BUG();
6747 out:
6748         btrfs_free_path(path);
6749         return ret;
6750 }
6751
6752 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6753                                 struct btrfs_root *extent_root,
6754                                 struct btrfs_ref_path *ref_path,
6755                                 u64 extent_start)
6756 {
6757         memset(ref_path, 0, sizeof(*ref_path));
6758         ref_path->extent_start = extent_start;
6759
6760         return __next_ref_path(trans, extent_root, ref_path, 1);
6761 }
6762
6763 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6764                                struct btrfs_root *extent_root,
6765                                struct btrfs_ref_path *ref_path)
6766 {
6767         return __next_ref_path(trans, extent_root, ref_path, 0);
6768 }
6769
6770 static noinline int get_new_locations(struct inode *reloc_inode,
6771                                       struct btrfs_key *extent_key,
6772                                       u64 offset, int no_fragment,
6773                                       struct disk_extent **extents,
6774                                       int *nr_extents)
6775 {
6776         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6777         struct btrfs_path *path;
6778         struct btrfs_file_extent_item *fi;
6779         struct extent_buffer *leaf;
6780         struct disk_extent *exts = *extents;
6781         struct btrfs_key found_key;
6782         u64 cur_pos;
6783         u64 last_byte;
6784         u32 nritems;
6785         int nr = 0;
6786         int max = *nr_extents;
6787         int ret;
6788
6789         WARN_ON(!no_fragment && *extents);
6790         if (!exts) {
6791                 max = 1;
6792                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6793                 if (!exts)
6794                         return -ENOMEM;
6795         }
6796
6797         path = btrfs_alloc_path();
6798         BUG_ON(!path);
6799
6800         cur_pos = extent_key->objectid - offset;
6801         last_byte = extent_key->objectid + extent_key->offset;
6802         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6803                                        cur_pos, 0);
6804         if (ret < 0)
6805                 goto out;
6806         if (ret > 0) {
6807                 ret = -ENOENT;
6808                 goto out;
6809         }
6810
6811         while (1) {
6812                 leaf = path->nodes[0];
6813                 nritems = btrfs_header_nritems(leaf);
6814                 if (path->slots[0] >= nritems) {
6815                         ret = btrfs_next_leaf(root, path);
6816                         if (ret < 0)
6817                                 goto out;
6818                         if (ret > 0)
6819                                 break;
6820                         leaf = path->nodes[0];
6821                 }
6822
6823                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6824                 if (found_key.offset != cur_pos ||
6825                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6826                     found_key.objectid != reloc_inode->i_ino)
6827                         break;
6828
6829                 fi = btrfs_item_ptr(leaf, path->slots[0],
6830                                     struct btrfs_file_extent_item);
6831                 if (btrfs_file_extent_type(leaf, fi) !=
6832                     BTRFS_FILE_EXTENT_REG ||
6833                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6834                         break;
6835
6836                 if (nr == max) {
6837                         struct disk_extent *old = exts;
6838                         max *= 2;
6839                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6840                         memcpy(exts, old, sizeof(*exts) * nr);
6841                         if (old != *extents)
6842                                 kfree(old);
6843                 }
6844
6845                 exts[nr].disk_bytenr =
6846                         btrfs_file_extent_disk_bytenr(leaf, fi);
6847                 exts[nr].disk_num_bytes =
6848                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6849                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6850                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6851                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6852                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6853                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6854                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6855                                                                            fi);
6856                 BUG_ON(exts[nr].offset > 0);
6857                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6858                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6859
6860                 cur_pos += exts[nr].num_bytes;
6861                 nr++;
6862
6863                 if (cur_pos + offset >= last_byte)
6864                         break;
6865
6866                 if (no_fragment) {
6867                         ret = 1;
6868                         goto out;
6869                 }
6870                 path->slots[0]++;
6871         }
6872
6873         BUG_ON(cur_pos + offset > last_byte);
6874         if (cur_pos + offset < last_byte) {
6875                 ret = -ENOENT;
6876                 goto out;
6877         }
6878         ret = 0;
6879 out:
6880         btrfs_free_path(path);
6881         if (ret) {
6882                 if (exts != *extents)
6883                         kfree(exts);
6884         } else {
6885                 *extents = exts;
6886                 *nr_extents = nr;
6887         }
6888         return ret;
6889 }
6890
6891 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6892                                         struct btrfs_root *root,
6893                                         struct btrfs_path *path,
6894                                         struct btrfs_key *extent_key,
6895                                         struct btrfs_key *leaf_key,
6896                                         struct btrfs_ref_path *ref_path,
6897                                         struct disk_extent *new_extents,
6898                                         int nr_extents)
6899 {
6900         struct extent_buffer *leaf;
6901         struct btrfs_file_extent_item *fi;
6902         struct inode *inode = NULL;
6903         struct btrfs_key key;
6904         u64 lock_start = 0;
6905         u64 lock_end = 0;
6906         u64 num_bytes;
6907         u64 ext_offset;
6908         u64 search_end = (u64)-1;
6909         u32 nritems;
6910         int nr_scaned = 0;
6911         int extent_locked = 0;
6912         int extent_type;
6913         int ret;
6914
6915         memcpy(&key, leaf_key, sizeof(key));
6916         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6917                 if (key.objectid < ref_path->owner_objectid ||
6918                     (key.objectid == ref_path->owner_objectid &&
6919                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6920                         key.objectid = ref_path->owner_objectid;
6921                         key.type = BTRFS_EXTENT_DATA_KEY;
6922                         key.offset = 0;
6923                 }
6924         }
6925
6926         while (1) {
6927                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6928                 if (ret < 0)
6929                         goto out;
6930
6931                 leaf = path->nodes[0];
6932                 nritems = btrfs_header_nritems(leaf);
6933 next:
6934                 if (extent_locked && ret > 0) {
6935                         /*
6936                          * the file extent item was modified by someone
6937                          * before the extent got locked.
