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         ctl = cache->caching_ctl;
 246         atomic_inc(&ctl->count);
 247         spin_unlock(&cache->lock);
 248         return ctl;
 249 }
 250
 251 static void put_caching_control(struct btrfs_caching_control *ctl)
 252 {
 253         if (atomic_dec_and_test(&ctl->count))
 254                 kfree(ctl);
 255 }
 256
 257 /*
 258  * this is only called by cache_block_group, since we could have freed extents
 259  * we need to check the pinned_extents for any extents that can't be used yet
 260  * since their free space will be released as soon as the transaction commits.
 261  */
 262 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 263                               struct btrfs_fs_info *info, u64 start, u64 end)
 264 {
 265         u64 extent_start, extent_end, size, total_added = 0;
 266         int ret;
 267
 268         while (start < end) {
 269                 ret = find_first_extent_bit(info->pinned_extents, start,
 270                                             &extent_start, &extent_end,
 271                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 272                 if (ret)
 273                         break;
 274
 275                 if (extent_start <= start) {
 276                         start = extent_end + 1;
 277                 } else if (extent_start > start && extent_start < end) {
 278                         size = extent_start - start;
 279                         total_added += size;
 280                         ret = btrfs_add_free_space(block_group, start,
 281                                                    size);
 282                         BUG_ON(ret);
 283                         start = extent_end + 1;
 284                 } else {
 285                         break;
 286                 }
 287         }
 288
 289         if (start < end) {
 290                 size = end - start;
 291                 total_added += size;
 292                 ret = btrfs_add_free_space(block_group, start, size);
 293                 BUG_ON(ret);
 294         }
 295
 296         return total_added;
 297 }
 298
 299 static int caching_kthread(void *data)
 300 {
 301         struct btrfs_block_group_cache *block_group = data;
 302         struct btrfs_fs_info *fs_info = block_group->fs_info;
 303         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 304         struct btrfs_root *extent_root = fs_info->extent_root;
 305         struct btrfs_path *path;
 306         struct extent_buffer *leaf;
 307         struct btrfs_key key;
 308         u64 total_found = 0;
 309         u64 last = 0;
 310         u32 nritems;
 311         int ret = 0;
 312
 313         path = btrfs_alloc_path();
 314         if (!path)
 315                 return -ENOMEM;
 316
 317         exclude_super_stripes(extent_root, block_group);
 318         spin_lock(&block_group->space_info->lock);
 319         block_group->space_info->bytes_readonly += block_group->bytes_super;
 320         spin_unlock(&block_group->space_info->lock);
 321
 322         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 323
 324         /*
 325          * We don't want to deadlock with somebody trying to allocate a new
 326          * extent for the extent root while also trying to search the extent
 327          * root to add free space.  So we skip locking and search the commit
 328          * root, since its read-only
 329          */
 330         path->skip_locking = 1;
 331         path->search_commit_root = 1;
 332         path->reada = 2;
 333
 334         key.objectid = last;
 335         key.offset = 0;
 336         key.type = BTRFS_EXTENT_ITEM_KEY;
 337 again:
 338         mutex_lock(&caching_ctl->mutex);
 339         /* need to make sure the commit_root doesn't disappear */
 340         down_read(&fs_info->extent_commit_sem);
 341
 342         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 343         if (ret < 0)
 344                 goto err;
 345
 346         leaf = path->nodes[0];
 347         nritems = btrfs_header_nritems(leaf);
 348
 349         while (1) {
 350                 smp_mb();
 351                 if (fs_info->closing > 1) {
 352                         last = (u64)-1;
 353                         break;
 354                 }
 355
 356                 if (path->slots[0] < nritems) {
 357                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 358                 } else {
 359                         ret = find_next_key(path, 0, &key);
 360                         if (ret)
 361                                 break;
 362
 363                         caching_ctl->progress = last;
 364                         btrfs_release_path(extent_root, path);
 365                         up_read(&fs_info->extent_commit_sem);
 366                         mutex_unlock(&caching_ctl->mutex);
 367                         if (btrfs_transaction_in_commit(fs_info))
 368                                 schedule_timeout(1);
 369                         else
 370                                 cond_resched();
 371                         goto again;
 372                 }
 373
 374                 if (key.objectid < block_group->key.objectid) {
 375                         path->slots[0]++;
 376                         continue;
 377                 }
 378
 379                 if (key.objectid >= block_group->key.objectid +
 380                     block_group->key.offset)
 381                         break;
 382
 383                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 384                         total_found += add_new_free_space(block_group,
 385                                                           fs_info, last,
 386                                                           key.objectid);
 387                         last = key.objectid + key.offset;
 388
 389                         if (total_found > (1024 * 1024 * 2)) {
 390                                 total_found = 0;
 391                                 wake_up(&caching_ctl->wait);
 392                         }
 393                 }
 394                 path->slots[0]++;
 395         }
 396         ret = 0;
 397
 398         total_found += add_new_free_space(block_group, fs_info, last,
 399                                           block_group->key.objectid +
 400                                           block_group->key.offset);
 401         caching_ctl->progress = (u64)-1;
 402
 403         spin_lock(&block_group->lock);
 404         block_group->caching_ctl = NULL;
 405         block_group->cached = BTRFS_CACHE_FINISHED;
 406         spin_unlock(&block_group->lock);
 407
 408 err:
 409         btrfs_free_path(path);
 410         up_read(&fs_info->extent_commit_sem);
 411
 412         free_excluded_extents(extent_root, block_group);
 413
 414         mutex_unlock(&caching_ctl->mutex);
 415         wake_up(&caching_ctl->wait);
 416
 417         put_caching_control(caching_ctl);
 418         atomic_dec(&block_group->space_info->caching_threads);
 419         btrfs_put_block_group(block_group);
 420
 421         return 0;
 422 }
 423
 424 static int cache_block_group(struct btrfs_block_group_cache *cache)
 425 {
 426         struct btrfs_fs_info *fs_info = cache->fs_info;
 427         struct btrfs_caching_control *caching_ctl;
 428         struct task_struct *tsk;
 429         int ret = 0;
 430
 431         smp_mb();
 432         if (cache->cached != BTRFS_CACHE_NO)
 433                 return 0;
 434
 435         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
 436         BUG_ON(!caching_ctl);
 437
 438         INIT_LIST_HEAD(&caching_ctl->list);
 439         mutex_init(&caching_ctl->mutex);
 440         init_waitqueue_head(&caching_ctl->wait);
 441         caching_ctl->block_group = cache;
 442         caching_ctl->progress = cache->key.objectid;
 443         /* one for caching kthread, one for caching block group list */
 444         atomic_set(&caching_ctl->count, 2);
 445
 446         spin_lock(&cache->lock);
 447         if (cache->cached != BTRFS_CACHE_NO) {
 448                 spin_unlock(&cache->lock);
 449                 kfree(caching_ctl);
 450                 return 0;
 451         }
 452         cache->caching_ctl = caching_ctl;
 453         cache->cached = BTRFS_CACHE_STARTED;
 454         spin_unlock(&cache->lock);
 455
 456         down_write(&fs_info->extent_commit_sem);
 457         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 458         up_write(&fs_info->extent_commit_sem);
 459
 460         atomic_inc(&cache->space_info->caching_threads);
 461         btrfs_get_block_group(cache);
 462
 463         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 464                           cache->key.objectid);
 465         if (IS_ERR(tsk)) {
 466                 ret = PTR_ERR(tsk);
 467                 printk(KERN_ERR "error running thread %d\n", ret);
 468                 BUG();
 469         }
 470
 471         return ret;
 472 }
 473
 474 /*
 475  * return the block group that starts at or after bytenr
 476  */
 477 static struct btrfs_block_group_cache *
 478 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 479 {
 480         struct btrfs_block_group_cache *cache;
 481
 482         cache = block_group_cache_tree_search(info, bytenr, 0);
 483
 484         return cache;
 485 }
 486
 487 /*
 488  * return the block group that contains the given bytenr
 489  */
 490 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 491                                                  struct btrfs_fs_info *info,
 492                                                  u64 bytenr)
 493 {
 494         struct btrfs_block_group_cache *cache;
 495
 496         cache = block_group_cache_tree_search(info, bytenr, 1);
 497
 498         return cache;
 499 }
 500
 501 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 502                                                   u64 flags)
 503 {
 504         struct list_head *head = &info->space_info;
 505         struct btrfs_space_info *found;
 506
 507         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 508                  BTRFS_BLOCK_GROUP_METADATA;
 509
 510         rcu_read_lock();
 511         list_for_each_entry_rcu(found, head, list) {
 512                 if (found->flags == flags) {
 513                         rcu_read_unlock();
 514                         return found;
 515                 }
 516         }
 517         rcu_read_unlock();
 518         return NULL;
 519 }
 520
 521 /*
 522  * after adding space to the filesystem, we need to clear the full flags
 523  * on all the space infos.
 524  */
 525 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 526 {
 527         struct list_head *head = &info->space_info;
 528         struct btrfs_space_info *found;
 529
 530         rcu_read_lock();
 531         list_for_each_entry_rcu(found, head, list)
 532                 found->full = 0;
 533         rcu_read_unlock();
 534 }
 535
 536 static u64 div_factor(u64 num, int factor)
 537 {
 538         if (factor == 10)
 539                 return num;
 540         num *= factor;
 541         do_div(num, 10);
 542         return num;
 543 }
 544
 545 u64 btrfs_find_block_group(struct btrfs_root *root,
 546                            u64 search_start, u64 search_hint, int owner)
 547 {
 548         struct btrfs_block_group_cache *cache;
 549         u64 used;
 550         u64 last = max(search_hint, search_start);
 551         u64 group_start = 0;
 552         int full_search = 0;
 553         int factor = 9;
 554         int wrapped = 0;
 555 again:
 556         while (1) {
 557                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 558                 if (!cache)
 559                         break;
 560
 561                 spin_lock(&cache->lock);
 562                 last = cache->key.objectid + cache->key.offset;
 563                 used = btrfs_block_group_used(&cache->item);
 564
 565                 if ((full_search || !cache->ro) &&
 566                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 567                         if (used + cache->pinned + cache->reserved <
 568                             div_factor(cache->key.offset, factor)) {
 569                                 group_start = cache->key.objectid;
 570                                 spin_unlock(&cache->lock);
 571                                 btrfs_put_block_group(cache);
 572                                 goto found;
 573                         }
 574                 }
 575                 spin_unlock(&cache->lock);
 576                 btrfs_put_block_group(cache);
 577                 cond_resched();
 578         }
 579         if (!wrapped) {
 580                 last = search_start;
 581                 wrapped = 1;
 582                 goto again;
 583         }
 584         if (!full_search && factor < 10) {
 585                 last = search_start;
 586                 full_search = 1;
 587                 factor = 10;
 588                 goto again;
 589         }
 590 found:
 591         return group_start;
 592 }
 593
 594 /* simple helper to search for an existing extent at a given offset */
 595 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 596 {
 597         int ret;
 598         struct btrfs_key key;
 599         struct btrfs_path *path;
 600
 601         path = btrfs_alloc_path();
 602         BUG_ON(!path);
 603         key.objectid = start;
 604         key.offset = len;
 605         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 606         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 607                                 0, 0);
 608         btrfs_free_path(path);
 609         return ret;
 610 }
 611
 612 /*
 613  * helper function to lookup reference count and flags of extent.
 614  *
 615  * the head node for delayed ref is used to store the sum of all the
 616  * reference count modifications queued up in the rbtree. the head
 617  * node may also store the extent flags to set. This way you can check
 618  * to see what the reference count and extent flags would be if all of
 619  * the delayed refs are not processed.
 620  */
 621 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 622                              struct btrfs_root *root, u64 bytenr,
 623                              u64 num_bytes, u64 *refs, u64 *flags)
 624 {
 625         struct btrfs_delayed_ref_head *head;
 626         struct btrfs_delayed_ref_root *delayed_refs;
 627         struct btrfs_path *path;
 628         struct btrfs_extent_item *ei;
 629         struct extent_buffer *leaf;
 630         struct btrfs_key key;
 631         u32 item_size;
 632         u64 num_refs;
 633         u64 extent_flags;
 634         int ret;
 635
 636         path = btrfs_alloc_path();
 637         if (!path)
 638                 return -ENOMEM;
 639
 640         key.objectid = bytenr;
 641         key.type = BTRFS_EXTENT_ITEM_KEY;
 642         key.offset = num_bytes;
 643         if (!trans) {
 644                 path->skip_locking = 1;
 645                 path->search_commit_root = 1;
 646         }
 647 again:
 648         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 649                                 &key, path, 0, 0);
 650         if (ret < 0)
 651                 goto out_free;
 652
 653         if (ret == 0) {
 654                 leaf = path->nodes[0];
 655                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 656                 if (item_size >= sizeof(*ei)) {
 657                         ei = btrfs_item_ptr(leaf, path->slots[0],
 658                                             struct btrfs_extent_item);
 659                         num_refs = btrfs_extent_refs(leaf, ei);
 660                         extent_flags = btrfs_extent_flags(leaf, ei);
 661                 } else {
 662 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 663                         struct btrfs_extent_item_v0 *ei0;
 664                         BUG_ON(item_size != sizeof(*ei0));
 665                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 666                                              struct btrfs_extent_item_v0);
 667                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 668                         /* FIXME: this isn't correct for data */
 669                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 670 #else
 671                         BUG();
 672 #endif
 673                 }
 674                 BUG_ON(num_refs == 0);
 675         } else {
 676                 num_refs = 0;
 677                 extent_flags = 0;
 678                 ret = 0;
 679         }
 680
 681         if (!trans)
 682                 goto out;
 683
 684         delayed_refs = &trans->transaction->delayed_refs;
 685         spin_lock(&delayed_refs->lock);
 686         head = btrfs_find_delayed_ref_head(trans, bytenr);
 687         if (head) {
 688                 if (!mutex_trylock(&head->mutex)) {
 689                         atomic_inc(&head->node.refs);
 690                         spin_unlock(&delayed_refs->lock);
 691
 692                         btrfs_release_path(root->fs_info->extent_root, path);
 693
 694                         mutex_lock(&head->mutex);
 695                         mutex_unlock(&head->mutex);
 696                         btrfs_put_delayed_ref(&head->node);
 697                         goto again;
 698                 }
 699                 if (head->extent_op && head->extent_op->update_flags)
 700                         extent_flags |= head->extent_op->flags_to_set;
 701                 else
 702                         BUG_ON(num_refs == 0);
 703
 704                 num_refs += head->node.ref_mod;
 705                 mutex_unlock(&head->mutex);
 706         }
 707         spin_unlock(&delayed_refs->lock);
 708 out:
 709         WARN_ON(num_refs == 0);
 710         if (refs)
 711                 *refs = num_refs;
 712         if (flags)
 713                 *flags = extent_flags;
 714 out_free:
 715         btrfs_free_path(path);
 716         return ret;
 717 }
 718
 719 /*
 720  * Back reference rules.  Back refs have three main goals:
 721  *
 722  * 1) differentiate between all holders of references to an extent so that
 723  *    when a reference is dropped we can make sure it was a valid reference
 724  *    before freeing the extent.