6938                          */
6939                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6940                                       lock_end, GFP_NOFS);
6941                         extent_locked = 0;
6942                 }
6943
6944                 if (path->slots[0] >= nritems) {
6945                         if (++nr_scaned > 2)
6946                                 break;
6947
6948                         BUG_ON(extent_locked);
6949                         ret = btrfs_next_leaf(root, path);
6950                         if (ret < 0)
6951                                 goto out;
6952                         if (ret > 0)
6953                                 break;
6954                         leaf = path->nodes[0];
6955                         nritems = btrfs_header_nritems(leaf);
6956                 }
6957
6958                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6959
6960                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6961                         if ((key.objectid > ref_path->owner_objectid) ||
6962                             (key.objectid == ref_path->owner_objectid &&
6963                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6964                             key.offset >= search_end)
6965                                 break;
6966                 }
6967
6968                 if (inode && key.objectid != inode->i_ino) {
6969                         BUG_ON(extent_locked);
6970                         btrfs_release_path(root, path);
6971                         mutex_unlock(&inode->i_mutex);
6972                         iput(inode);
6973                         inode = NULL;
6974                         continue;
6975                 }
6976
6977                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6978                         path->slots[0]++;
6979                         ret = 1;
6980                         goto next;
6981                 }
6982                 fi = btrfs_item_ptr(leaf, path->slots[0],
6983                                     struct btrfs_file_extent_item);
6984                 extent_type = btrfs_file_extent_type(leaf, fi);
6985                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6986                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6987                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6988                      extent_key->objectid)) {
6989                         path->slots[0]++;
6990                         ret = 1;
6991                         goto next;
6992                 }
6993
6994                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6995                 ext_offset = btrfs_file_extent_offset(leaf, fi);
6996
6997                 if (search_end == (u64)-1) {
6998                         search_end = key.offset - ext_offset +
6999                                 btrfs_file_extent_ram_bytes(leaf, fi);
7000                 }
7001
7002                 if (!extent_locked) {
7003                         lock_start = key.offset;
7004                         lock_end = lock_start + num_bytes - 1;
7005                 } else {
7006                         if (lock_start > key.offset ||
7007                             lock_end + 1 < key.offset + num_bytes) {
7008                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7009                                               lock_start, lock_end, GFP_NOFS);
7010                                 extent_locked = 0;
7011                         }
7012                 }
7013
7014                 if (!inode) {
7015                         btrfs_release_path(root, path);
7016
7017                         inode = btrfs_iget_locked(root->fs_info->sb,
7018                                                   key.objectid, root);
7019                         if (inode->i_state & I_NEW) {
7020                                 BTRFS_I(inode)->root = root;
7021                                 BTRFS_I(inode)->location.objectid =
7022                                         key.objectid;
7023                                 BTRFS_I(inode)->location.type =
7024                                         BTRFS_INODE_ITEM_KEY;
7025                                 BTRFS_I(inode)->location.offset = 0;
7026                                 btrfs_read_locked_inode(inode);
7027                                 unlock_new_inode(inode);
7028                         }
7029                         /*
7030                          * some code call btrfs_commit_transaction while
7031                          * holding the i_mutex, so we can't use mutex_lock
7032                          * here.
7033                          */
7034                         if (is_bad_inode(inode) ||
7035                             !mutex_trylock(&inode->i_mutex)) {
7036                                 iput(inode);
7037                                 inode = NULL;
7038                                 key.offset = (u64)-1;
7039                                 goto skip;
7040                         }
7041                 }
7042
7043                 if (!extent_locked) {
7044                         struct btrfs_ordered_extent *ordered;
7045
7046                         btrfs_release_path(root, path);
7047
7048                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7049                                     lock_end, GFP_NOFS);
7050                         ordered = btrfs_lookup_first_ordered_extent(inode,
7051                                                                     lock_end);
7052                         if (ordered &&
7053                             ordered->file_offset <= lock_end &&
7054                             ordered->file_offset + ordered->len > lock_start) {
7055                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7056                                               lock_start, lock_end, GFP_NOFS);
7057                                 btrfs_start_ordered_extent(inode, ordered, 1);
7058                                 btrfs_put_ordered_extent(ordered);
7059                                 key.offset += num_bytes;
7060                                 goto skip;
7061                         }
7062                         if (ordered)
7063                                 btrfs_put_ordered_extent(ordered);
7064
7065                         extent_locked = 1;
7066                         continue;
7067                 }
7068
7069                 if (nr_extents == 1) {
7070                         /* update extent pointer in place */
7071                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7072                                                 new_extents[0].disk_bytenr);
7073                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7074                                                 new_extents[0].disk_num_bytes);
7075                         btrfs_mark_buffer_dirty(leaf);
7076
7077                         btrfs_drop_extent_cache(inode, key.offset,
7078                                                 key.offset + num_bytes - 1, 0);
7079
7080                         ret = btrfs_inc_extent_ref(trans, root,
7081                                                 new_extents[0].disk_bytenr,
7082                                                 new_extents[0].disk_num_bytes,
7083                                                 leaf->start,
7084                                                 root->root_key.objectid,
7085                                                 trans->transid,
7086                                                 key.objectid);
7087                         BUG_ON(ret);
7088
7089                         ret = btrfs_free_extent(trans, root,
7090                                                 extent_key->objectid,
7091                                                 extent_key->offset,
7092                                                 leaf->start,
7093                                                 btrfs_header_owner(leaf),
7094                                                 btrfs_header_generation(leaf),
7095                                                 key.objectid, 0);
7096                         BUG_ON(ret);
7097
7098                         btrfs_release_path(root, path);
7099                         key.offset += num_bytes;
7100                 } else {
7101                         BUG_ON(1);
7102 #if 0
7103                         u64 alloc_hint;
7104                         u64 extent_len;
7105                         int i;
7106                         /*
7107                          * drop old extent pointer at first, then insert the
7108                          * new pointers one bye one
7109                          */
7110                         btrfs_release_path(root, path);
7111                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7112                                                  key.offset + num_bytes,
7113                                                  key.offset, &alloc_hint);
7114                         BUG_ON(ret);
7115
7116                         for (i = 0; i < nr_extents; i++) {
7117                                 if (ext_offset >= new_extents[i].num_bytes) {
7118                                         ext_offset -= new_extents[i].num_bytes;
7119                                         continue;
7120                                 }
7121                                 extent_len = min(new_extents[i].num_bytes -
7122                                                  ext_offset, num_bytes);
7123
7124                                 ret = btrfs_insert_empty_item(trans, root,
7125                                                               path, &key,
7126                                                               sizeof(*fi));
7127                                 BUG_ON(ret);
7128
7129                                 leaf = path->nodes[0];
7130                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7131                                                 struct btrfs_file_extent_item);
7132                                 btrfs_set_file_extent_generation(leaf, fi,
7133                                                         trans->transid);
7134                                 btrfs_set_file_extent_type(leaf, fi,
7135                                                         BTRFS_FILE_EXTENT_REG);
7136                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7137                                                 new_extents[i].disk_bytenr);
7138                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7139                                                 new_extents[i].disk_num_bytes);
7140                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7141                                                 new_extents[i].ram_bytes);
7142
7143                                 btrfs_set_file_extent_compression(leaf, fi,
7144                                                 new_extents[i].compression);
7145                                 btrfs_set_file_extent_encryption(leaf, fi,
7146                                                 new_extents[i].encryption);
7147                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7148                                                 new_extents[i].other_encoding);
7149
7150                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7151                                                         extent_len);
7152                                 ext_offset += new_extents[i].offset;
7153                                 btrfs_set_file_extent_offset(leaf, fi,
7154                                                         ext_offset);
7155                                 btrfs_mark_buffer_dirty(leaf);
7156
7157                                 btrfs_drop_extent_cache(inode, key.offset,
7158                                                 key.offset + extent_len - 1, 0);
7159
7160                                 ret = btrfs_inc_extent_ref(trans, root,
7161                                                 new_extents[i].disk_bytenr,
7162                                                 new_extents[i].disk_num_bytes,
7163                                                 leaf->start,
7164                                                 root->root_key.objectid,
7165                                                 trans->transid, key.objectid);
7166                                 BUG_ON(ret);
7167                                 btrfs_release_path(root, path);
7168
7169                                 inode_add_bytes(inode, extent_len);
7170
7171                                 ext_offset = 0;
7172                                 num_bytes -= extent_len;
7173                                 key.offset += extent_len;
7174
7175                                 if (num_bytes == 0)
7176                                         break;
7177                         }
7178                         BUG_ON(i >= nr_extents);
7179 #endif
7180                 }
7181
7182                 if (extent_locked) {
7183                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7184                                       lock_end, GFP_NOFS);
7185                         extent_locked = 0;
7186                 }
7187 skip:
7188                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7189                     key.offset >= search_end)
7190                         break;
7191
7192                 cond_resched();
7193         }
7194         ret = 0;
7195 out:
7196         btrfs_release_path(root, path);
7197         if (inode) {
7198                 mutex_unlock(&inode->i_mutex);
7199                 if (extent_locked) {
7200                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7201                                       lock_end, GFP_NOFS);
7202                 }
7203                 iput(inode);
7204         }
7205         return ret;
7206 }
7207
7208 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7209                                struct btrfs_root *root,
7210                                struct extent_buffer *buf, u64 orig_start)
7211 {
7212         int level;
7213         int ret;
7214
7215         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7216         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7217
7218         level = btrfs_header_level(buf);
7219         if (level == 0) {
7220                 struct btrfs_leaf_ref *ref;
7221                 struct btrfs_leaf_ref *orig_ref;
7222
7223                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7224                 if (!orig_ref)
7225                         return -ENOENT;
7226
7227                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7228                 if (!ref) {
7229                         btrfs_free_leaf_ref(root, orig_ref);
7230                         return -ENOMEM;
7231                 }
7232
7233                 ref->nritems = orig_ref->nritems;
7234                 memcpy(ref->extents, orig_ref->extents,
7235                         sizeof(ref->extents[0]) * ref->nritems);
7236
7237                 btrfs_free_leaf_ref(root, orig_ref);
7238
7239                 ref->root_gen = trans->transid;
7240                 ref->bytenr = buf->start;
7241                 ref->owner = btrfs_header_owner(buf);
7242                 ref->generation = btrfs_header_generation(buf);
7243
7244                 ret = btrfs_add_leaf_ref(root, ref, 0);
7245                 WARN_ON(ret);
7246                 btrfs_free_leaf_ref(root, ref);
7247         }
7248         return 0;
7249 }
7250
7251 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7252                                         struct extent_buffer *leaf,
7253                                         struct btrfs_block_group_cache *group,
7254                                         struct btrfs_root *target_root)
7255 {
7256         struct btrfs_key key;
7257         struct inode *inode = NULL;
7258         struct btrfs_file_extent_item *fi;
7259         struct extent_state *cached_state = NULL;
7260         u64 num_bytes;
7261         u64 skip_objectid = 0;
7262         u32 nritems;
7263         u32 i;
7264
7265         nritems = btrfs_header_nritems(leaf);
7266         for (i = 0; i < nritems; i++) {
7267                 btrfs_item_key_to_cpu(leaf, &key, i);
7268                 if (key.objectid == skip_objectid ||
7269                     key.type != BTRFS_EXTENT_DATA_KEY)
7270                         continue;
7271                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7272                 if (btrfs_file_extent_type(leaf, fi) ==
7273                     BTRFS_FILE_EXTENT_INLINE)
7274                         continue;
7275                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7276                         continue;
7277                 if (!inode || inode->i_ino != key.objectid) {
7278                         iput(inode);
7279                         inode = btrfs_ilookup(target_root->fs_info->sb,
7280                                               key.objectid, target_root, 1);
7281                 }
7282                 if (!inode) {
7283                         skip_objectid = key.objectid;
7284                         continue;
7285                 }
7286                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7287
7288                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7289                                  key.offset + num_bytes - 1, 0, &cached_state,
7290                                  GFP_NOFS);
7291                 btrfs_drop_extent_cache(inode, key.offset,
7292                                         key.offset + num_bytes - 1, 1);
7293                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7294                                      key.offset + num_bytes - 1, &cached_state,
7295                                      GFP_NOFS);
7296                 cond_resched();
7297         }
7298         iput(inode);
7299         return 0;
7300 }
7301
7302 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7303                                         struct btrfs_root *root,
7304                                         struct extent_buffer *leaf,
7305                                         struct btrfs_block_group_cache *group,
7306                                         struct inode *reloc_inode)
7307 {
7308         struct btrfs_key key;
7309         struct btrfs_key extent_key;
7310         struct btrfs_file_extent_item *fi;
7311         struct btrfs_leaf_ref *ref;
7312         struct disk_extent *new_extent;
7313         u64 bytenr;
7314         u64 num_bytes;
7315         u32 nritems;
7316         u32 i;
7317         int ext_index;
7318         int nr_extent;
7319         int ret;
7320
7321         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7322         BUG_ON(!new_extent);
7323
7324         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7325         BUG_ON(!ref);
7326
7327         ext_index = -1;
7328         nritems = btrfs_header_nritems(leaf);
7329         for (i = 0; i < nritems; i++) {
7330                 btrfs_item_key_to_cpu(leaf, &key, i);
7331                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7332                         continue;
7333                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7334                 if (btrfs_file_extent_type(leaf, fi) ==
7335                     BTRFS_FILE_EXTENT_INLINE)
7336                         continue;
7337                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7338                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7339                 if (bytenr == 0)
7340                         continue;
7341
7342                 ext_index++;
7343                 if (bytenr >= group->key.objectid + group->key.offset ||
7344                     bytenr + num_bytes <= group->key.objectid)
7345                         continue;
7346
7347                 extent_key.objectid = bytenr;
7348                 extent_key.offset = num_bytes;
7349                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7350                 nr_extent = 1;
7351                 ret = get_new_locations(reloc_inode, &extent_key,
7352                                         group->key.objectid, 1,
7353                                         &new_extent, &nr_extent);
7354                 if (ret > 0)
7355                         continue;
7356                 BUG_ON(ret < 0);
7357
7358                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7359                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7360                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7361                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7362
7363                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7364                                                 new_extent->disk_bytenr);
7365                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7366                                                 new_extent->disk_num_bytes);
7367                 btrfs_mark_buffer_dirty(leaf);
7368
7369                 ret = btrfs_inc_extent_ref(trans, root,
7370                                         new_extent->disk_bytenr,
7371                                         new_extent->disk_num_bytes,
7372                                         leaf->start,
7373                                         root->root_key.objectid,
7374                                         trans->transid, key.objectid);
7375                 BUG_ON(ret);
7376
7377                 ret = btrfs_free_extent(trans, root,
7378                                         bytenr, num_bytes, leaf->start,
7379                                         btrfs_header_owner(leaf),
7380                                         btrfs_header_generation(leaf),
7381                                         key.objectid, 0);
7382                 BUG_ON(ret);
7383                 cond_resched();
7384         }
7385         kfree(new_extent);
7386         BUG_ON(ext_index + 1 != ref->nritems);
7387         btrfs_free_leaf_ref(root, ref);
7388         return 0;