 725  *
 726  * 2) Provide enough information to quickly find the holders of an extent
 727  *    if we notice a given block is corrupted or bad.
 728  *
 729  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 730  *    maintenance.  This is actually the same as #2, but with a slightly
 731  *    different use case.
 732  *
 733  * There are two kinds of back refs. The implicit back refs is optimized
 734  * for pointers in non-shared tree blocks. For a given pointer in a block,
 735  * back refs of this kind provide information about the block's owner tree
 736  * and the pointer's key. These information allow us to find the block by
 737  * b-tree searching. The full back refs is for pointers in tree blocks not
 738  * referenced by their owner trees. The location of tree block is recorded
 739  * in the back refs. Actually the full back refs is generic, and can be
 740  * used in all cases the implicit back refs is used. The major shortcoming
 741  * of the full back refs is its overhead. Every time a tree block gets
 742  * COWed, we have to update back refs entry for all pointers in it.
 743  *
 744  * For a newly allocated tree block, we use implicit back refs for
 745  * pointers in it. This means most tree related operations only involve
 746  * implicit back refs. For a tree block created in old transaction, the
 747  * only way to drop a reference to it is COW it. So we can detect the
 748  * event that tree block loses its owner tree's reference and do the
 749  * back refs conversion.
 750  *
 751  * When a tree block is COW'd through a tree, there are four cases:
 752  *
 753  * The reference count of the block is one and the tree is the block's
 754  * owner tree. Nothing to do in this case.
 755  *
 756  * The reference count of the block is one and the tree is not the
 757  * block's owner tree. In this case, full back refs is used for pointers
 758  * in the block. Remove these full back refs, add implicit back refs for
 759  * every pointers in the new block.
 760  *
 761  * The reference count of the block is greater than one and the tree is
 762  * the block's owner tree. In this case, implicit back refs is used for
 763  * pointers in the block. Add full back refs for every pointers in the
 764  * block, increase lower level extents' reference counts. The original
 765  * implicit back refs are entailed to the new block.
 766  *
 767  * The reference count of the block is greater than one and the tree is
 768  * not the block's owner tree. Add implicit back refs for every pointer in
 769  * the new block, increase lower level extents' reference count.
 770  *
 771  * Back Reference Key composing:
 772  *
 773  * The key objectid corresponds to the first byte in the extent,
 774  * The key type is used to differentiate between types of back refs.
 775  * There are different meanings of the key offset for different types
 776  * of back refs.
 777  *
 778  * File extents can be referenced by:
 779  *
 780  * - multiple snapshots, subvolumes, or different generations in one subvol
 781  * - different files inside a single subvolume
 782  * - different offsets inside a file (bookend extents in file.c)
 783  *
 784  * The extent ref structure for the implicit back refs has fields for:
 785  *
 786  * - Objectid of the subvolume root
 787  * - objectid of the file holding the reference
 788  * - original offset in the file
 789  * - how many bookend extents
 790  *
 791  * The key offset for the implicit back refs is hash of the first
 792  * three fields.
 793  *
 794  * The extent ref structure for the full back refs has field for:
 795  *
 796  * - number of pointers in the tree leaf
 797  *
 798  * The key offset for the implicit back refs is the first byte of
 799  * the tree leaf
 800  *
 801  * When a file extent is allocated, The implicit back refs is used.
 802  * the fields are filled in:
 803  *
 804  *     (root_key.objectid, inode objectid, offset in file, 1)
 805  *
 806  * When a file extent is removed file truncation, we find the
 807  * corresponding implicit back refs and check the following fields:
 808  *
 809  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 810  *
 811  * Btree extents can be referenced by:
 812  *
 813  * - Different subvolumes
 814  *
 815  * Both the implicit back refs and the full back refs for tree blocks
 816  * only consist of key. The key offset for the implicit back refs is
 817  * objectid of block's owner tree. The key offset for the full back refs
 818  * is the first byte of parent block.
 819  *
 820  * When implicit back refs is used, information about the lowest key and
 821  * level of the tree block are required. These information are stored in
 822  * tree block info structure.
 823  */
 824
 825 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 826 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 827                                   struct btrfs_root *root,
 828                                   struct btrfs_path *path,
 829                                   u64 owner, u32 extra_size)
 830 {
 831         struct btrfs_extent_item *item;
 832         struct btrfs_extent_item_v0 *ei0;
 833         struct btrfs_extent_ref_v0 *ref0;
 834         struct btrfs_tree_block_info *bi;
 835         struct extent_buffer *leaf;
 836         struct btrfs_key key;
 837         struct btrfs_key found_key;
 838         u32 new_size = sizeof(*item);
 839         u64 refs;
 840         int ret;
 841
 842         leaf = path->nodes[0];
 843         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 844
 845         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 846         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 847                              struct btrfs_extent_item_v0);
 848         refs = btrfs_extent_refs_v0(leaf, ei0);
 849
 850         if (owner == (u64)-1) {
 851                 while (1) {
 852                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 853                                 ret = btrfs_next_leaf(root, path);
 854                                 if (ret < 0)
 855                                         return ret;
 856                                 BUG_ON(ret > 0);
 857                                 leaf = path->nodes[0];
 858                         }
 859                         btrfs_item_key_to_cpu(leaf, &found_key,
 860                                               path->slots[0]);
 861                         BUG_ON(key.objectid != found_key.objectid);
 862                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 863                                 path->slots[0]++;
 864                                 continue;
 865                         }
 866                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 867                                               struct btrfs_extent_ref_v0);
 868                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 869                         break;
 870                 }
 871         }
 872         btrfs_release_path(root, path);
 873
 874         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 875                 new_size += sizeof(*bi);
 876
 877         new_size -= sizeof(*ei0);
 878         ret = btrfs_search_slot(trans, root, &key, path,
 879                                 new_size + extra_size, 1);
 880         if (ret < 0)
 881                 return ret;
 882         BUG_ON(ret);
 883
 884         ret = btrfs_extend_item(trans, root, path, new_size);
 885         BUG_ON(ret);
 886
 887         leaf = path->nodes[0];
 888         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 889         btrfs_set_extent_refs(leaf, item, refs);
 890         /* FIXME: get real generation */
 891         btrfs_set_extent_generation(leaf, item, 0);
 892         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 893                 btrfs_set_extent_flags(leaf, item,
 894                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 895                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 896                 bi = (struct btrfs_tree_block_info *)(item + 1);
 897                 /* FIXME: get first key of the block */
 898                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 899                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 900         } else {
 901                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 902         }
 903         btrfs_mark_buffer_dirty(leaf);
 904         return 0;
 905 }
 906 #endif
 907
 908 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 909 {
 910         u32 high_crc = ~(u32)0;
 911         u32 low_crc = ~(u32)0;
 912         __le64 lenum;
 913
 914         lenum = cpu_to_le64(root_objectid);
 915         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 916         lenum = cpu_to_le64(owner);
 917         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 918         lenum = cpu_to_le64(offset);
 919         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 920
 921         return ((u64)high_crc << 31) ^ (u64)low_crc;
 922 }
 923
 924 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 925                                      struct btrfs_extent_data_ref *ref)
 926 {
 927         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 928                                     btrfs_extent_data_ref_objectid(leaf, ref),
 929                                     btrfs_extent_data_ref_offset(leaf, ref));
 930 }
 931
 932 static int match_extent_data_ref(struct extent_buffer *leaf,
 933                                  struct btrfs_extent_data_ref *ref,
 934                                  u64 root_objectid, u64 owner, u64 offset)
 935 {
 936         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 937             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 938             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 939                 return 0;
 940         return 1;
 941 }
 942
 943 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 944                                            struct btrfs_root *root,
 945                                            struct btrfs_path *path,
 946                                            u64 bytenr, u64 parent,
 947                                            u64 root_objectid,
 948                                            u64 owner, u64 offset)
 949 {
 950         struct btrfs_key key;
 951         struct btrfs_extent_data_ref *ref;
 952         struct extent_buffer *leaf;
 953         u32 nritems;
 954         int ret;
 955         int recow;
 956         int err = -ENOENT;
 957
 958         key.objectid = bytenr;
 959         if (parent) {
 960                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 961                 key.offset = parent;
 962         } else {
 963                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 964                 key.offset = hash_extent_data_ref(root_objectid,
 965                                                   owner, offset);
 966         }
 967 again:
 968         recow = 0;
 969         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 970         if (ret < 0) {
 971                 err = ret;
 972                 goto fail;
 973         }
 974
 975         if (parent) {
 976                 if (!ret)
 977                         return 0;
 978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 979                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 980                 btrfs_release_path(root, path);
 981                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 982                 if (ret < 0) {
 983                         err = ret;
 984                         goto fail;
 985                 }
 986                 if (!ret)
 987                         return 0;
 988 #endif
 989                 goto fail;
 990         }
 991
 992         leaf = path->nodes[0];
 993         nritems = btrfs_header_nritems(leaf);
 994         while (1) {
 995                 if (path->slots[0] >= nritems) {
 996                         ret = btrfs_next_leaf(root, path);
 997                         if (ret < 0)
 998                                 err = ret;
 999                         if (ret)
1000                                 goto fail;
1001
1002                         leaf = path->nodes[0];
1003                         nritems = btrfs_header_nritems(leaf);
1004                         recow = 1;
1005                 }
1006
1007                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1008                 if (key.objectid != bytenr ||
1009                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1010                         goto fail;
1011
1012                 ref = btrfs_item_ptr(leaf, path->slots[0],
1013                                      struct btrfs_extent_data_ref);
1014
1015                 if (match_extent_data_ref(leaf, ref, root_objectid,
1016                                           owner, offset)) {
1017                         if (recow) {
1018                                 btrfs_release_path(root, path);
1019                                 goto again;
1020                         }
1021                         err = 0;
1022                         break;
1023                 }
1024                 path->slots[0]++;
1025         }
1026 fail:
1027         return err;
1028 }
1029
1030 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1031                                            struct btrfs_root *root,
1032                                            struct btrfs_path *path,
1033                                            u64 bytenr, u64 parent,
1034                                            u64 root_objectid, u64 owner,
1035                                            u64 offset, int refs_to_add)
1036 {
1037         struct btrfs_key key;
1038         struct extent_buffer *leaf;
1039         u32 size;
1040         u32 num_refs;
1041         int ret;
1042
1043         key.objectid = bytenr;
1044         if (parent) {
1045                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1046                 key.offset = parent;
1047                 size = sizeof(struct btrfs_shared_data_ref);
1048         } else {
1049                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1050                 key.offset = hash_extent_data_ref(root_objectid,
1051                                                   owner, offset);
1052                 size = sizeof(struct btrfs_extent_data_ref);
1053         }
1054
1055         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1056         if (ret && ret != -EEXIST)
1057                 goto fail;
1058
1059         leaf = path->nodes[0];
1060         if (parent) {
1061                 struct btrfs_shared_data_ref *ref;
1062                 ref = btrfs_item_ptr(leaf, path->slots[0],
1063                                      struct btrfs_shared_data_ref);
1064                 if (ret == 0) {
1065                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1066                 } else {
1067                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1068                         num_refs += refs_to_add;
1069                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1070                 }
1071         } else {
1072                 struct btrfs_extent_data_ref *ref;
1073                 while (ret == -EEXIST) {
1074                         ref = btrfs_item_ptr(leaf, path->slots[0],
1075                                              struct btrfs_extent_data_ref);
1076                         if (match_extent_data_ref(leaf, ref, root_objectid,
1077                                                   owner, offset))
1078                                 break;
1079                         btrfs_release_path(root, path);
1080                         key.offset++;
1081                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1082                                                       size);
1083                         if (ret && ret != -EEXIST)
1084                                 goto fail;
1085
1086                         leaf = path->nodes[0];
1087                 }
1088                 ref = btrfs_item_ptr(leaf, path->slots[0],
1089                                      struct btrfs_extent_data_ref);
1090                 if (ret == 0) {
1091                         btrfs_set_extent_data_ref_root(leaf, ref,
1092                                                        root_objectid);
1093                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1094                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1095                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1096                 } else {
1097                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1098                         num_refs += refs_to_add;
1099                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1100                 }
1101         }
1102         btrfs_mark_buffer_dirty(leaf);
1103         ret = 0;
1104 fail:
1105         btrfs_release_path(root, path);
1106         return ret;
1107 }
1108
1109 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1110                                            struct btrfs_root *root,
1111                                            struct btrfs_path *path,
1112                                            int refs_to_drop)
1113 {
1114         struct btrfs_key key;
1115         struct btrfs_extent_data_ref *ref1 = NULL;
1116         struct btrfs_shared_data_ref *ref2 = NULL;
1117         struct extent_buffer *leaf;
1118         u32 num_refs = 0;
1119         int ret = 0;
1120
1121         leaf = path->nodes[0];
1122         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1123
1124         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1125                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1126                                       struct btrfs_extent_data_ref);
1127                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1128         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1129                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1130                                       struct btrfs_shared_data_ref);
1131                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1132 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1133         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1134                 struct btrfs_extent_ref_v0 *ref0;
1135                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1136                                       struct btrfs_extent_ref_v0);
1137                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1138 #endif
1139         } else {
1140                 BUG();
1141         }
1142
1143         BUG_ON(num_refs < refs_to_drop);
1144         num_refs -= refs_to_drop;
1145
1146         if (num_refs == 0) {
1147                 ret = btrfs_del_item(trans, root, path);
1148         } else {
1149                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1150                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1151                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1152                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1153 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1154                 else {
1155                         struct btrfs_extent_ref_v0 *ref0;
1156                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1157                                         struct btrfs_extent_ref_v0);
1158                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1159                 }
1160 #endif
1161                 btrfs_mark_buffer_dirty(leaf);
1162         }
1163         return ret;
1164 }
1165
1166 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1167                                           struct btrfs_path *path,
1168                                           struct btrfs_extent_inline_ref *iref)
1169 {
1170         struct btrfs_key key;
1171         struct extent_buffer *leaf;
1172         struct btrfs_extent_data_ref *ref1;
1173         struct btrfs_shared_data_ref *ref2;
1174         u32 num_refs = 0;
1175
1176         leaf = path->nodes[0];
1177         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1178         if (iref) {
1179                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1180                     BTRFS_EXTENT_DATA_REF_KEY) {
1181                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1182                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1183                 } else {
1184                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1185                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1186                 }
1187         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1188                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1189                                       struct btrfs_extent_data_ref);
1190                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1191         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1192                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1193                                       struct btrfs_shared_data_ref);
1194                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1195 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1196         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1197                 struct btrfs_extent_ref_v0 *ref0;
1198                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1199                                       struct btrfs_extent_ref_v0);
1200                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1201 #endif
1202         } else {
1203                 WARN_ON(1);
1204         }
1205         return num_refs;
1206 }
1207
1208 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1209                                           struct btrfs_root *root,
1210                                           struct btrfs_path *path,
1211                                           u64 bytenr, u64 parent,
1212                                           u64 root_objectid)
1213 {
1214         struct btrfs_key key;
1215         int ret;
1216
1217         key.objectid = bytenr;
1218         if (parent) {
1219                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1220                 key.offset = parent;
1221         } else {
1222                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1223                 key.offset = root_objectid;
1224         }
1225
1226         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1227         if (ret > 0)
1228                 ret = -ENOENT;
1229 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1230         if (ret == -ENOENT && parent) {
1231                 btrfs_release_path(root, path);
1232                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1233                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1234                 if (ret > 0)
1235                         ret = -ENOENT;
1236         }
1237 #endif
1238         return ret;
1239 }
1240
1241 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1242                                           struct btrfs_root *root,
1243                                           struct btrfs_path *path,
1244                                           u64 bytenr, u64 parent,
1245                                           u64 root_objectid)
1246 {
1247         struct btrfs_key key;
1248         int ret;
1249
1250         key.objectid = bytenr;
1251         if (parent) {
1252                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1253                 key.offset = parent;
1254         } else {
1255                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1256                 key.offset = root_objectid;
1257         }
1258
1259         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1260         btrfs_release_path(root, path);
1261         return ret;
1262 }
1263
1264 static inline int extent_ref_type(u64 parent, u64 owner)
1265 {
1266         int type;
1267         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1268                 if (parent > 0)
1269                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1270                 else
1271                         type = BTRFS_TREE_BLOCK_REF_KEY;
1272         } else {
1273                 if (parent > 0)
1274                         type = BTRFS_SHARED_DATA_REF_KEY;
1275                 else
1276                         type = BTRFS_EXTENT_DATA_REF_KEY;
1277         }
1278         return type;
1279 }
1280
1281 static int find_next_key(struct btrfs_path *path, int level,
1282                          struct btrfs_key *key)
1283
1284 {
1285         for (; level < BTRFS_MAX_LEVEL; level++) {
1286                 if (!path->nodes[level])
1287                         break;
1288                 if (path->slots[level] + 1 >=
1289                     btrfs_header_nritems(path->nodes[level]))
1290                         continue;
1291                 if (level == 0)
1292                         btrfs_item_key_to_cpu(path->nodes[level], key,
1293                                               path->slots[level] + 1);
1294                 else
1295                         btrfs_node_key_to_cpu(path->nodes[level], key,
1296                                               path->slots[level] + 1);
1297                 return 0;
1298         }
1299         return 1;
1300 }
1301
1302 /*
1303  * look for inline back ref. if back ref is found, *ref_ret is set
1304  * to the address of inline back ref, and 0 is returned.