7389 }
7390
7391 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7392                           struct btrfs_root *root)
7393 {
7394         struct btrfs_root *reloc_root;
7395         int ret;
7396
7397         if (root->reloc_root) {
7398                 reloc_root = root->reloc_root;
7399                 root->reloc_root = NULL;
7400                 list_add(&reloc_root->dead_list,
7401                          &root->fs_info->dead_reloc_roots);
7402
7403                 btrfs_set_root_bytenr(&reloc_root->root_item,
7404                                       reloc_root->node->start);
7405                 btrfs_set_root_level(&root->root_item,
7406                                      btrfs_header_level(reloc_root->node));
7407                 memset(&reloc_root->root_item.drop_progress, 0,
7408                         sizeof(struct btrfs_disk_key));
7409                 reloc_root->root_item.drop_level = 0;
7410
7411                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7412                                         &reloc_root->root_key,
7413                                         &reloc_root->root_item);
7414                 BUG_ON(ret);
7415         }
7416         return 0;
7417 }
7418
7419 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7420 {
7421         struct btrfs_trans_handle *trans;
7422         struct btrfs_root *reloc_root;
7423         struct btrfs_root *prev_root = NULL;
7424         struct list_head dead_roots;
7425         int ret;
7426         unsigned long nr;
7427
7428         INIT_LIST_HEAD(&dead_roots);
7429         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7430
7431         while (!list_empty(&dead_roots)) {
7432                 reloc_root = list_entry(dead_roots.prev,
7433                                         struct btrfs_root, dead_list);
7434                 list_del_init(&reloc_root->dead_list);
7435
7436                 BUG_ON(reloc_root->commit_root != NULL);
7437                 while (1) {
7438                         trans = btrfs_join_transaction(root, 1);
7439                         BUG_ON(!trans);
7440
7441                         mutex_lock(&root->fs_info->drop_mutex);
7442                         ret = btrfs_drop_snapshot(trans, reloc_root);
7443                         if (ret != -EAGAIN)
7444                                 break;
7445                         mutex_unlock(&root->fs_info->drop_mutex);
7446
7447                         nr = trans->blocks_used;
7448                         ret = btrfs_end_transaction(trans, root);
7449                         BUG_ON(ret);
7450                         btrfs_btree_balance_dirty(root, nr);
7451                 }
7452
7453                 free_extent_buffer(reloc_root->node);
7454
7455                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7456                                      &reloc_root->root_key);
7457                 BUG_ON(ret);
7458                 mutex_unlock(&root->fs_info->drop_mutex);
7459
7460                 nr = trans->blocks_used;
7461                 ret = btrfs_end_transaction(trans, root);
7462                 BUG_ON(ret);
7463                 btrfs_btree_balance_dirty(root, nr);
7464
7465                 kfree(prev_root);
7466                 prev_root = reloc_root;
7467         }
7468         if (prev_root) {
7469                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7470                 kfree(prev_root);
7471         }
7472         return 0;
7473 }
7474
7475 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7476 {
7477         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7478         return 0;
7479 }
7480
7481 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7482 {
7483         struct btrfs_root *reloc_root;
7484         struct btrfs_trans_handle *trans;
7485         struct btrfs_key location;
7486         int found;
7487         int ret;
7488
7489         mutex_lock(&root->fs_info->tree_reloc_mutex);
7490         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7491         BUG_ON(ret);
7492         found = !list_empty(&root->fs_info->dead_reloc_roots);
7493         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7494
7495         if (found) {
7496                 trans = btrfs_start_transaction(root, 1);
7497                 BUG_ON(!trans);
7498                 ret = btrfs_commit_transaction(trans, root);
7499                 BUG_ON(ret);
7500         }
7501
7502         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7503         location.offset = (u64)-1;
7504         location.type = BTRFS_ROOT_ITEM_KEY;
7505
7506         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7507         BUG_ON(!reloc_root);
7508         btrfs_orphan_cleanup(reloc_root);
7509         return 0;
7510 }
7511
7512 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7513                                     struct btrfs_root *root)
7514 {
7515         struct btrfs_root *reloc_root;
7516         struct extent_buffer *eb;
7517         struct btrfs_root_item *root_item;
7518         struct btrfs_key root_key;
7519         int ret;
7520
7521         BUG_ON(!root->ref_cows);
7522         if (root->reloc_root)
7523                 return 0;
7524
7525         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7526         BUG_ON(!root_item);
7527
7528         ret = btrfs_copy_root(trans, root, root->commit_root,
7529                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7530         BUG_ON(ret);
7531
7532         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7533         root_key.offset = root->root_key.objectid;
7534         root_key.type = BTRFS_ROOT_ITEM_KEY;
7535
7536         memcpy(root_item, &root->root_item, sizeof(root_item));
7537         btrfs_set_root_refs(root_item, 0);
7538         btrfs_set_root_bytenr(root_item, eb->start);
7539         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7540         btrfs_set_root_generation(root_item, trans->transid);
7541
7542         btrfs_tree_unlock(eb);
7543         free_extent_buffer(eb);
7544
7545         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7546                                 &root_key, root_item);
7547         BUG_ON(ret);
7548         kfree(root_item);
7549
7550         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7551                                                  &root_key);
7552         BUG_ON(!reloc_root);
7553         reloc_root->last_trans = trans->transid;
7554         reloc_root->commit_root = NULL;
7555         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7556
7557         root->reloc_root = reloc_root;
7558         return 0;
7559 }
7560
7561 /*
7562  * Core function of space balance.
7563  *
7564  * The idea is using reloc trees to relocate tree blocks in reference
7565  * counted roots. There is one reloc tree for each subvol, and all
7566  * reloc trees share same root key objectid. Reloc trees are snapshots
7567  * of the latest committed roots of subvols (root->commit_root).
7568  *
7569  * To relocate a tree block referenced by a subvol, there are two steps.
7570  * COW the block through subvol's reloc tree, then update block pointer
7571  * in the subvol to point to the new block. Since all reloc trees share
7572  * same root key objectid, doing special handing for tree blocks owned
7573  * by them is easy. Once a tree block has been COWed in one reloc tree,
7574  * we can use the resulting new block directly when the same block is
7575  * required to COW again through other reloc trees. By this way, relocated
7576  * tree blocks are shared between reloc trees, so they are also shared
7577  * between subvols.
7578  */
7579 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7580                                       struct btrfs_root *root,
7581                                       struct btrfs_path *path,
7582                                       struct btrfs_key *first_key,
7583                                       struct btrfs_ref_path *ref_path,
7584                                       struct btrfs_block_group_cache *group,
7585                                       struct inode *reloc_inode)
7586 {
7587         struct btrfs_root *reloc_root;
7588         struct extent_buffer *eb = NULL;
7589         struct btrfs_key *keys;
7590         u64 *nodes;
7591         int level;
7592         int shared_level;
7593         int lowest_level = 0;
7594         int ret;
7595
7596         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7597                 lowest_level = ref_path->owner_objectid;
7598
7599         if (!root->ref_cows) {
7600                 path->lowest_level = lowest_level;
7601                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7602                 BUG_ON(ret < 0);
7603                 path->lowest_level = 0;
7604                 btrfs_release_path(root, path);
7605                 return 0;
7606         }
7607
7608         mutex_lock(&root->fs_info->tree_reloc_mutex);
7609         ret = init_reloc_tree(trans, root);
7610         BUG_ON(ret);
7611         reloc_root = root->reloc_root;
7612
7613         shared_level = ref_path->shared_level;
7614         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7615
7616         keys = ref_path->node_keys;
7617         nodes = ref_path->new_nodes;
7618         memset(&keys[shared_level + 1], 0,
7619                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7620         memset(&nodes[shared_level + 1], 0,
7621                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7622
7623         if (nodes[lowest_level] == 0) {
7624                 path->lowest_level = lowest_level;
7625                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7626                                         0, 1);
7627                 BUG_ON(ret);
7628                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7629                         eb = path->nodes[level];
7630                         if (!eb || eb == reloc_root->node)
7631                                 break;
7632                         nodes[level] = eb->start;
7633                         if (level == 0)
7634                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7635                         else
7636                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7637                 }
7638                 if (nodes[0] &&
7639                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7640                         eb = path->nodes[0];
7641                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7642                                                       group, reloc_inode);
7643                         BUG_ON(ret);
7644                 }
7645                 btrfs_release_path(reloc_root, path);
7646         } else {
7647                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7648                                        lowest_level);
7649                 BUG_ON(ret);
7650         }
7651
7652         /*
7653          * replace tree blocks in the fs tree with tree blocks in
7654          * the reloc tree.