1305  *
1306  * if back ref isn't found, *ref_ret is set to the address where it
1307  * should be inserted, and -ENOENT is returned.
1308  *
1309  * if insert is true and there are too many inline back refs, the path
1310  * points to the extent item, and -EAGAIN is returned.
1311  *
1312  * NOTE: inline back refs are ordered in the same way that back ref
1313  *       items in the tree are ordered.
1314  */
1315 static noinline_for_stack
1316 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1317                                  struct btrfs_root *root,
1318                                  struct btrfs_path *path,
1319                                  struct btrfs_extent_inline_ref **ref_ret,
1320                                  u64 bytenr, u64 num_bytes,
1321                                  u64 parent, u64 root_objectid,
1322                                  u64 owner, u64 offset, int insert)
1323 {
1324         struct btrfs_key key;
1325         struct extent_buffer *leaf;
1326         struct btrfs_extent_item *ei;
1327         struct btrfs_extent_inline_ref *iref;
1328         u64 flags;
1329         u64 item_size;
1330         unsigned long ptr;
1331         unsigned long end;
1332         int extra_size;
1333         int type;
1334         int want;
1335         int ret;
1336         int err = 0;
1337
1338         key.objectid = bytenr;
1339         key.type = BTRFS_EXTENT_ITEM_KEY;
1340         key.offset = num_bytes;
1341
1342         want = extent_ref_type(parent, owner);
1343         if (insert) {
1344                 extra_size = btrfs_extent_inline_ref_size(want);
1345                 path->keep_locks = 1;
1346         } else
1347                 extra_size = -1;
1348         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1349         if (ret < 0) {
1350                 err = ret;
1351                 goto out;
1352         }
1353         BUG_ON(ret);
1354
1355         leaf = path->nodes[0];
1356         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358         if (item_size < sizeof(*ei)) {
1359                 if (!insert) {
1360                         err = -ENOENT;
1361                         goto out;
1362                 }
1363                 ret = convert_extent_item_v0(trans, root, path, owner,
1364                                              extra_size);
1365                 if (ret < 0) {
1366                         err = ret;
1367                         goto out;
1368                 }
1369                 leaf = path->nodes[0];
1370                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1371         }
1372 #endif
1373         BUG_ON(item_size < sizeof(*ei));
1374
1375         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1376         flags = btrfs_extent_flags(leaf, ei);
1377
1378         ptr = (unsigned long)(ei + 1);
1379         end = (unsigned long)ei + item_size;
1380
1381         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1382                 ptr += sizeof(struct btrfs_tree_block_info);
1383                 BUG_ON(ptr > end);
1384         } else {
1385                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1386         }
1387
1388         err = -ENOENT;
1389         while (1) {
1390                 if (ptr >= end) {
1391                         WARN_ON(ptr > end);
1392                         break;
1393                 }
1394                 iref = (struct btrfs_extent_inline_ref *)ptr;
1395                 type = btrfs_extent_inline_ref_type(leaf, iref);
1396                 if (want < type)
1397                         break;
1398                 if (want > type) {
1399                         ptr += btrfs_extent_inline_ref_size(type);
1400                         continue;
1401                 }
1402
1403                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1404                         struct btrfs_extent_data_ref *dref;
1405                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1406                         if (match_extent_data_ref(leaf, dref, root_objectid,
1407                                                   owner, offset)) {
1408                                 err = 0;
1409                                 break;
1410                         }
1411                         if (hash_extent_data_ref_item(leaf, dref) <
1412                             hash_extent_data_ref(root_objectid, owner, offset))
1413                                 break;
1414                 } else {
1415                         u64 ref_offset;
1416                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1417                         if (parent > 0) {
1418                                 if (parent == ref_offset) {
1419                                         err = 0;
1420                                         break;
1421                                 }
1422                                 if (ref_offset < parent)
1423                                         break;
1424                         } else {
1425                                 if (root_objectid == ref_offset) {
1426                                         err = 0;
1427                                         break;
1428                                 }
1429                                 if (ref_offset < root_objectid)
1430                                         break;
1431                         }
1432                 }
1433                 ptr += btrfs_extent_inline_ref_size(type);
1434         }
1435         if (err == -ENOENT && insert) {
1436                 if (item_size + extra_size >=
1437                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1438                         err = -EAGAIN;
1439                         goto out;
1440                 }
1441                 /*
1442                  * To add new inline back ref, we have to make sure
1443                  * there is no corresponding back ref item.
1444                  * For simplicity, we just do not add new inline back
1445                  * ref if there is any kind of item for this block
1446                  */
1447                 if (find_next_key(path, 0, &key) == 0 &&
1448                     key.objectid == bytenr &&
1449                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1450                         err = -EAGAIN;
1451                         goto out;
1452                 }
1453         }
1454         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1455 out:
1456         if (insert) {
1457                 path->keep_locks = 0;
1458                 btrfs_unlock_up_safe(path, 1);
1459         }
1460         return err;
1461 }
1462
1463 /*
1464  * helper to add new inline back ref
1465  */
1466 static noinline_for_stack
1467 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1468                                 struct btrfs_root *root,
1469                                 struct btrfs_path *path,
1470                                 struct btrfs_extent_inline_ref *iref,
1471                                 u64 parent, u64 root_objectid,
1472                                 u64 owner, u64 offset, int refs_to_add,
1473                                 struct btrfs_delayed_extent_op *extent_op)
1474 {
1475         struct extent_buffer *leaf;
1476         struct btrfs_extent_item *ei;
1477         unsigned long ptr;
1478         unsigned long end;
1479         unsigned long item_offset;
1480         u64 refs;
1481         int size;
1482         int type;
1483         int ret;
1484
1485         leaf = path->nodes[0];
1486         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1487         item_offset = (unsigned long)iref - (unsigned long)ei;
1488
1489         type = extent_ref_type(parent, owner);
1490         size = btrfs_extent_inline_ref_size(type);
1491
1492         ret = btrfs_extend_item(trans, root, path, size);
1493         BUG_ON(ret);
1494
1495         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1496         refs = btrfs_extent_refs(leaf, ei);
1497         refs += refs_to_add;
1498         btrfs_set_extent_refs(leaf, ei, refs);
1499         if (extent_op)
1500                 __run_delayed_extent_op(extent_op, leaf, ei);
1501
1502         ptr = (unsigned long)ei + item_offset;
1503         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1504         if (ptr < end - size)
1505                 memmove_extent_buffer(leaf, ptr + size, ptr,
1506                                       end - size - ptr);
1507
1508         iref = (struct btrfs_extent_inline_ref *)ptr;
1509         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1510         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1511                 struct btrfs_extent_data_ref *dref;
1512                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1513                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1514                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1515                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1516                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1517         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1518                 struct btrfs_shared_data_ref *sref;
1519                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1520                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1521                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1522         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1523                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1524         } else {
1525                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1526         }
1527         btrfs_mark_buffer_dirty(leaf);
1528         return 0;
1529 }
1530
1531 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1532                                  struct btrfs_root *root,
1533                                  struct btrfs_path *path,
1534                                  struct btrfs_extent_inline_ref **ref_ret,
1535                                  u64 bytenr, u64 num_bytes, u64 parent,
1536                                  u64 root_objectid, u64 owner, u64 offset)
1537 {
1538         int ret;
1539
1540         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1541                                            bytenr, num_bytes, parent,
1542                                            root_objectid, owner, offset, 0);
1543         if (ret != -ENOENT)
1544                 return ret;
1545
1546         btrfs_release_path(root, path);
1547         *ref_ret = NULL;
1548
1549         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1550                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1551                                             root_objectid);
1552         } else {
1553                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1554                                              root_objectid, owner, offset);
1555         }
1556         return ret;
1557 }
1558
1559 /*
1560  * helper to update/remove inline back ref
1561  */
1562 static noinline_for_stack
1563 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1564                                  struct btrfs_root *root,
1565                                  struct btrfs_path *path,
1566                                  struct btrfs_extent_inline_ref *iref,
1567                                  int refs_to_mod,
1568                                  struct btrfs_delayed_extent_op *extent_op)
1569 {
1570         struct extent_buffer *leaf;
1571         struct btrfs_extent_item *ei;
1572         struct btrfs_extent_data_ref *dref = NULL;
1573         struct btrfs_shared_data_ref *sref = NULL;
1574         unsigned long ptr;
1575         unsigned long end;
1576         u32 item_size;
1577         int size;
1578         int type;
1579         int ret;
1580         u64 refs;
1581
1582         leaf = path->nodes[0];
1583         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1584         refs = btrfs_extent_refs(leaf, ei);
1585         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1586         refs += refs_to_mod;
1587         btrfs_set_extent_refs(leaf, ei, refs);
1588         if (extent_op)
1589                 __run_delayed_extent_op(extent_op, leaf, ei);
1590
1591         type = btrfs_extent_inline_ref_type(leaf, iref);
1592
1593         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1594                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1595                 refs = btrfs_extent_data_ref_count(leaf, dref);
1596         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1597                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1598                 refs = btrfs_shared_data_ref_count(leaf, sref);
1599         } else {
1600                 refs = 1;
1601                 BUG_ON(refs_to_mod != -1);
1602         }
1603
1604         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1605         refs += refs_to_mod;
1606
1607         if (refs > 0) {
1608                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1609                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1610                 else
1611                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1612         } else {
1613                 size =  btrfs_extent_inline_ref_size(type);
1614                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1615                 ptr = (unsigned long)iref;
1616                 end = (unsigned long)ei + item_size;
1617                 if (ptr + size < end)
1618                         memmove_extent_buffer(leaf, ptr, ptr + size,
1619                                               end - ptr - size);
1620                 item_size -= size;
1621                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1622                 BUG_ON(ret);
1623         }
1624         btrfs_mark_buffer_dirty(leaf);
1625         return 0;
1626 }
1627
1628 static noinline_for_stack
1629 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1630                                  struct btrfs_root *root,
1631                                  struct btrfs_path *path,
1632                                  u64 bytenr, u64 num_bytes, u64 parent,
1633                                  u64 root_objectid, u64 owner,
1634                                  u64 offset, int refs_to_add,
1635                                  struct btrfs_delayed_extent_op *extent_op)
1636 {
1637         struct btrfs_extent_inline_ref *iref;
1638         int ret;
1639
1640         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1641                                            bytenr, num_bytes, parent,
1642                                            root_objectid, owner, offset, 1);
1643         if (ret == 0) {
1644                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1645                 ret = update_inline_extent_backref(trans, root, path, iref,
1646                                                    refs_to_add, extent_op);
1647         } else if (ret == -ENOENT) {
1648                 ret = setup_inline_extent_backref(trans, root, path, iref,
1649                                                   parent, root_objectid,
1650                                                   owner, offset, refs_to_add,
1651                                                   extent_op);
1652         }
1653         return ret;
1654 }
1655
1656 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1657                                  struct btrfs_root *root,
1658                                  struct btrfs_path *path,
1659                                  u64 bytenr, u64 parent, u64 root_objectid,
1660                                  u64 owner, u64 offset, int refs_to_add)
1661 {
1662         int ret;
1663         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1664                 BUG_ON(refs_to_add != 1);
1665                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1666                                             parent, root_objectid);
1667         } else {
1668                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1669                                              parent, root_objectid,
1670                                              owner, offset, refs_to_add);
1671         }
1672         return ret;
1673 }
1674
1675 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1676                                  struct btrfs_root *root,
1677                                  struct btrfs_path *path,
1678                                  struct btrfs_extent_inline_ref *iref,
1679                                  int refs_to_drop, int is_data)
1680 {
1681         int ret;
1682
1683         BUG_ON(!is_data && refs_to_drop != 1);
1684         if (iref) {
1685                 ret = update_inline_extent_backref(trans, root, path, iref,
1686                                                    -refs_to_drop, NULL);
1687         } else if (is_data) {
1688                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1689         } else {
1690                 ret = btrfs_del_item(trans, root, path);
1691         }
1692         return ret;
1693 }
1694
1695 static void btrfs_issue_discard(struct block_device *bdev,
1696                                 u64 start, u64 len)
1697 {
1698         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1699                         BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
1700 }
1701
1702 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1703                                 u64 num_bytes)
1704 {
1705         int ret;
1706         u64 map_length = num_bytes;
1707         struct btrfs_multi_bio *multi = NULL;
1708
1709         if (!btrfs_test_opt(root, DISCARD))
1710                 return 0;
1711
1712         /* Tell the block device(s) that the sectors can be discarded */
1713         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1714                               bytenr, &map_length, &multi, 0);
1715         if (!ret) {
1716                 struct btrfs_bio_stripe *stripe = multi->stripes;
1717                 int i;
1718
1719                 if (map_length > num_bytes)
1720                         map_length = num_bytes;
1721
1722                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1723                         btrfs_issue_discard(stripe->dev->bdev,
1724                                             stripe->physical,
1725                                             map_length);
1726                 }
1727                 kfree(multi);
1728         }
1729
1730         return ret;
1731 }
1732
1733 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1734                          struct btrfs_root *root,
1735                          u64 bytenr, u64 num_bytes, u64 parent,
1736                          u64 root_objectid, u64 owner, u64 offset)
1737 {
1738         int ret;
1739         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1740                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1741
1742         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1743                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1744                                         parent, root_objectid, (int)owner,
1745                                         BTRFS_ADD_DELAYED_REF, NULL);
1746         } else {
1747                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1748                                         parent, root_objectid, owner, offset,
1749                                         BTRFS_ADD_DELAYED_REF, NULL);
1750         }
1751         return ret;
1752 }
1753
1754 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1755                                   struct btrfs_root *root,
1756                                   u64 bytenr, u64 num_bytes,
1757                                   u64 parent, u64 root_objectid,
1758                                   u64 owner, u64 offset, int refs_to_add,
1759                                   struct btrfs_delayed_extent_op *extent_op)
1760 {
1761         struct btrfs_path *path;
1762         struct extent_buffer *leaf;
1763         struct btrfs_extent_item *item;
1764         u64 refs;
1765         int ret;
1766         int err = 0;
1767
1768         path = btrfs_alloc_path();
1769         if (!path)
1770                 return -ENOMEM;
1771
1772         path->reada = 1;
1773         path->leave_spinning = 1;
1774         /* this will setup the path even if it fails to insert the back ref */
1775         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1776                                            path, bytenr, num_bytes, parent,
1777                                            root_objectid, owner, offset,
1778                                            refs_to_add, extent_op);
1779         if (ret == 0)
1780                 goto out;
1781
1782         if (ret != -EAGAIN) {
1783                 err = ret;
1784                 goto out;
1785         }
1786
1787         leaf = path->nodes[0];
1788         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1789         refs = btrfs_extent_refs(leaf, item);
1790         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1791         if (extent_op)
1792                 __run_delayed_extent_op(extent_op, leaf, item);
1793
1794         btrfs_mark_buffer_dirty(leaf);
1795         btrfs_release_path(root->fs_info->extent_root, path);
1796
1797         path->reada = 1;
1798         path->leave_spinning = 1;
1799
1800         /* now insert the actual backref */
1801         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1802                                     path, bytenr, parent, root_objectid,
1803                                     owner, offset, refs_to_add);
1804         BUG_ON(ret);
1805 out:
1806         btrfs_free_path(path);
1807         return err;
1808 }
1809
1810 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1811                                 struct btrfs_root *root,
1812                                 struct btrfs_delayed_ref_node *node,
1813                                 struct btrfs_delayed_extent_op *extent_op,
1814                                 int insert_reserved)
1815 {
1816         int ret = 0;
1817         struct btrfs_delayed_data_ref *ref;
1818         struct btrfs_key ins;
1819         u64 parent = 0;
1820         u64 ref_root = 0;
1821         u64 flags = 0;
1822
1823         ins.objectid = node->bytenr;
1824         ins.offset = node->num_bytes;
1825         ins.type = BTRFS_EXTENT_ITEM_KEY;
1826
1827         ref = btrfs_delayed_node_to_data_ref(node);
1828         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1829                 parent = ref->parent;
1830         else
1831                 ref_root = ref->root;
1832
1833         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1834                 if (extent_op) {
1835                         BUG_ON(extent_op->update_key);
1836                         flags |= extent_op->flags_to_set;
1837                 }
1838                 ret = alloc_reserved_file_extent(trans, root,
1839                                                  parent, ref_root, flags,
1840                                                  ref->objectid, ref->offset,
1841                                                  &ins, node->ref_mod);
1842         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1843                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1844                                              node->num_bytes, parent,
1845                                              ref_root, ref->objectid,
1846                                              ref->offset, node->ref_mod,
1847                                              extent_op);
1848         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1849                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1850                                           node->num_bytes, parent,
1851                                           ref_root, ref->objectid,
1852                                           ref->offset, node->ref_mod,
1853                                           extent_op);
1854         } else {
1855                 BUG();
1856         }
1857         return ret;
1858 }
1859
1860 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1861                                     struct extent_buffer *leaf,
1862                                     struct btrfs_extent_item *ei)
1863 {
1864         u64 flags = btrfs_extent_flags(leaf, ei);
1865         if (extent_op->update_flags) {
1866                 flags |= extent_op->flags_to_set;
1867                 btrfs_set_extent_flags(leaf, ei, flags);
1868         }
1869
1870         if (extent_op->update_key) {
1871                 struct btrfs_tree_block_info *bi;
1872                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1873                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1874                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1875         }
1876 }
1877
1878 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1879                                  struct btrfs_root *root,
1880                                  struct btrfs_delayed_ref_node *node,
1881                                  struct btrfs_delayed_extent_op *extent_op)
1882 {
1883         struct btrfs_key key;
1884         struct btrfs_path *path;
1885         struct btrfs_extent_item *ei;
1886         struct extent_buffer *leaf;
1887         u32 item_size;
1888         int ret;
1889         int err = 0;
1890
1891         path = btrfs_alloc_path();
1892         if (!path)
1893                 return -ENOMEM;
1894
1895         key.objectid = node->bytenr;
1896         key.type = BTRFS_EXTENT_ITEM_KEY;
1897         key.offset = node->num_bytes;
1898
1899         path->reada = 1;
1900         path->leave_spinning = 1;
1901         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1902                                 path, 0, 1);
1903         if (ret < 0) {
1904                 err = ret;
1905                 goto out;
1906         }
1907         if (ret > 0) {
1908                 err = -EIO;
1909                 goto out;
1910         }
1911
1912         leaf = path->nodes[0];
1913         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1914 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1915         if (item_size < sizeof(*ei)) {
1916                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1917                                              path, (u64)-1, 0);
1918                 if (ret < 0) {
1919                         err = ret;
1920                         goto out;
1921                 }
1922                 leaf = path->nodes[0];
1923                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1924         }
1925 #endif
1926         BUG_ON(item_size < sizeof(*ei));
1927         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1928         __run_delayed_extent_op(extent_op, leaf, ei);
1929
1930         btrfs_mark_buffer_dirty(leaf);
1931 out:
1932         btrfs_free_path(path);
1933         return err;
1934 }
1935
1936 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1937                                 struct btrfs_root *root,
1938                                 struct btrfs_delayed_ref_node *node,
1939                                 struct btrfs_delayed_extent_op *extent_op,
1940                                 int insert_reserved)
1941 {
1942         int ret = 0;
1943         struct btrfs_delayed_tree_ref *ref;
1944         struct btrfs_key ins;
1945         u64 parent = 0;
1946         u64 ref_root = 0;
1947
1948         ins.objectid = node->bytenr;
1949         ins.offset = node->num_bytes;
1950         ins.type = BTRFS_EXTENT_ITEM_KEY;
1951
1952         ref = btrfs_delayed_node_to_tree_ref(node);
1953         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1954                 parent = ref->parent;
1955         else
1956                 ref_root = ref->root;
1957
1958         BUG_ON(node->ref_mod != 1);
1959         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1960                 BUG_ON(!extent_op || !extent_op->update_flags ||
1961                        !extent_op->update_key);
1962                 ret = alloc_reserved_tree_block(trans, root,
1963                                                 parent, ref_root,
1964                                                 extent_op->flags_to_set,
1965                                                 &extent_op->key,
1966                                                 ref->level, &ins);
1967         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1968                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1969                                              node->num_bytes, parent, ref_root,
1970                                              ref->level, 0, 1, extent_op);
1971         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1972                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1973                                           node->num_bytes, parent, ref_root,
1974                                           ref->level, 0, 1, extent_op);
1975         } else {
1976                 BUG();
1977         }
1978         return ret;
1979 }
1980
1981 /* helper function to actually process a single delayed ref entry */
1982 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1983                                struct btrfs_root *root,
1984                                struct btrfs_delayed_ref_node *node,
1985                                struct btrfs_delayed_extent_op *extent_op,
1986                                int insert_reserved)
1987 {
1988         int ret;
1989         if (btrfs_delayed_ref_is_head(node)) {
1990                 struct btrfs_delayed_ref_head *head;
1991                 /*
1992                  * we've hit the end of the chain and we were supposed
1993                  * to insert this extent into the tree.  But, it got
1994                  * deleted before we ever needed to insert it, so all
1995                  * we have to do is clean up the accounting
1996                  */
1997                 BUG_ON(extent_op);
1998                 head = btrfs_delayed_node_to_head(node);
1999                 if (insert_reserved) {
2000                         btrfs_pin_extent(root, node->bytenr,
2001                                          node->num_bytes, 1);
2002                         if (head->is_data) {
2003                                 ret = btrfs_del_csums(trans, root,
2004                                                       node->bytenr,
2005                                                       node->num_bytes);
2006                                 BUG_ON(ret);
2007                         }
2008                 }
2009                 mutex_unlock(&head->mutex);
2010                 return 0;
2011         }
2012
2013         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2014             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2015                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2016                                            insert_reserved);
2017         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2018                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2019                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2020                                            insert_reserved);
2021         else
2022                 BUG();
2023         return ret;
2024 }
2025
2026 static noinline struct btrfs_delayed_ref_node *
2027 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2028 {
2029         struct rb_node *node;
2030         struct btrfs_delayed_ref_node *ref;
2031         int action = BTRFS_ADD_DELAYED_REF;
2032 again:
2033         /*
2034          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2035          * this prevents ref count from going down to zero when
2036          * there still are pending delayed ref.
2037          */
2038         node = rb_prev(&head->node.rb_node);
2039         while (1) {
2040                 if (!node)
2041                         break;
2042                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2043                                 rb_node);
2044                 if (ref->bytenr != head->node.bytenr)
2045                         break;
2046                 if (ref->action == action)
2047                         return ref;
2048                 node = rb_prev(node);
2049         }
2050         if (action == BTRFS_ADD_DELAYED_REF) {
2051                 action = BTRFS_DROP_DELAYED_REF;
2052                 goto again;
2053         }
2054         return NULL;
2055 }
2056
2057 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2058                                        struct btrfs_root *root,
2059                                        struct list_head *cluster)
2060 {
2061         struct btrfs_delayed_ref_root *delayed_refs;
2062         struct btrfs_delayed_ref_node *ref;
2063         struct btrfs_delayed_ref_head *locked_ref = NULL;
2064         struct btrfs_delayed_extent_op *extent_op;
2065         int ret;
2066         int count = 0;
2067         int must_insert_reserved = 0;
2068
2069         delayed_refs = &trans->transaction->delayed_refs;
2070         while (1) {
2071                 if (!locked_ref) {
2072                         /* pick a new head ref from the cluster list */
2073                         if (list_empty(cluster))
2074                                 break;
2075
2076                         locked_ref = list_entry(cluster->next,
2077                                      struct btrfs_delayed_ref_head, cluster);
2078
2079                         /* grab the lock that says we are going to process
2080                          * all the refs for this head */
2081                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2082
2083                         /*
2084                          * we may have dropped the spin lock to get the head
2085                          * mutex lock, and that might have given someone else
2086                          * time to free the head.  If that's true, it has been
2087                          * removed from our list and we can move on.
2088                          */
2089                         if (ret == -EAGAIN) {
2090                                 locked_ref = NULL;
2091                                 count++;
2092                                 continue;
2093                         }
2094                 }
2095
2096                 /*
2097                  * record the must insert reserved flag before we
2098                  * drop the spin lock.