7655          */
7656         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7657         BUG_ON(ret < 0);
7658
7659         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7660                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7661                                         0, 0);
7662                 BUG_ON(ret);
7663                 extent_buffer_get(path->nodes[0]);
7664                 eb = path->nodes[0];
7665                 btrfs_release_path(reloc_root, path);
7666                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7667                 BUG_ON(ret);
7668                 free_extent_buffer(eb);
7669         }
7670
7671         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7672         path->lowest_level = 0;
7673         return 0;
7674 }
7675
7676 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7677                                         struct btrfs_root *root,
7678                                         struct btrfs_path *path,
7679                                         struct btrfs_key *first_key,
7680                                         struct btrfs_ref_path *ref_path)
7681 {
7682         int ret;
7683
7684         ret = relocate_one_path(trans, root, path, first_key,
7685                                 ref_path, NULL, NULL);
7686         BUG_ON(ret);
7687
7688         return 0;
7689 }
7690
7691 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7692                                     struct btrfs_root *extent_root,
7693                                     struct btrfs_path *path,
7694                                     struct btrfs_key *extent_key)
7695 {
7696         int ret;
7697
7698         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7699         if (ret)
7700                 goto out;
7701         ret = btrfs_del_item(trans, extent_root, path);
7702 out:
7703         btrfs_release_path(extent_root, path);
7704         return ret;
7705 }
7706
7707 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7708                                                 struct btrfs_ref_path *ref_path)
7709 {
7710         struct btrfs_key root_key;
7711
7712         root_key.objectid = ref_path->root_objectid;
7713         root_key.type = BTRFS_ROOT_ITEM_KEY;
7714         if (is_cowonly_root(ref_path->root_objectid))
7715                 root_key.offset = 0;
7716         else
7717                 root_key.offset = (u64)-1;
7718
7719         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7720 }
7721
7722 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7723                                         struct btrfs_path *path,
7724                                         struct btrfs_key *extent_key,
7725                                         struct btrfs_block_group_cache *group,
7726                                         struct inode *reloc_inode, int pass)
7727 {
7728         struct btrfs_trans_handle *trans;
7729         struct btrfs_root *found_root;
7730         struct btrfs_ref_path *ref_path = NULL;
7731         struct disk_extent *new_extents = NULL;
7732         int nr_extents = 0;
7733         int loops;
7734         int ret;
7735         int level;
7736         struct btrfs_key first_key;
7737         u64 prev_block = 0;
7738
7739
7740         trans = btrfs_start_transaction(extent_root, 1);
7741         BUG_ON(!trans);
7742
7743         if (extent_key->objectid == 0) {
7744                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7745                 goto out;
7746         }
7747
7748         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7749         if (!ref_path) {
7750                 ret = -ENOMEM;
7751                 goto out;
7752         }
7753
7754         for (loops = 0; ; loops++) {
7755                 if (loops == 0) {
7756                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7757                                                    extent_key->objectid);
7758                 } else {
7759                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7760                 }
7761                 if (ret < 0)
7762                         goto out;
7763                 if (ret > 0)
7764                         break;
7765
7766                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7767                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7768                         continue;
7769
7770                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7771                 BUG_ON(!found_root);
7772                 /*
7773                  * for reference counted tree, only process reference paths
7774                  * rooted at the latest committed root.
7775                  */
7776                 if (found_root->ref_cows &&
7777                     ref_path->root_generation != found_root->root_key.offset)
7778                         continue;
7779
7780                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7781                         if (pass == 0) {
7782                                 /*
7783                                  * copy data extents to new locations
7784                                  */
7785                                 u64 group_start = group->key.objectid;
7786                                 ret = relocate_data_extent(reloc_inode,
7787                                                            extent_key,
7788                                                            group_start);
7789                                 if (ret < 0)
7790                                         goto out;
7791                                 break;
7792                         }
7793                         level = 0;
7794                 } else {
7795                         level = ref_path->owner_objectid;
7796                 }
7797
7798                 if (prev_block != ref_path->nodes[level]) {
7799                         struct extent_buffer *eb;
7800                         u64 block_start = ref_path->nodes[level];
7801                         u64 block_size = btrfs_level_size(found_root, level);
7802
7803                         eb = read_tree_block(found_root, block_start,
7804                                              block_size, 0);
7805                         btrfs_tree_lock(eb);
7806                         BUG_ON(level != btrfs_header_level(eb));
7807
7808                         if (level == 0)
7809                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7810                         else
7811                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7812
7813                         btrfs_tree_unlock(eb);
7814                         free_extent_buffer(eb);
7815                         prev_block = block_start;
7816                 }
7817
7818                 mutex_lock(&extent_root->fs_info->trans_mutex);
7819                 btrfs_record_root_in_trans(found_root);
7820                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7821                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7822                         /*
7823                          * try to update data extent references while
7824                          * keeping metadata shared between snapshots.
7825                          */
7826                         if (pass == 1) {
7827                                 ret = relocate_one_path(trans, found_root,
7828                                                 path, &first_key, ref_path,
7829                                                 group, reloc_inode);
7830                                 if (ret < 0)
7831                                         goto out;
7832                                 continue;
7833                         }
7834                         /*
7835                          * use fallback method to process the remaining
7836                          * references.