2099                  */
2100                 must_insert_reserved = locked_ref->must_insert_reserved;
2101                 locked_ref->must_insert_reserved = 0;
2102
2103                 extent_op = locked_ref->extent_op;
2104                 locked_ref->extent_op = NULL;
2105
2106                 /*
2107                  * locked_ref is the head node, so we have to go one
2108                  * node back for any delayed ref updates
2109                  */
2110                 ref = select_delayed_ref(locked_ref);
2111                 if (!ref) {
2112                         /* All delayed refs have been processed, Go ahead
2113                          * and send the head node to run_one_delayed_ref,
2114                          * so that any accounting fixes can happen
2115                          */
2116                         ref = &locked_ref->node;
2117
2118                         if (extent_op && must_insert_reserved) {
2119                                 kfree(extent_op);
2120                                 extent_op = NULL;
2121                         }
2122
2123                         if (extent_op) {
2124                                 spin_unlock(&delayed_refs->lock);
2125
2126                                 ret = run_delayed_extent_op(trans, root,
2127                                                             ref, extent_op);
2128                                 BUG_ON(ret);
2129                                 kfree(extent_op);
2130
2131                                 cond_resched();
2132                                 spin_lock(&delayed_refs->lock);
2133                                 continue;
2134                         }
2135
2136                         list_del_init(&locked_ref->cluster);
2137                         locked_ref = NULL;
2138                 }
2139
2140                 ref->in_tree = 0;
2141                 rb_erase(&ref->rb_node, &delayed_refs->root);
2142                 delayed_refs->num_entries--;
2143
2144                 spin_unlock(&delayed_refs->lock);
2145
2146                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2147                                           must_insert_reserved);
2148                 BUG_ON(ret);
2149
2150                 btrfs_put_delayed_ref(ref);
2151                 kfree(extent_op);
2152                 count++;
2153
2154                 cond_resched();
2155                 spin_lock(&delayed_refs->lock);
2156         }
2157         return count;
2158 }
2159
2160 /*
2161  * this starts processing the delayed reference count updates and
2162  * extent insertions we have queued up so far.  count can be
2163  * 0, which means to process everything in the tree at the start
2164  * of the run (but not newly added entries), or it can be some target
2165  * number you'd like to process.
2166  */
2167 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2168                            struct btrfs_root *root, unsigned long count)
2169 {
2170         struct rb_node *node;
2171         struct btrfs_delayed_ref_root *delayed_refs;
2172         struct btrfs_delayed_ref_node *ref;
2173         struct list_head cluster;
2174         int ret;
2175         int run_all = count == (unsigned long)-1;
2176         int run_most = 0;
2177
2178         if (root == root->fs_info->extent_root)
2179                 root = root->fs_info->tree_root;
2180
2181         delayed_refs = &trans->transaction->delayed_refs;
2182         INIT_LIST_HEAD(&cluster);
2183 again:
2184         spin_lock(&delayed_refs->lock);
2185         if (count == 0) {
2186                 count = delayed_refs->num_entries * 2;
2187                 run_most = 1;
2188         }
2189         while (1) {
2190                 if (!(run_all || run_most) &&
2191                     delayed_refs->num_heads_ready < 64)
2192                         break;
2193
2194                 /*
2195                  * go find something we can process in the rbtree.  We start at
2196                  * the beginning of the tree, and then build a cluster
2197                  * of refs to process starting at the first one we are able to
2198                  * lock
2199                  */
2200                 ret = btrfs_find_ref_cluster(trans, &cluster,
2201                                              delayed_refs->run_delayed_start);
2202                 if (ret)
2203                         break;
2204
2205                 ret = run_clustered_refs(trans, root, &cluster);
2206                 BUG_ON(ret < 0);
2207
2208                 count -= min_t(unsigned long, ret, count);
2209
2210                 if (count == 0)
2211                         break;
2212         }
2213
2214         if (run_all) {
2215                 node = rb_first(&delayed_refs->root);
2216                 if (!node)
2217                         goto out;
2218                 count = (unsigned long)-1;
2219
2220                 while (node) {
2221                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2222                                        rb_node);
2223                         if (btrfs_delayed_ref_is_head(ref)) {
2224                                 struct btrfs_delayed_ref_head *head;
2225
2226                                 head = btrfs_delayed_node_to_head(ref);
2227                                 atomic_inc(&ref->refs);
2228
2229                                 spin_unlock(&delayed_refs->lock);
2230                                 mutex_lock(&head->mutex);
2231                                 mutex_unlock(&head->mutex);
2232
2233                                 btrfs_put_delayed_ref(ref);
2234                                 cond_resched();
2235                                 goto again;
2236                         }
2237                         node = rb_next(node);
2238                 }
2239                 spin_unlock(&delayed_refs->lock);
2240                 schedule_timeout(1);
2241                 goto again;
2242         }
2243 out:
2244         spin_unlock(&delayed_refs->lock);
2245         return 0;
2246 }
2247
2248 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2249                                 struct btrfs_root *root,
2250                                 u64 bytenr, u64 num_bytes, u64 flags,
2251                                 int is_data)
2252 {
2253         struct btrfs_delayed_extent_op *extent_op;
2254         int ret;
2255
2256         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2257         if (!extent_op)
2258                 return -ENOMEM;
2259
2260         extent_op->flags_to_set = flags;
2261         extent_op->update_flags = 1;
2262         extent_op->update_key = 0;
2263         extent_op->is_data = is_data ? 1 : 0;
2264
2265         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2266         if (ret)
2267                 kfree(extent_op);
2268         return ret;
2269 }
2270
2271 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2272                                       struct btrfs_root *root,
2273                                       struct btrfs_path *path,
2274                                       u64 objectid, u64 offset, u64 bytenr)
2275 {
2276         struct btrfs_delayed_ref_head *head;
2277         struct btrfs_delayed_ref_node *ref;
2278         struct btrfs_delayed_data_ref *data_ref;
2279         struct btrfs_delayed_ref_root *delayed_refs;
2280         struct rb_node *node;
2281         int ret = 0;
2282
2283         ret = -ENOENT;
2284         delayed_refs = &trans->transaction->delayed_refs;
2285         spin_lock(&delayed_refs->lock);
2286         head = btrfs_find_delayed_ref_head(trans, bytenr);
2287         if (!head)
2288                 goto out;
2289
2290         if (!mutex_trylock(&head->mutex)) {
2291                 atomic_inc(&head->node.refs);
2292                 spin_unlock(&delayed_refs->lock);
2293
2294                 btrfs_release_path(root->fs_info->extent_root, path);
2295
2296                 mutex_lock(&head->mutex);
2297                 mutex_unlock(&head->mutex);
2298                 btrfs_put_delayed_ref(&head->node);
2299                 return -EAGAIN;
2300         }
2301
2302         node = rb_prev(&head->node.rb_node);
2303         if (!node)
2304                 goto out_unlock;
2305
2306         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2307
2308         if (ref->bytenr != bytenr)
2309                 goto out_unlock;
2310
2311         ret = 1;
2312         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2313                 goto out_unlock;
2314
2315         data_ref = btrfs_delayed_node_to_data_ref(ref);
2316
2317         node = rb_prev(node);
2318         if (node) {
2319                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2320                 if (ref->bytenr == bytenr)
2321                         goto out_unlock;
2322         }
2323
2324         if (data_ref->root != root->root_key.objectid ||
2325             data_ref->objectid != objectid || data_ref->offset != offset)
2326                 goto out_unlock;
2327
2328         ret = 0;
2329 out_unlock:
2330         mutex_unlock(&head->mutex);
2331 out:
2332         spin_unlock(&delayed_refs->lock);
2333         return ret;
2334 }
2335
2336 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2337                                         struct btrfs_root *root,
2338                                         struct btrfs_path *path,
2339                                         u64 objectid, u64 offset, u64 bytenr)
2340 {
2341         struct btrfs_root *extent_root = root->fs_info->extent_root;
2342         struct extent_buffer *leaf;
2343         struct btrfs_extent_data_ref *ref;
2344         struct btrfs_extent_inline_ref *iref;
2345         struct btrfs_extent_item *ei;
2346         struct btrfs_key key;
2347         u32 item_size;
2348         int ret;
2349
2350         key.objectid = bytenr;
2351         key.offset = (u64)-1;
2352         key.type = BTRFS_EXTENT_ITEM_KEY;
2353
2354         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2355         if (ret < 0)
2356                 goto out;
2357         BUG_ON(ret == 0);
2358
2359         ret = -ENOENT;
2360         if (path->slots[0] == 0)
2361                 goto out;
2362
2363         path->slots[0]--;
2364         leaf = path->nodes[0];
2365         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2366
2367         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2368                 goto out;
2369
2370         ret = 1;
2371         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2372 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2373         if (item_size < sizeof(*ei)) {
2374                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2375                 goto out;
2376         }
2377 #endif
2378         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2379
2380         if (item_size != sizeof(*ei) +
2381             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2382                 goto out;
2383
2384         if (btrfs_extent_generation(leaf, ei) <=
2385             btrfs_root_last_snapshot(&root->root_item))
2386                 goto out;
2387
2388         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2389         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2390             BTRFS_EXTENT_DATA_REF_KEY)
2391                 goto out;
2392
2393         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2394         if (btrfs_extent_refs(leaf, ei) !=
2395             btrfs_extent_data_ref_count(leaf, ref) ||
2396             btrfs_extent_data_ref_root(leaf, ref) !=
2397             root->root_key.objectid ||
2398             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2399             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2400                 goto out;
2401
2402         ret = 0;
2403 out:
2404         return ret;
2405 }
2406
2407 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2408                           struct btrfs_root *root,
2409                           u64 objectid, u64 offset, u64 bytenr)
2410 {
2411         struct btrfs_path *path;
2412         int ret;
2413         int ret2;
2414
2415         path = btrfs_alloc_path();
2416         if (!path)
2417                 return -ENOENT;
2418
2419         do {
2420                 ret = check_committed_ref(trans, root, path, objectid,
2421                                           offset, bytenr);
2422                 if (ret && ret != -ENOENT)
2423                         goto out;
2424
2425                 ret2 = check_delayed_ref(trans, root, path, objectid,
2426                                          offset, bytenr);
2427         } while (ret2 == -EAGAIN);
2428
2429         if (ret2 && ret2 != -ENOENT) {
2430                 ret = ret2;
2431                 goto out;
2432         }
2433
2434         if (ret != -ENOENT || ret2 != -ENOENT)
2435                 ret = 0;
2436 out:
2437         btrfs_free_path(path);
2438         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2439                 WARN_ON(ret > 0);
2440         return ret;
2441 }
2442
2443 #if 0
2444 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2445                     struct extent_buffer *buf, u32 nr_extents)
2446 {
2447         struct btrfs_key key;
2448         struct btrfs_file_extent_item *fi;
2449         u64 root_gen;
2450         u32 nritems;
2451         int i;
2452         int level;
2453         int ret = 0;
2454         int shared = 0;
2455
2456         if (!root->ref_cows)
2457                 return 0;
2458
2459         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2460                 shared = 0;
2461                 root_gen = root->root_key.offset;
2462         } else {
2463                 shared = 1;
2464                 root_gen = trans->transid - 1;
2465         }
2466
2467         level = btrfs_header_level(buf);
2468         nritems = btrfs_header_nritems(buf);
2469
2470         if (level == 0) {
2471                 struct btrfs_leaf_ref *ref;
2472                 struct btrfs_extent_info *info;
2473
2474                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2475                 if (!ref) {
2476                         ret = -ENOMEM;
2477                         goto out;
2478                 }
2479
2480                 ref->root_gen = root_gen;
2481                 ref->bytenr = buf->start;
2482                 ref->owner = btrfs_header_owner(buf);
2483                 ref->generation = btrfs_header_generation(buf);
2484                 ref->nritems = nr_extents;
2485                 info = ref->extents;
2486
2487                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2488                         u64 disk_bytenr;
2489                         btrfs_item_key_to_cpu(buf, &key, i);
2490                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2491                                 continue;
2492                         fi = btrfs_item_ptr(buf, i,
2493                                             struct btrfs_file_extent_item);
2494                         if (btrfs_file_extent_type(buf, fi) ==
2495                             BTRFS_FILE_EXTENT_INLINE)
2496                                 continue;
2497                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2498                         if (disk_bytenr == 0)
2499                                 continue;
2500
2501                         info->bytenr = disk_bytenr;
2502                         info->num_bytes =
2503                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2504                         info->objectid = key.objectid;
2505                         info->offset = key.offset;
2506                         info++;
2507                 }
2508
2509                 ret = btrfs_add_leaf_ref(root, ref, shared);
2510                 if (ret == -EEXIST && shared) {
2511                         struct btrfs_leaf_ref *old;
2512                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2513                         BUG_ON(!old);
2514                         btrfs_remove_leaf_ref(root, old);
2515                         btrfs_free_leaf_ref(root, old);
2516                         ret = btrfs_add_leaf_ref(root, ref, shared);
2517                 }
2518                 WARN_ON(ret);
2519                 btrfs_free_leaf_ref(root, ref);
2520         }
2521 out:
2522         return ret;
2523 }
2524
2525 /* when a block goes through cow, we update the reference counts of
2526  * everything that block points to.  The internal pointers of the block
2527  * can be in just about any order, and it is likely to have clusters of
2528  * things that are close together and clusters of things that are not.
2529  *
2530  * To help reduce the seeks that come with updating all of these reference
2531  * counts, sort them by byte number before actual updates are done.
2532  *
2533  * struct refsort is used to match byte number to slot in the btree block.
2534  * we sort based on the byte number and then use the slot to actually
2535  * find the item.
2536  *
2537  * struct refsort is smaller than strcut btrfs_item and smaller than
2538  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2539  * for a btree block, there's no way for a kmalloc of refsorts for a
2540  * single node to be bigger than a page.
2541  */
2542 struct refsort {
2543         u64 bytenr;
2544         u32 slot;
2545 };
2546
2547 /*
2548  * for passing into sort()
2549  */
2550 static int refsort_cmp(const void *a_void, const void *b_void)
2551 {
2552         const struct refsort *a = a_void;
2553         const struct refsort *b = b_void;
2554
2555         if (a->bytenr < b->bytenr)
2556                 return -1;
2557         if (a->bytenr > b->bytenr)
2558                 return 1;
2559         return 0;
2560 }
2561 #endif
2562
2563 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2564                            struct btrfs_root *root,
2565                            struct extent_buffer *buf,
2566                            int full_backref, int inc)
2567 {
2568         u64 bytenr;
2569         u64 num_bytes;
2570         u64 parent;
2571         u64 ref_root;
2572         u32 nritems;
2573         struct btrfs_key key;
2574         struct btrfs_file_extent_item *fi;
2575         int i;
2576         int level;
2577         int ret = 0;
2578         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2579                             u64, u64, u64, u64, u64, u64);
2580
2581         ref_root = btrfs_header_owner(buf);
2582         nritems = btrfs_header_nritems(buf);
2583         level = btrfs_header_level(buf);
2584
2585         if (!root->ref_cows && level == 0)
2586                 return 0;
2587
2588         if (inc)
2589                 process_func = btrfs_inc_extent_ref;
2590         else
2591                 process_func = btrfs_free_extent;
2592
2593         if (full_backref)
2594                 parent = buf->start;
2595         else
2596                 parent = 0;
2597
2598         for (i = 0; i < nritems; i++) {
2599                 if (level == 0) {
2600                         btrfs_item_key_to_cpu(buf, &key, i);
2601                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2602                                 continue;
2603                         fi = btrfs_item_ptr(buf, i,
2604                                             struct btrfs_file_extent_item);
2605                         if (btrfs_file_extent_type(buf, fi) ==
2606                             BTRFS_FILE_EXTENT_INLINE)
2607                                 continue;
2608                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2609                         if (bytenr == 0)
2610                                 continue;
2611
2612                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2613                         key.offset -= btrfs_file_extent_offset(buf, fi);
2614                         ret = process_func(trans, root, bytenr, num_bytes,
2615                                            parent, ref_root, key.objectid,
2616                                            key.offset);
2617                         if (ret)
2618                                 goto fail;
2619                 } else {
2620                         bytenr = btrfs_node_blockptr(buf, i);
2621                         num_bytes = btrfs_level_size(root, level - 1);
2622                         ret = process_func(trans, root, bytenr, num_bytes,
2623                                            parent, ref_root, level - 1, 0);
2624                         if (ret)
2625                                 goto fail;
2626                 }
2627         }
2628         return 0;
2629 fail:
2630         BUG();
2631         return ret;
2632 }
2633
2634 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2635                   struct extent_buffer *buf, int full_backref)
2636 {
2637         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2638 }
2639
2640 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2641                   struct extent_buffer *buf, int full_backref)
2642 {
2643         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2644 }
2645
2646 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2647                                  struct btrfs_root *root,
2648                                  struct btrfs_path *path,
2649                                  struct btrfs_block_group_cache *cache)
2650 {
2651         int ret;
2652         struct btrfs_root *extent_root = root->fs_info->extent_root;
2653         unsigned long bi;
2654         struct extent_buffer *leaf;
2655
2656         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2657         if (ret < 0)
2658                 goto fail;
2659         BUG_ON(ret);
2660
2661         leaf = path->nodes[0];
2662         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2663         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2664         btrfs_mark_buffer_dirty(leaf);
2665         btrfs_release_path(extent_root, path);
2666 fail:
2667         if (ret)
2668                 return ret;
2669         return 0;
2670
2671 }
2672
2673 static struct btrfs_block_group_cache *
2674 next_block_group(struct btrfs_root *root,
2675                  struct btrfs_block_group_cache *cache)
2676 {
2677         struct rb_node *node;
2678         spin_lock(&root->fs_info->block_group_cache_lock);
2679         node = rb_next(&cache->cache_node);
2680         btrfs_put_block_group(cache);
2681         if (node) {
2682                 cache = rb_entry(node, struct btrfs_block_group_cache,
2683                                  cache_node);
2684                 btrfs_get_block_group(cache);
2685         } else
2686                 cache = NULL;
2687         spin_unlock(&root->fs_info->block_group_cache_lock);
2688         return cache;
2689 }
2690
2691 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2692                             struct btrfs_trans_handle *trans,
2693                             struct btrfs_path *path)
2694 {
2695         struct btrfs_root *root = block_group->fs_info->tree_root;
2696         struct inode *inode = NULL;
2697         u64 alloc_hint = 0;
2698         int num_pages = 0;
2699         int retries = 0;
2700         int ret = 0;
2701
2702         /*
2703          * If this block group is smaller than 100 megs don't bother caching the
2704          * block group.
2705          */
2706         if (block_group->key.offset < (100 * 1024 * 1024)) {
2707                 spin_lock(&block_group->lock);
2708                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2709                 spin_unlock(&block_group->lock);
2710                 return 0;
2711         }
2712
2713 again:
2714         inode = lookup_free_space_inode(root, block_group, path);
2715         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2716                 ret = PTR_ERR(inode);
2717                 btrfs_release_path(root, path);
2718                 goto out;
2719         }
2720
2721         if (IS_ERR(inode)) {
2722                 BUG_ON(retries);
2723                 retries++;
2724
2725                 if (block_group->ro)
2726                         goto out_free;
2727
2728                 ret = create_free_space_inode(root, trans, block_group, path);
2729                 if (ret)
2730                         goto out_free;
2731                 goto again;
2732         }
2733
2734         /*
2735          * We want to set the generation to 0, that way if anything goes wrong
2736          * from here on out we know not to trust this cache when we load up next
2737          * time.
2738          */
2739         BTRFS_I(inode)->generation = 0;
2740         ret = btrfs_update_inode(trans, root, inode);
2741         WARN_ON(ret);
2742
2743         if (i_size_read(inode) > 0) {
2744                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2745                                                       inode);
2746                 if (ret)
2747                         goto out_put;
2748         }
2749
2750         spin_lock(&block_group->lock);
2751         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2752                 spin_unlock(&block_group->lock);
2753                 goto out_put;
2754         }
2755         spin_unlock(&block_group->lock);
2756
2757         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2758         if (!num_pages)
2759                 num_pages = 1;
2760
2761         /*
2762          * Just to make absolutely sure we have enough space, we're going to
2763          * preallocate 12 pages worth of space for each block group.  In
2764          * practice we ought to use at most 8, but we need extra space so we can
2765          * add our header and have a terminator between the extents and the
2766          * bitmaps.
2767          */
2768         num_pages *= 16;
2769         num_pages *= PAGE_CACHE_SIZE;
2770
2771         ret = btrfs_check_data_free_space(inode, num_pages);
2772         if (ret)
2773                 goto out_put;
2774
2775         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2776                                               num_pages, num_pages,
2777                                               &alloc_hint);
2778         btrfs_free_reserved_data_space(inode, num_pages);
2779 out_put:
2780         iput(inode);
2781 out_free:
2782         btrfs_release_path(root, path);
2783 out:
2784         spin_lock(&block_group->lock);
2785         if (ret)
2786                 block_group->disk_cache_state = BTRFS_DC_ERROR;
2787         else
2788                 block_group->disk_cache_state = BTRFS_DC_SETUP;
2789         spin_unlock(&block_group->lock);
2790
2791         return ret;
2792 }
2793
2794 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2795                                    struct btrfs_root *root)
2796 {
2797         struct btrfs_block_group_cache *cache;
2798         int err = 0;
2799         struct btrfs_path *path;
2800         u64 last = 0;
2801
2802         path = btrfs_alloc_path();
2803         if (!path)
2804                 return -ENOMEM;
2805
2806 again:
2807         while (1) {
2808                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2809                 while (cache) {
2810                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2811                                 break;
2812                         cache = next_block_group(root, cache);
2813                 }
2814                 if (!cache) {
2815                         if (last == 0)
2816                                 break;
2817                         last = 0;
2818                         continue;
2819                 }
2820                 err = cache_save_setup(cache, trans, path);
2821                 last = cache->key.objectid + cache->key.offset;
2822                 btrfs_put_block_group(cache);
2823         }
2824
2825         while (1) {
2826                 if (last == 0) {
2827                         err = btrfs_run_delayed_refs(trans, root,
2828                                                      (unsigned long)-1);
2829                         BUG_ON(err);
2830                 }
2831
2832                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2833                 while (cache) {
2834                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2835                                 btrfs_put_block_group(cache);
2836                                 goto again;
2837                         }
2838
2839                         if (cache->dirty)
2840                                 break;
2841                         cache = next_block_group(root, cache);
2842                 }
2843                 if (!cache) {
2844                         if (last == 0)
2845                                 break;
2846                         last = 0;
2847                         continue;
2848                 }
2849
2850                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2851                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2852                 cache->dirty = 0;
2853                 last = cache->key.objectid + cache->key.offset;
2854
2855                 err = write_one_cache_group(trans, root, path, cache);
2856                 BUG_ON(err);
2857                 btrfs_put_block_group(cache);
2858         }
2859
2860         while (1) {
2861                 /*
2862                  * I don't think this is needed since we're just marking our
2863                  * preallocated extent as written, but just in case it can't
2864                  * hurt.
2865                  */
2866                 if (last == 0) {
2867                         err = btrfs_run_delayed_refs(trans, root,
2868                                                      (unsigned long)-1);
2869                         BUG_ON(err);
2870                 }
2871
2872                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2873                 while (cache) {
2874                         /*
2875                          * Really this shouldn't happen, but it could if we
2876                          * couldn't write the entire preallocated extent and
2877                          * splitting the extent resulted in a new block.
2878                          */
2879                         if (cache->dirty) {
2880                                 btrfs_put_block_group(cache);
2881                                 goto again;
2882                         }
2883                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2884                                 break;
2885                         cache = next_block_group(root, cache);
2886                 }
2887                 if (!cache) {
2888                         if (last == 0)
2889                                 break;
2890                         last = 0;
2891                         continue;
2892                 }
2893
2894                 btrfs_write_out_cache(root, trans, cache, path);
2895
2896                 /*
2897                  * If we didn't have an error then the cache state is still
2898                  * NEED_WRITE, so we can set it to WRITTEN.
2899                  */
2900                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2901                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2902                 last = cache->key.objectid + cache->key.offset;
2903                 btrfs_put_block_group(cache);
2904         }
2905
2906         btrfs_free_path(path);
2907         return 0;
2908 }
2909
2910 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2911 {
2912         struct btrfs_block_group_cache *block_group;
2913         int readonly = 0;
2914
2915         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2916         if (!block_group || block_group->ro)
2917                 readonly = 1;
2918         if (block_group)
2919                 btrfs_put_block_group(block_group);
2920         return readonly;
2921 }
2922
2923 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2924                              u64 total_bytes, u64 bytes_used,
2925                              struct btrfs_space_info **space_info)
2926 {
2927         struct btrfs_space_info *found;
2928         int i;
2929         int factor;
2930
2931         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2932                      BTRFS_BLOCK_GROUP_RAID10))
2933                 factor = 2;
2934         else
2935                 factor = 1;
2936
2937         found = __find_space_info(info, flags);
2938         if (found) {
2939                 spin_lock(&found->lock);
2940                 found->total_bytes += total_bytes;
2941                 found->bytes_used += bytes_used;
2942                 found->disk_used += bytes_used * factor;
2943                 found->full = 0;
2944                 spin_unlock(&found->lock);
2945                 *space_info = found;
2946                 return 0;
2947         }
2948         found = kzalloc(sizeof(*found), GFP_NOFS);
2949         if (!found)
2950                 return -ENOMEM;
2951
2952         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2953                 INIT_LIST_HEAD(&found->block_groups[i]);
2954         init_rwsem(&found->groups_sem);
2955         spin_lock_init(&found->lock);
2956         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2957                                 BTRFS_BLOCK_GROUP_SYSTEM |
2958                                 BTRFS_BLOCK_GROUP_METADATA);
2959         found->total_bytes = total_bytes;
2960         found->bytes_used = bytes_used;
2961         found->disk_used = bytes_used * factor;
2962         found->bytes_pinned = 0;
2963         found->bytes_reserved = 0;
2964         found->bytes_readonly = 0;
2965         found->bytes_may_use = 0;
2966         found->full = 0;
2967         found->force_alloc = 0;
2968         *space_info = found;
2969         list_add_rcu(&found->list, &info->space_info);
2970         atomic_set(&found->caching_threads, 0);
2971         return 0;
2972 }
2973
2974 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2975 {
2976         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2977                                    BTRFS_BLOCK_GROUP_RAID1 |
2978                                    BTRFS_BLOCK_GROUP_RAID10 |
2979                                    BTRFS_BLOCK_GROUP_DUP);
2980         if (extra_flags) {
2981                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2982                         fs_info->avail_data_alloc_bits |= extra_flags;
2983                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2984                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2985                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2986                         fs_info->avail_system_alloc_bits |= extra_flags;
2987         }
2988 }
2989
2990 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2991 {
2992         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2993
2994         if (num_devices == 1)
2995                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2996         if (num_devices < 4)
2997                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2998
2999         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3000             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3001                       BTRFS_BLOCK_GROUP_RAID10))) {
3002                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3003         }
3004
3005         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3006             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3007                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3008         }
3009
3010         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3011             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3012              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3013              (flags & BTRFS_BLOCK_GROUP_DUP)))
3014                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3015         return flags;
3016 }
3017
3018 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3019 {
3020         if (flags & BTRFS_BLOCK_GROUP_DATA)
3021                 flags |= root->fs_info->avail_data_alloc_bits &
3022                          root->fs_info->data_alloc_profile;
3023         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3024                 flags |= root->fs_info->avail_system_alloc_bits &
3025                          root->fs_info->system_alloc_profile;
3026         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3027                 flags |= root->fs_info->avail_metadata_alloc_bits &
3028                          root->fs_info->metadata_alloc_profile;
3029         return btrfs_reduce_alloc_profile(root, flags);
3030 }
3031
3032 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3033 {
3034         u64 flags;
3035
3036         if (data)
3037                 flags = BTRFS_BLOCK_GROUP_DATA;
3038         else if (root == root->fs_info->chunk_root)
3039                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3040         else
3041                 flags = BTRFS_BLOCK_GROUP_METADATA;
3042
3043         return get_alloc_profile(root, flags);
3044 }
3045
3046 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3047 {
3048         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3049                                                        BTRFS_BLOCK_GROUP_DATA);
3050 }
3051
3052 /*
3053  * This will check the space that the inode allocates from to make sure we have
3054  * enough space for bytes.