7837                          */
7838                         if (!new_extents) {
7839                                 u64 group_start = group->key.objectid;
7840                                 new_extents = kmalloc(sizeof(*new_extents),
7841                                                       GFP_NOFS);
7842                                 nr_extents = 1;
7843                                 ret = get_new_locations(reloc_inode,
7844                                                         extent_key,
7845                                                         group_start, 1,
7846                                                         &new_extents,
7847                                                         &nr_extents);
7848                                 if (ret)
7849                                         goto out;
7850                         }
7851                         ret = replace_one_extent(trans, found_root,
7852                                                 path, extent_key,
7853                                                 &first_key, ref_path,
7854                                                 new_extents, nr_extents);
7855                 } else {
7856                         ret = relocate_tree_block(trans, found_root, path,
7857                                                   &first_key, ref_path);
7858                 }
7859                 if (ret < 0)
7860                         goto out;
7861         }
7862         ret = 0;
7863 out:
7864         btrfs_end_transaction(trans, extent_root);
7865         kfree(new_extents);
7866         kfree(ref_path);
7867         return ret;
7868 }
7869 #endif
7870
7871 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7872 {
7873         u64 num_devices;
7874         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7875                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7876
7877         num_devices = root->fs_info->fs_devices->rw_devices;
7878         if (num_devices == 1) {
7879                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7880                 stripped = flags & ~stripped;
7881
7882                 /* turn raid0 into single device chunks */
7883                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7884                         return stripped;
7885
7886                 /* turn mirroring into duplication */
7887                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7888                              BTRFS_BLOCK_GROUP_RAID10))
7889                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7890                 return flags;
7891         } else {
7892                 /* they already had raid on here, just return */
7893                 if (flags & stripped)
7894                         return flags;
7895
7896                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7897                 stripped = flags & ~stripped;
7898
7899                 /* switch duplicated blocks with raid1 */
7900                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7901                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7902
7903                 /* turn single device chunks into raid0 */
7904                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7905         }
7906         return flags;
7907 }
7908
7909 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
7910 {
7911         struct btrfs_space_info *sinfo = cache->space_info;
7912         u64 num_bytes;
7913         int ret = -ENOSPC;
7914
7915         if (cache->ro)
7916                 return 0;
7917
7918         spin_lock(&sinfo->lock);
7919         spin_lock(&cache->lock);
7920         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7921                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7922
7923         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7924             sinfo->bytes_may_use + sinfo->bytes_readonly +
7925             cache->reserved_pinned + num_bytes < sinfo->total_bytes) {
7926                 sinfo->bytes_readonly += num_bytes;
7927                 sinfo->bytes_reserved += cache->reserved_pinned;
7928                 cache->reserved_pinned = 0;
7929                 cache->ro = 1;
7930                 ret = 0;
7931         }
7932         spin_unlock(&cache->lock);
7933         spin_unlock(&sinfo->lock);
7934         return ret;
7935 }
7936
7937 int btrfs_set_block_group_ro(struct btrfs_root *root,
7938                              struct btrfs_block_group_cache *cache)
7939
7940 {
7941         struct btrfs_trans_handle *trans;
7942         u64 alloc_flags;
7943         int ret;
7944
7945         BUG_ON(cache->ro);
7946
7947         trans = btrfs_join_transaction(root, 1);
7948         BUG_ON(IS_ERR(trans));
7949
7950         alloc_flags = update_block_group_flags(root, cache->flags);
7951         if (alloc_flags != cache->flags)
7952                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7953
7954         ret = set_block_group_ro(cache);
7955         if (!ret)
7956                 goto out;
7957         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7958         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7959         if (ret < 0)
7960                 goto out;
7961         ret = set_block_group_ro(cache);
7962 out:
7963         btrfs_end_transaction(trans, root);
7964         return ret;
7965 }
7966
7967 int btrfs_set_block_group_rw(struct btrfs_root *root,
7968                               struct btrfs_block_group_cache *cache)
7969 {
7970         struct btrfs_space_info *sinfo = cache->space_info;
7971         u64 num_bytes;
7972
7973         BUG_ON(!cache->ro);
7974
7975         spin_lock(&sinfo->lock);
7976         spin_lock(&cache->lock);
7977         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7978                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7979         sinfo->bytes_readonly -= num_bytes;
7980         cache->ro = 0;
7981         spin_unlock(&cache->lock);
7982         spin_unlock(&sinfo->lock);
7983         return 0;
7984 }
7985
7986 /*
7987  * checks to see if its even possible to relocate this block group.
7988  *
7989  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7990  * ok to go ahead and try.
7991  */
7992 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7993 {
7994         struct btrfs_block_group_cache *block_group;
7995         struct btrfs_space_info *space_info;
7996         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7997         struct btrfs_device *device;
7998         int full = 0;
7999         int ret = 0;
8000
8001         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8002
8003         /* odd, couldn't find the block group, leave it alone */
8004         if (!block_group)
8005                 return -1;
8006
8007         /* no bytes used, we're good */
8008         if (!btrfs_block_group_used(&block_group->item))
8009                 goto out;
8010
8011         space_info = block_group->space_info;
8012         spin_lock(&space_info->lock);
8013
8014         full = space_info->full;
8015
8016         /*
8017          * if this is the last block group we have in this space, we can't
8018          * relocate it unless we're able to allocate a new chunk below.
8019          *
8020          * Otherwise, we need to make sure we have room in the space to handle
8021          * all of the extents from this block group.  If we can, we're good
8022          */
8023         if ((space_info->total_bytes != block_group->key.offset) &&
8024            (space_info->bytes_used + space_info->bytes_reserved +
8025             space_info->bytes_pinned + space_info->bytes_readonly +
8026             btrfs_block_group_used(&block_group->item) <
8027             space_info->total_bytes)) {
8028                 spin_unlock(&space_info->lock);
8029                 goto out;
8030         }
8031         spin_unlock(&space_info->lock);
8032
8033         /*
8034          * ok we don't have enough space, but maybe we have free space on our
8035          * devices to allocate new chunks for relocation, so loop through our
8036          * alloc devices and guess if we have enough space.  However, if we
8037          * were marked as full, then we know there aren't enough chunks, and we
8038          * can just return.
8039          */
8040         ret = -1;
8041         if (full)
8042                 goto out;
8043
8044         mutex_lock(&root->fs_info->chunk_mutex);
8045         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8046                 u64 min_free = btrfs_block_group_used(&block_group->item);
8047                 u64 dev_offset, max_avail;
8048
8049                 /*
8050                  * check to make sure we can actually find a chunk with enough
8051                  * space to fit our block group in.