3055  */
3056 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3057 {
3058         struct btrfs_space_info *data_sinfo;
3059         struct btrfs_root *root = BTRFS_I(inode)->root;
3060         u64 used;
3061         int ret = 0, committed = 0, alloc_chunk = 1;
3062
3063         /* make sure bytes are sectorsize aligned */
3064         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3065
3066         if (root == root->fs_info->tree_root) {
3067                 alloc_chunk = 0;
3068                 committed = 1;
3069         }
3070
3071         data_sinfo = BTRFS_I(inode)->space_info;
3072         if (!data_sinfo)
3073                 goto alloc;
3074
3075 again:
3076         /* make sure we have enough space to handle the data first */
3077         spin_lock(&data_sinfo->lock);
3078         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3079                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3080                 data_sinfo->bytes_may_use;
3081
3082         if (used + bytes > data_sinfo->total_bytes) {
3083                 struct btrfs_trans_handle *trans;
3084
3085                 /*
3086                  * if we don't have enough free bytes in this space then we need
3087                  * to alloc a new chunk.
3088                  */
3089                 if (!data_sinfo->full && alloc_chunk) {
3090                         u64 alloc_target;
3091
3092                         data_sinfo->force_alloc = 1;
3093                         spin_unlock(&data_sinfo->lock);
3094 alloc:
3095                         alloc_target = btrfs_get_alloc_profile(root, 1);
3096                         trans = btrfs_join_transaction(root, 1);
3097                         if (IS_ERR(trans))
3098                                 return PTR_ERR(trans);
3099
3100                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3101                                              bytes + 2 * 1024 * 1024,
3102                                              alloc_target, 0);
3103                         btrfs_end_transaction(trans, root);
3104                         if (ret < 0)
3105                                 return ret;
3106
3107                         if (!data_sinfo) {
3108                                 btrfs_set_inode_space_info(root, inode);
3109                                 data_sinfo = BTRFS_I(inode)->space_info;
3110                         }
3111                         goto again;
3112                 }
3113                 spin_unlock(&data_sinfo->lock);
3114
3115                 /* commit the current transaction and try again */
3116                 if (!committed && !root->fs_info->open_ioctl_trans) {
3117                         committed = 1;
3118                         trans = btrfs_join_transaction(root, 1);
3119                         if (IS_ERR(trans))
3120                                 return PTR_ERR(trans);
3121                         ret = btrfs_commit_transaction(trans, root);
3122                         if (ret)
3123                                 return ret;
3124                         goto again;
3125                 }
3126
3127 #if 0 /* I hope we never need this code again, just in case */
3128                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3129                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3130                        "%llu bytes_readonly, %llu may use %llu total\n",
3131                        (unsigned long long)bytes,
3132                        (unsigned long long)data_sinfo->bytes_used,
3133                        (unsigned long long)data_sinfo->bytes_reserved,
3134                        (unsigned long long)data_sinfo->bytes_pinned,
3135                        (unsigned long long)data_sinfo->bytes_readonly,
3136                        (unsigned long long)data_sinfo->bytes_may_use,
3137                        (unsigned long long)data_sinfo->total_bytes);
3138 #endif
3139                 return -ENOSPC;
3140         }
3141         data_sinfo->bytes_may_use += bytes;
3142         BTRFS_I(inode)->reserved_bytes += bytes;
3143         spin_unlock(&data_sinfo->lock);
3144
3145         return 0;
3146 }
3147
3148 /*
3149  * called when we are clearing an delalloc extent from the
3150  * inode's io_tree or there was an error for whatever reason
3151  * after calling btrfs_check_data_free_space
3152  */
3153 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3154 {
3155         struct btrfs_root *root = BTRFS_I(inode)->root;
3156         struct btrfs_space_info *data_sinfo;
3157
3158         /* make sure bytes are sectorsize aligned */
3159         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3160
3161         data_sinfo = BTRFS_I(inode)->space_info;
3162         spin_lock(&data_sinfo->lock);
3163         data_sinfo->bytes_may_use -= bytes;
3164         BTRFS_I(inode)->reserved_bytes -= bytes;
3165         spin_unlock(&data_sinfo->lock);
3166 }
3167
3168 static void force_metadata_allocation(struct btrfs_fs_info *info)
3169 {
3170         struct list_head *head = &info->space_info;
3171         struct btrfs_space_info *found;
3172
3173         rcu_read_lock();
3174         list_for_each_entry_rcu(found, head, list) {
3175                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3176                         found->force_alloc = 1;
3177         }
3178         rcu_read_unlock();
3179 }
3180
3181 static int should_alloc_chunk(struct btrfs_space_info *sinfo,
3182                               u64 alloc_bytes)
3183 {
3184         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3185
3186         if (sinfo->bytes_used + sinfo->bytes_reserved +
3187             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3188                 return 0;
3189
3190         if (sinfo->bytes_used + sinfo->bytes_reserved +
3191             alloc_bytes < div_factor(num_bytes, 8))
3192                 return 0;
3193
3194         return 1;
3195 }
3196
3197 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3198                           struct btrfs_root *extent_root, u64 alloc_bytes,
3199                           u64 flags, int force)
3200 {
3201         struct btrfs_space_info *space_info;
3202         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3203         int ret = 0;
3204
3205         mutex_lock(&fs_info->chunk_mutex);
3206
3207         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3208
3209         space_info = __find_space_info(extent_root->fs_info, flags);
3210         if (!space_info) {
3211                 ret = update_space_info(extent_root->fs_info, flags,
3212                                         0, 0, &space_info);
3213                 BUG_ON(ret);
3214         }
3215         BUG_ON(!space_info);
3216
3217         spin_lock(&space_info->lock);
3218         if (space_info->force_alloc)
3219                 force = 1;
3220         if (space_info->full) {
3221                 spin_unlock(&space_info->lock);
3222                 goto out;
3223         }
3224
3225         if (!force && !should_alloc_chunk(space_info, alloc_bytes)) {
3226                 spin_unlock(&space_info->lock);
3227                 goto out;
3228         }
3229         spin_unlock(&space_info->lock);
3230
3231         /*
3232          * if we're doing a data chunk, go ahead and make sure that
3233          * we keep a reasonable number of metadata chunks allocated in the
3234          * FS as well.
3235          */
3236         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3237                 fs_info->data_chunk_allocations++;
3238                 if (!(fs_info->data_chunk_allocations %
3239                       fs_info->metadata_ratio))
3240                         force_metadata_allocation(fs_info);
3241         }
3242
3243         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3244         spin_lock(&space_info->lock);
3245         if (ret)
3246                 space_info->full = 1;
3247         else
3248                 ret = 1;
3249         space_info->force_alloc = 0;
3250         spin_unlock(&space_info->lock);
3251 out:
3252         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3253         return ret;
3254 }
3255
3256 static int maybe_allocate_chunk(struct btrfs_trans_handle *trans,
3257                                 struct btrfs_root *root,
3258                                 struct btrfs_space_info *sinfo, u64 num_bytes)
3259 {
3260         int ret;
3261         int end_trans = 0;
3262
3263         if (sinfo->full)
3264                 return 0;
3265
3266         spin_lock(&sinfo->lock);
3267         ret = should_alloc_chunk(sinfo, num_bytes + 2 * 1024 * 1024);
3268         spin_unlock(&sinfo->lock);
3269         if (!ret)
3270                 return 0;
3271
3272         if (!trans) {
3273                 trans = btrfs_join_transaction(root, 1);
3274                 BUG_ON(IS_ERR(trans));
3275                 end_trans = 1;
3276         }
3277
3278         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3279                              num_bytes + 2 * 1024 * 1024,
3280                              get_alloc_profile(root, sinfo->flags), 0);
3281
3282         if (end_trans)
3283                 btrfs_end_transaction(trans, root);
3284
3285         return ret == 1 ? 1 : 0;
3286 }
3287
3288 /*
3289  * shrink metadata reservation for delalloc
3290  */
3291 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3292                            struct btrfs_root *root, u64 to_reclaim)
3293 {
3294         struct btrfs_block_rsv *block_rsv;
3295         u64 reserved;
3296         u64 max_reclaim;
3297         u64 reclaimed = 0;
3298         int pause = 1;
3299         int ret;
3300
3301         block_rsv = &root->fs_info->delalloc_block_rsv;
3302         spin_lock(&block_rsv->lock);
3303         reserved = block_rsv->reserved;
3304         spin_unlock(&block_rsv->lock);
3305
3306         if (reserved == 0)
3307                 return 0;
3308
3309         max_reclaim = min(reserved, to_reclaim);
3310
3311         while (1) {
3312                 ret = btrfs_start_one_delalloc_inode(root, trans ? 1 : 0);
3313                 if (!ret) {
3314                         __set_current_state(TASK_INTERRUPTIBLE);
3315                         schedule_timeout(pause);
3316                         pause <<= 1;
3317                         if (pause > HZ / 10)
3318                                 pause = HZ / 10;
3319                 } else {
3320                         pause = 1;
3321                 }
3322
3323                 spin_lock(&block_rsv->lock);
3324                 if (reserved > block_rsv->reserved)
3325                         reclaimed = reserved - block_rsv->reserved;
3326                 reserved = block_rsv->reserved;
3327                 spin_unlock(&block_rsv->lock);
3328
3329                 if (reserved == 0 || reclaimed >= max_reclaim)
3330                         break;
3331
3332                 if (trans && trans->transaction->blocked)
3333                         return -EAGAIN;
3334         }
3335         return reclaimed >= to_reclaim;
3336 }
3337
3338 static int should_retry_reserve(struct btrfs_trans_handle *trans,
3339                                 struct btrfs_root *root,
3340                                 struct btrfs_block_rsv *block_rsv,
3341                                 u64 num_bytes, int *retries)
3342 {
3343         struct btrfs_space_info *space_info = block_rsv->space_info;
3344         int ret;
3345
3346         if ((*retries) > 2)
3347                 return -ENOSPC;
3348
3349         ret = maybe_allocate_chunk(trans, root, space_info, num_bytes);
3350         if (ret)
3351                 return 1;
3352
3353         if (trans && trans->transaction->in_commit)
3354                 return -ENOSPC;
3355
3356         ret = shrink_delalloc(trans, root, num_bytes);
3357         if (ret)
3358                 return ret;
3359
3360         spin_lock(&space_info->lock);
3361         if (space_info->bytes_pinned < num_bytes)
3362                 ret = 1;
3363         spin_unlock(&space_info->lock);
3364         if (ret)
3365                 return -ENOSPC;
3366
3367         (*retries)++;
3368
3369         if (trans)
3370                 return -EAGAIN;
3371
3372         trans = btrfs_join_transaction(root, 1);
3373         BUG_ON(IS_ERR(trans));
3374         ret = btrfs_commit_transaction(trans, root);
3375         BUG_ON(ret);
3376
3377         return 1;
3378 }
3379
3380 static int reserve_metadata_bytes(struct btrfs_block_rsv *block_rsv,
3381                                   u64 num_bytes)
3382 {
3383         struct btrfs_space_info *space_info = block_rsv->space_info;
3384         u64 unused;
3385         int ret = -ENOSPC;
3386
3387         spin_lock(&space_info->lock);
3388         unused = space_info->bytes_used + space_info->bytes_reserved +
3389                  space_info->bytes_pinned + space_info->bytes_readonly;
3390
3391         if (unused < space_info->total_bytes)
3392                 unused = space_info->total_bytes - unused;
3393         else
3394                 unused = 0;
3395
3396         if (unused >= num_bytes) {
3397                 if (block_rsv->priority >= 10) {
3398                         space_info->bytes_reserved += num_bytes;
3399                         ret = 0;
3400                 } else {
3401                         if ((unused + block_rsv->reserved) *
3402                             block_rsv->priority >=
3403                             (num_bytes + block_rsv->reserved) * 10) {
3404                                 space_info->bytes_reserved += num_bytes;
3405                                 ret = 0;
3406                         }
3407                 }
3408         }
3409         spin_unlock(&space_info->lock);
3410
3411         return ret;
3412 }
3413
3414 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3415                                              struct btrfs_root *root)
3416 {
3417         struct btrfs_block_rsv *block_rsv;
3418         if (root->ref_cows)
3419                 block_rsv = trans->block_rsv;
3420         else
3421                 block_rsv = root->block_rsv;
3422
3423         if (!block_rsv)
3424                 block_rsv = &root->fs_info->empty_block_rsv;
3425
3426         return block_rsv;
3427 }
3428
3429 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3430                                u64 num_bytes)
3431 {
3432         int ret = -ENOSPC;
3433         spin_lock(&block_rsv->lock);
3434         if (block_rsv->reserved >= num_bytes) {
3435                 block_rsv->reserved -= num_bytes;
3436                 if (block_rsv->reserved < block_rsv->size)
3437                         block_rsv->full = 0;
3438                 ret = 0;
3439         }
3440         spin_unlock(&block_rsv->lock);
3441         return ret;
3442 }
3443
3444 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3445                                 u64 num_bytes, int update_size)
3446 {
3447         spin_lock(&block_rsv->lock);
3448         block_rsv->reserved += num_bytes;
3449         if (update_size)
3450                 block_rsv->size += num_bytes;
3451         else if (block_rsv->reserved >= block_rsv->size)
3452                 block_rsv->full = 1;
3453         spin_unlock(&block_rsv->lock);
3454 }
3455
3456 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3457                              struct btrfs_block_rsv *dest, u64 num_bytes)
3458 {
3459         struct btrfs_space_info *space_info = block_rsv->space_info;
3460
3461         spin_lock(&block_rsv->lock);
3462         if (num_bytes == (u64)-1)
3463                 num_bytes = block_rsv->size;
3464         block_rsv->size -= num_bytes;
3465         if (block_rsv->reserved >= block_rsv->size) {
3466                 num_bytes = block_rsv->reserved - block_rsv->size;
3467                 block_rsv->reserved = block_rsv->size;
3468                 block_rsv->full = 1;
3469         } else {
3470                 num_bytes = 0;
3471         }
3472         spin_unlock(&block_rsv->lock);
3473
3474         if (num_bytes > 0) {
3475                 if (dest) {
3476                         block_rsv_add_bytes(dest, num_bytes, 0);
3477                 } else {
3478                         spin_lock(&space_info->lock);
3479                         space_info->bytes_reserved -= num_bytes;
3480                         spin_unlock(&space_info->lock);
3481                 }
3482         }
3483 }
3484
3485 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3486                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3487 {
3488         int ret;
3489
3490         ret = block_rsv_use_bytes(src, num_bytes);
3491         if (ret)
3492                 return ret;
3493
3494         block_rsv_add_bytes(dst, num_bytes, 1);
3495         return 0;
3496 }
3497
3498 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3499 {
3500         memset(rsv, 0, sizeof(*rsv));
3501         spin_lock_init(&rsv->lock);
3502         atomic_set(&rsv->usage, 1);
3503         rsv->priority = 6;
3504         INIT_LIST_HEAD(&rsv->list);
3505 }
3506
3507 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3508 {
3509         struct btrfs_block_rsv *block_rsv;
3510         struct btrfs_fs_info *fs_info = root->fs_info;
3511         u64 alloc_target;
3512
3513         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3514         if (!block_rsv)
3515                 return NULL;
3516
3517         btrfs_init_block_rsv(block_rsv);
3518
3519         alloc_target = btrfs_get_alloc_profile(root, 0);
3520         block_rsv->space_info = __find_space_info(fs_info,
3521                                                   BTRFS_BLOCK_GROUP_METADATA);
3522
3523         return block_rsv;
3524 }
3525
3526 void btrfs_free_block_rsv(struct btrfs_root *root,
3527                           struct btrfs_block_rsv *rsv)
3528 {
3529         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3530                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3531                 if (!rsv->durable)
3532                         kfree(rsv);
3533         }
3534 }
3535
3536 /*
3537  * make the block_rsv struct be able to capture freed space.