8052                  */
8053                 if (device->total_bytes > device->bytes_used + min_free) {
8054                         ret = find_free_dev_extent(NULL, device, min_free,
8055                                                    &dev_offset, &max_avail);
8056                         if (!ret)
8057                                 break;
8058                         ret = -1;
8059                 }
8060         }
8061         mutex_unlock(&root->fs_info->chunk_mutex);
8062 out:
8063         btrfs_put_block_group(block_group);
8064         return ret;
8065 }
8066
8067 static int find_first_block_group(struct btrfs_root *root,
8068                 struct btrfs_path *path, struct btrfs_key *key)
8069 {
8070         int ret = 0;
8071         struct btrfs_key found_key;
8072         struct extent_buffer *leaf;
8073         int slot;
8074
8075         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8076         if (ret < 0)
8077                 goto out;
8078
8079         while (1) {
8080                 slot = path->slots[0];
8081                 leaf = path->nodes[0];
8082                 if (slot >= btrfs_header_nritems(leaf)) {
8083                         ret = btrfs_next_leaf(root, path);
8084                         if (ret == 0)
8085                                 continue;
8086                         if (ret < 0)
8087                                 goto out;
8088                         break;
8089                 }
8090                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8091
8092                 if (found_key.objectid >= key->objectid &&
8093                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8094                         ret = 0;
8095                         goto out;
8096                 }
8097                 path->slots[0]++;
8098         }
8099 out:
8100         return ret;
8101 }
8102
8103 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8104 {
8105         struct btrfs_block_group_cache *block_group;
8106         u64 last = 0;
8107
8108         while (1) {
8109                 struct inode *inode;
8110
8111                 block_group = btrfs_lookup_first_block_group(info, last);
8112                 while (block_group) {
8113                         spin_lock(&block_group->lock);
8114                         if (block_group->iref)
8115                                 break;
8116                         spin_unlock(&block_group->lock);
8117                         block_group = next_block_group(info->tree_root,
8118                                                        block_group);
8119                 }
8120                 if (!block_group) {
8121                         if (last == 0)
8122                                 break;
8123                         last = 0;
8124                         continue;
8125                 }
8126
8127                 inode = block_group->inode;
8128                 block_group->iref = 0;
8129                 block_group->inode = NULL;
8130                 spin_unlock(&block_group->lock);
8131                 iput(inode);
8132                 last = block_group->key.objectid + block_group->key.offset;
8133                 btrfs_put_block_group(block_group);
8134         }
8135 }
8136
8137 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8138 {
8139         struct btrfs_block_group_cache *block_group;
8140         struct btrfs_space_info *space_info;
8141         struct btrfs_caching_control *caching_ctl;
8142         struct rb_node *n;
8143
8144         down_write(&info->extent_commit_sem);
8145         while (!list_empty(&info->caching_block_groups)) {
8146                 caching_ctl = list_entry(info->caching_block_groups.next,
8147                                          struct btrfs_caching_control, list);
8148                 list_del(&caching_ctl->list);
8149                 put_caching_control(caching_ctl);
8150         }
8151         up_write(&info->extent_commit_sem);
8152
8153         spin_lock(&info->block_group_cache_lock);
8154         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8155                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8156                                        cache_node);
8157                 rb_erase(&block_group->cache_node,
8158                          &info->block_group_cache_tree);
8159                 spin_unlock(&info->block_group_cache_lock);
8160
8161                 down_write(&block_group->space_info->groups_sem);
8162                 list_del(&block_group->list);
8163                 up_write(&block_group->space_info->groups_sem);
8164
8165                 if (block_group->cached == BTRFS_CACHE_STARTED)
8166                         wait_block_group_cache_done(block_group);
8167
8168                 btrfs_remove_free_space_cache(block_group);
8169                 btrfs_put_block_group(block_group);
8170
8171                 spin_lock(&info->block_group_cache_lock);
8172         }
8173         spin_unlock(&info->block_group_cache_lock);
8174
8175         /* now that all the block groups are freed, go through and
8176          * free all the space_info structs.  This is only called during
8177          * the final stages of unmount, and so we know nobody is
8178          * using them.  We call synchronize_rcu() once before we start,
8179          * just to be on the safe side.
8180          */
8181         synchronize_rcu();
8182
8183         release_global_block_rsv(info);
8184
8185         while(!list_empty(&info->space_info)) {
8186                 space_info = list_entry(info->space_info.next,
8187                                         struct btrfs_space_info,
8188                                         list);
8189                 if (space_info->bytes_pinned > 0 ||
8190                     space_info->bytes_reserved > 0) {
8191                         WARN_ON(1);
8192                         dump_space_info(space_info, 0, 0);
8193                 }
8194                 list_del(&space_info->list);
8195                 kfree(space_info);
8196         }
8197         return 0;
8198 }
8199
8200 static void __link_block_group(struct btrfs_space_info *space_info,
8201                                struct btrfs_block_group_cache *cache)
8202 {
8203         int index = get_block_group_index(cache);
8204
8205         down_write(&space_info->groups_sem);
8206         list_add_tail(&cache->list, &space_info->block_groups[index]);
8207         up_write(&space_info->groups_sem);
8208 }
8209
8210 int btrfs_read_block_groups(struct btrfs_root *root)
8211 {
8212         struct btrfs_path *path;
8213         int ret;
8214         struct btrfs_block_group_cache *cache;
8215         struct btrfs_fs_info *info = root->fs_info;
8216         struct btrfs_space_info *space_info;
8217         struct btrfs_key key;
8218         struct btrfs_key found_key;
8219         struct extent_buffer *leaf;
8220         int need_clear = 0;
8221         u64 cache_gen;
8222
8223         root = info->extent_root;
8224         key.objectid = 0;
8225         key.offset = 0;
8226         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8227         path = btrfs_alloc_path();
8228         if (!path)
8229                 return -ENOMEM;
8230
8231         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8232         if (cache_gen != 0 &&
8233             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8234                 need_clear = 1;
8235         if (btrfs_test_opt(root, CLEAR_CACHE))
8236                 need_clear = 1;
8237         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8238                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8239
8240         while (1) {
8241                 ret = find_first_block_group(root, path, &key);
8242                 if (ret > 0)
8243                         break;
8244                 if (ret != 0)
8245                         goto error;
8246
8247                 leaf = path->nodes[0];
8248                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8249                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8250                 if (!cache) {
8251                         ret = -ENOMEM;
8252                         goto error;
8253                 }
8254
8255                 atomic_set(&cache->count, 1);
8256                 spin_lock_init(&cache->lock);
8257                 spin_lock_init(&cache->tree_lock);
8258                 cache->fs_info = info;
8259                 INIT_LIST_HEAD(&cache->list);
8260                 INIT_LIST_HEAD(&cache->cluster_list);
8261
8262                 if (need_clear)
8263                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8264
8265                 /*
8266                  * we only want to have 32k of ram per block group for keeping
8267                  * track of free space, and if we pass 1/2 of that we want to
8268                  * start converting things over to using bitmaps
8269                  */
8270                 cache->extents_thresh = ((1024 * 32) / 2) /
8271                         sizeof(struct btrfs_free_space);
8272
8273                 read_extent_buffer(leaf, &cache->item,
8274                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8275                                    sizeof(cache->item));
8276                 memcpy(&cache->key, &found_key, sizeof(found_key));
8277
8278                 key.objectid = found_key.objectid + found_key.offset;
8279                 btrfs_release_path(root, path);
8280                 cache->flags = btrfs_block_group_flags(&cache->item);
8281                 cache->sectorsize = root->sectorsize;
8282
8283                 /*
8284                  * check for two cases, either we are full, and therefore
8285                  * don't need to bother with the caching work since we won't
8286                  * find any space, or we are empty, and we can just add all
8287                  * the space in and be done with it.  This saves us _alot_ of
8288                  * time, particularly in the full case.
8289                  */
8290                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8291                         exclude_super_stripes(root, cache);
8292                         cache->last_byte_to_unpin = (u64)-1;
8293                         cache->cached = BTRFS_CACHE_FINISHED;
8294                         free_excluded_extents(root, cache);
8295                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8296                         exclude_super_stripes(root, cache);
8297                         cache->last_byte_to_unpin = (u64)-1;
8298                         cache->cached = BTRFS_CACHE_FINISHED;
8299                         add_new_free_space(cache, root->fs_info,
8300                                            found_key.objectid,
8301                                            found_key.objectid +
8302                                            found_key.offset);
8303                         free_excluded_extents(root, cache);
8304                 }
8305
8306                 ret = update_space_info(info, cache->flags, found_key.offset,
8307                                         btrfs_block_group_used(&cache->item),
8308                                         &space_info);
8309                 BUG_ON(ret);
8310                 cache->space_info = space_info;
8311                 spin_lock(&cache->space_info->lock);
8312                 cache->space_info->bytes_readonly += cache->bytes_super;
8313                 spin_unlock(&cache->space_info->lock);
8314
8315                 __link_block_group(space_info, cache);
8316
8317                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8318                 BUG_ON(ret);
8319
8320                 set_avail_alloc_bits(root->fs_info, cache->flags);
8321                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8322                         set_block_group_ro(cache);
8323         }
8324
8325         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8326                 if (!(get_alloc_profile(root, space_info->flags) &
8327                       (BTRFS_BLOCK_GROUP_RAID10 |
8328                        BTRFS_BLOCK_GROUP_RAID1 |
8329                        BTRFS_BLOCK_GROUP_DUP)))
8330                         continue;
8331                 /*
8332                  * avoid allocating from un-mirrored block group if there are
8333                  * mirrored block groups.
8334                  */
8335                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8336                         set_block_group_ro(cache);
8337                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8338                         set_block_group_ro(cache);
8339         }
8340
8341         init_global_block_rsv(info);
8342         ret = 0;
8343 error:
8344         btrfs_free_path(path);
8345         return ret;
8346 }
8347
8348 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8349                            struct btrfs_root *root, u64 bytes_used,
8350                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8351                            u64 size)
8352 {
8353         int ret;
8354         struct btrfs_root *extent_root;
8355         struct btrfs_block_group_cache *cache;
8356
8357         extent_root = root->fs_info->extent_root;
8358
8359         root->fs_info->last_trans_log_full_commit = trans->transid;
8360
8361         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8362         if (!cache)
8363                 return -ENOMEM;
8364
8365         cache->key.objectid = chunk_offset;
8366         cache->key.offset = size;
8367         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8368         cache->sectorsize = root->sectorsize;
8369         cache->fs_info = root->fs_info;
8370
8371         /*
8372          * we only want to have 32k of ram per block group for keeping track
8373          * of free space, and if we pass 1/2 of that we want to start
8374          * converting things over to using bitmaps
8375          */
8376         cache->extents_thresh = ((1024 * 32) / 2) /
8377                 sizeof(struct btrfs_free_space);
8378         atomic_set(&cache->count, 1);
8379         spin_lock_init(&cache->lock);
8380         spin_lock_init(&cache->tree_lock);
8381         INIT_LIST_HEAD(&cache->list);
8382         INIT_LIST_HEAD(&cache->cluster_list);
8383
8384         btrfs_set_block_group_used(&cache->item, bytes_used);
8385         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8386         cache->flags = type;
8387         btrfs_set_block_group_flags(&cache->item, type);
8388
8389         cache->last_byte_to_unpin = (u64)-1;
8390         cache->cached = BTRFS_CACHE_FINISHED;
8391         exclude_super_stripes(root, cache);
8392
8393         add_new_free_space(cache, root->fs_info, chunk_offset,
8394                            chunk_offset + size);
8395
8396         free_excluded_extents(root, cache);
8397
8398         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8399                                 &cache->space_info);
8400         BUG_ON(ret);
8401
8402         spin_lock(&cache->space_info->lock);
8403         cache->space_info->bytes_readonly += cache->bytes_super;
8404         spin_unlock(&cache->space_info->lock);
8405
8406         __link_block_group(cache->space_info, cache);
8407
8408         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8409         BUG_ON(ret);
8410
8411         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8412                                 sizeof(cache->item));
8413         BUG_ON(ret);
8414
8415         set_avail_alloc_bits(extent_root->fs_info, type);
8416
8417         return 0;
8418 }
8419
8420 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8421                              struct btrfs_root *root, u64 group_start)
8422 {
8423         struct btrfs_path *path;
8424         struct btrfs_block_group_cache *block_group;
8425         struct btrfs_free_cluster *cluster;
8426         struct btrfs_root *tree_root = root->fs_info->tree_root;
8427         struct btrfs_key key;
8428         struct inode *inode;
8429         int ret;
8430         int factor;
8431
8432         root = root->fs_info->extent_root;
8433
8434         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8435         BUG_ON(!block_group);
8436         BUG_ON(!block_group->ro);
8437
8438         memcpy(&key, &block_group->key, sizeof(key));
8439         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8440                                   BTRFS_BLOCK_GROUP_RAID1 |
8441                                   BTRFS_BLOCK_GROUP_RAID10))
8442                 factor = 2;
8443         else
8444                 factor = 1;
8445
8446         /* make sure this block group isn't part of an allocation cluster */
8447         cluster = &root->fs_info->data_alloc_cluster;
8448         spin_lock(&cluster->refill_lock);
8449         btrfs_return_cluster_to_free_space(block_group, cluster);
8450         spin_unlock(&cluster->refill_lock);
8451
8452         /*
8453          * make sure this block group isn't part of a metadata
8454          * allocation cluster
8455          */
8456         cluster = &root->fs_info->meta_alloc_cluster;
8457         spin_lock(&cluster->refill_lock);
8458         btrfs_return_cluster_to_free_space(block_group, cluster);
8459         spin_unlock(&cluster->refill_lock);
8460
8461         path = btrfs_alloc_path();
8462         BUG_ON(!path);
8463
8464         inode = lookup_free_space_inode(root, block_group, path);
8465         if (!IS_ERR(inode)) {
8466                 btrfs_orphan_add(trans, inode);
8467                 clear_nlink(inode);
8468                 /* One for the block groups ref */
8469                 spin_lock(&block_group->lock);
8470                 if (block_group->iref) {
8471                         block_group->iref = 0;
8472                         block_group->inode = NULL;
8473                         spin_unlock(&block_group->lock);
8474                         iput(inode);
8475                 } else {
8476                         spin_unlock(&block_group->lock);
8477                 }
8478                 /* One for our lookup ref */
8479                 iput(inode);
8480         }
8481
8482         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8483         key.offset = block_group->key.objectid;
8484         key.type = 0;
8485
8486         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8487         if (ret < 0)
8488                 goto out;
8489         if (ret > 0)
8490                 btrfs_release_path(tree_root, path);
8491         if (ret == 0) {
8492                 ret = btrfs_del_item(trans, tree_root, path);
8493                 if (ret)
8494                         goto out;
8495                 btrfs_release_path(tree_root, path);
8496         }
8497
8498         spin_lock(&root->fs_info->block_group_cache_lock);
8499         rb_erase(&block_group->cache_node,
8500                  &root->fs_info->block_group_cache_tree);
8501         spin_unlock(&root->fs_info->block_group_cache_lock);
8502
8503         down_write(&block_group->space_info->groups_sem);
8504         /*
8505          * we must use list_del_init so people can check to see if they
8506          * are still on the list after taking the semaphore
8507          */
8508         list_del_init(&block_group->list);
8509         up_write(&block_group->space_info->groups_sem);
8510
8511         if (block_group->cached == BTRFS_CACHE_STARTED)
8512                 wait_block_group_cache_done(block_group);
8513
8514         btrfs_remove_free_space_cache(block_group);
8515
8516         spin_lock(&block_group->space_info->lock);
8517         block_group->space_info->total_bytes -= block_group->key.offset;
8518         block_group->space_info->bytes_readonly -= block_group->key.offset;
8519         block_group->space_info->disk_total -= block_group->key.offset * factor;
8520         spin_unlock(&block_group->space_info->lock);
8521
8522         memcpy(&key, &block_group->key, sizeof(key));
8523
8524         btrfs_clear_space_info_full(root->fs_info);
8525
8526         btrfs_put_block_group(block_group);
8527         btrfs_put_block_group(block_group);
8528
8529         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8530         if (ret > 0)
8531                 ret = -EIO;
8532         if (ret < 0)
8533                 goto out;
8534
8535         ret = btrfs_del_item(trans, root, path);
8536 out:
8537         btrfs_free_path(path);
8538         return ret;
8539 }