3538  * the captured space will re-add to the the block_rsv struct
3539  * after transaction commit
3540  */
3541 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3542                                  struct btrfs_block_rsv *block_rsv)
3543 {
3544         block_rsv->durable = 1;
3545         mutex_lock(&fs_info->durable_block_rsv_mutex);
3546         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3547         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3548 }
3549
3550 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3551                         struct btrfs_root *root,
3552                         struct btrfs_block_rsv *block_rsv,
3553                         u64 num_bytes, int *retries)
3554 {
3555         int ret;
3556
3557         if (num_bytes == 0)
3558                 return 0;
3559 again:
3560         ret = reserve_metadata_bytes(block_rsv, num_bytes);
3561         if (!ret) {
3562                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3563                 return 0;
3564         }
3565
3566         ret = should_retry_reserve(trans, root, block_rsv, num_bytes, retries);
3567         if (ret > 0)
3568                 goto again;
3569
3570         return ret;
3571 }
3572
3573 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3574                           struct btrfs_root *root,
3575                           struct btrfs_block_rsv *block_rsv,
3576                           u64 min_reserved, int min_factor)
3577 {
3578         u64 num_bytes = 0;
3579         int commit_trans = 0;
3580         int ret = -ENOSPC;
3581
3582         if (!block_rsv)
3583                 return 0;
3584
3585         spin_lock(&block_rsv->lock);
3586         if (min_factor > 0)
3587                 num_bytes = div_factor(block_rsv->size, min_factor);
3588         if (min_reserved > num_bytes)
3589                 num_bytes = min_reserved;
3590
3591         if (block_rsv->reserved >= num_bytes) {
3592                 ret = 0;
3593         } else {
3594                 num_bytes -= block_rsv->reserved;
3595                 if (block_rsv->durable &&
3596                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3597                         commit_trans = 1;
3598         }
3599         spin_unlock(&block_rsv->lock);
3600         if (!ret)
3601                 return 0;
3602
3603         if (block_rsv->refill_used) {
3604                 ret = reserve_metadata_bytes(block_rsv, num_bytes);
3605                 if (!ret) {
3606                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3607                         return 0;
3608                 }
3609         }
3610
3611         if (commit_trans) {
3612                 if (trans)
3613                         return -EAGAIN;
3614
3615                 trans = btrfs_join_transaction(root, 1);
3616                 BUG_ON(IS_ERR(trans));
3617                 ret = btrfs_commit_transaction(trans, root);
3618                 return 0;
3619         }
3620
3621         WARN_ON(1);
3622         printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
3623                 block_rsv->size, block_rsv->reserved,
3624                 block_rsv->freed[0], block_rsv->freed[1]);
3625
3626         return -ENOSPC;
3627 }
3628
3629 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3630                             struct btrfs_block_rsv *dst_rsv,
3631                             u64 num_bytes)
3632 {
3633         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3634 }
3635
3636 void btrfs_block_rsv_release(struct btrfs_root *root,
3637                              struct btrfs_block_rsv *block_rsv,
3638                              u64 num_bytes)
3639 {
3640         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3641         if (global_rsv->full || global_rsv == block_rsv ||
3642             block_rsv->space_info != global_rsv->space_info)
3643                 global_rsv = NULL;
3644         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3645 }
3646
3647 /*
3648  * helper to calculate size of global block reservation.
3649  * the desired value is sum of space used by extent tree,
3650  * checksum tree and root tree
3651  */
3652 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3653 {
3654         struct btrfs_space_info *sinfo;
3655         u64 num_bytes;
3656         u64 meta_used;
3657         u64 data_used;
3658         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3659 #if 0
3660         /*
3661          * per tree used space accounting can be inaccuracy, so we
3662          * can't rely on it.
3663          */
3664         spin_lock(&fs_info->extent_root->accounting_lock);
3665         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3666         spin_unlock(&fs_info->extent_root->accounting_lock);
3667
3668         spin_lock(&fs_info->csum_root->accounting_lock);
3669         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3670         spin_unlock(&fs_info->csum_root->accounting_lock);
3671
3672         spin_lock(&fs_info->tree_root->accounting_lock);
3673         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3674         spin_unlock(&fs_info->tree_root->accounting_lock);
3675 #endif
3676         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3677         spin_lock(&sinfo->lock);
3678         data_used = sinfo->bytes_used;
3679         spin_unlock(&sinfo->lock);
3680
3681         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3682         spin_lock(&sinfo->lock);
3683         meta_used = sinfo->bytes_used;
3684         spin_unlock(&sinfo->lock);
3685
3686         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3687                     csum_size * 2;
3688         num_bytes += div64_u64(data_used + meta_used, 50);
3689
3690         if (num_bytes * 3 > meta_used)
3691                 num_bytes = div64_u64(meta_used, 3);
3692
3693         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3694 }
3695
3696 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3697 {
3698         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3699         struct btrfs_space_info *sinfo = block_rsv->space_info;
3700         u64 num_bytes;
3701
3702         num_bytes = calc_global_metadata_size(fs_info);
3703
3704         spin_lock(&block_rsv->lock);
3705         spin_lock(&sinfo->lock);
3706
3707         block_rsv->size = num_bytes;
3708
3709         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3710                     sinfo->bytes_reserved + sinfo->bytes_readonly;
3711
3712         if (sinfo->total_bytes > num_bytes) {
3713                 num_bytes = sinfo->total_bytes - num_bytes;
3714                 block_rsv->reserved += num_bytes;
3715                 sinfo->bytes_reserved += num_bytes;
3716         }
3717
3718         if (block_rsv->reserved >= block_rsv->size) {
3719                 num_bytes = block_rsv->reserved - block_rsv->size;
3720                 sinfo->bytes_reserved -= num_bytes;
3721                 block_rsv->reserved = block_rsv->size;
3722                 block_rsv->full = 1;
3723         }
3724 #if 0
3725         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3726                 block_rsv->size, block_rsv->reserved);
3727 #endif
3728         spin_unlock(&sinfo->lock);
3729         spin_unlock(&block_rsv->lock);
3730 }
3731
3732 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3733 {
3734         struct btrfs_space_info *space_info;
3735
3736         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3737         fs_info->chunk_block_rsv.space_info = space_info;
3738         fs_info->chunk_block_rsv.priority = 10;
3739
3740         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3741         fs_info->global_block_rsv.space_info = space_info;
3742         fs_info->global_block_rsv.priority = 10;
3743         fs_info->global_block_rsv.refill_used = 1;
3744         fs_info->delalloc_block_rsv.space_info = space_info;
3745         fs_info->trans_block_rsv.space_info = space_info;
3746         fs_info->empty_block_rsv.space_info = space_info;
3747         fs_info->empty_block_rsv.priority = 10;
3748
3749         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3750         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3751         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3752         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3753         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3754
3755         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3756
3757         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3758
3759         update_global_block_rsv(fs_info);
3760 }
3761
3762 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3763 {
3764         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3765         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3766         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3767         WARN_ON(fs_info->trans_block_rsv.size > 0);
3768         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3769         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3770         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3771 }
3772
3773 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3774 {
3775         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3776                 3 * num_items;
3777 }
3778
3779 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3780                                  struct btrfs_root *root,
3781                                  int num_items, int *retries)
3782 {
3783         u64 num_bytes;
3784         int ret;
3785
3786         if (num_items == 0 || root->fs_info->chunk_root == root)
3787                 return 0;
3788
3789         num_bytes = calc_trans_metadata_size(root, num_items);
3790         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3791                                   num_bytes, retries);
3792         if (!ret) {
3793                 trans->bytes_reserved += num_bytes;
3794                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3795         }
3796         return ret;
3797 }
3798
3799 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3800                                   struct btrfs_root *root)
3801 {
3802         if (!trans->bytes_reserved)
3803                 return;
3804
3805         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3806         btrfs_block_rsv_release(root, trans->block_rsv,
3807                                 trans->bytes_reserved);
3808         trans->bytes_reserved = 0;
3809 }
3810
3811 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3812                                   struct inode *inode)
3813 {
3814         struct btrfs_root *root = BTRFS_I(inode)->root;
3815         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3816         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3817
3818         /*
3819          * one for deleting orphan item, one for updating inode and
3820          * two for calling btrfs_truncate_inode_items.
3821          *
3822          * btrfs_truncate_inode_items is a delete operation, it frees
3823          * more space than it uses in most cases. So two units of
3824          * metadata space should be enough for calling it many times.
3825          * If all of the metadata space is used, we can commit
3826          * transaction and use space it freed.
3827          */
3828         u64 num_bytes = calc_trans_metadata_size(root, 4);
3829         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3830 }
3831
3832 void btrfs_orphan_release_metadata(struct inode *inode)
3833 {
3834         struct btrfs_root *root = BTRFS_I(inode)->root;
3835         u64 num_bytes = calc_trans_metadata_size(root, 4);
3836         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3837 }
3838
3839 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3840                                 struct btrfs_pending_snapshot *pending)
3841 {
3842         struct btrfs_root *root = pending->root;
3843         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3844         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3845         /*
3846          * two for root back/forward refs, two for directory entries
3847          * and one for root of the snapshot.
3848          */
3849         u64 num_bytes = calc_trans_metadata_size(root, 5);
3850         dst_rsv->space_info = src_rsv->space_info;
3851         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3852 }
3853
3854 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3855 {
3856         return num_bytes >>= 3;
3857 }
3858
3859 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3860 {
3861         struct btrfs_root *root = BTRFS_I(inode)->root;
3862         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3863         u64 to_reserve;
3864         int nr_extents;
3865         int retries = 0;
3866         int ret;
3867
3868         if (btrfs_transaction_in_commit(root->fs_info))
3869                 schedule_timeout(1);
3870
3871         num_bytes = ALIGN(num_bytes, root->sectorsize);
3872 again:
3873         spin_lock(&BTRFS_I(inode)->accounting_lock);
3874         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
3875         if (nr_extents > BTRFS_I(inode)->reserved_extents) {
3876                 nr_extents -= BTRFS_I(inode)->reserved_extents;
3877                 to_reserve = calc_trans_metadata_size(root, nr_extents);
3878         } else {
3879                 nr_extents = 0;
3880                 to_reserve = 0;
3881         }
3882
3883         to_reserve += calc_csum_metadata_size(inode, num_bytes);
3884         ret = reserve_metadata_bytes(block_rsv, to_reserve);
3885         if (ret) {
3886                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3887                 ret = should_retry_reserve(NULL, root, block_rsv, to_reserve,
3888                                            &retries);
3889                 if (ret > 0)
3890                         goto again;
3891                 return ret;
3892         }
3893
3894         BTRFS_I(inode)->reserved_extents += nr_extents;
3895         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
3896         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3897
3898         block_rsv_add_bytes(block_rsv, to_reserve, 1);
3899
3900         if (block_rsv->size > 512 * 1024 * 1024)
3901                 shrink_delalloc(NULL, root, to_reserve);
3902
3903         return 0;
3904 }
3905
3906 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
3907 {
3908         struct btrfs_root *root = BTRFS_I(inode)->root;
3909         u64 to_free;
3910         int nr_extents;
3911
3912         num_bytes = ALIGN(num_bytes, root->sectorsize);
3913         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
3914
3915         spin_lock(&BTRFS_I(inode)->accounting_lock);
3916         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
3917         if (nr_extents < BTRFS_I(inode)->reserved_extents) {
3918                 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
3919                 BTRFS_I(inode)->reserved_extents -= nr_extents;
3920         } else {
3921                 nr_extents = 0;
3922         }
3923         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3924
3925         to_free = calc_csum_metadata_size(inode, num_bytes);
3926         if (nr_extents > 0)
3927                 to_free += calc_trans_metadata_size(root, nr_extents);
3928
3929         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
3930                                 to_free);
3931 }
3932
3933 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
3934 {
3935         int ret;
3936
3937         ret = btrfs_check_data_free_space(inode, num_bytes);
3938         if (ret)
3939                 return ret;
3940
3941         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
3942         if (ret) {
3943                 btrfs_free_reserved_data_space(inode, num_bytes);
3944                 return ret;
3945         }
3946
3947         return 0;
3948 }
3949
3950 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
3951 {
3952         btrfs_delalloc_release_metadata(inode, num_bytes);
3953         btrfs_free_reserved_data_space(inode, num_bytes);
3954 }
3955
3956 static int update_block_group(struct btrfs_trans_handle *trans,
3957                               struct btrfs_root *root,
3958                               u64 bytenr, u64 num_bytes, int alloc)
3959 {
3960         struct btrfs_block_group_cache *cache = NULL;
3961         struct btrfs_fs_info *info = root->fs_info;
3962         u64 total = num_bytes;
3963         u64 old_val;
3964         u64 byte_in_group;
3965         int factor;
3966
3967         /* block accounting for super block */
3968         spin_lock(&info->delalloc_lock);
3969         old_val = btrfs_super_bytes_used(&info->super_copy);
3970         if (alloc)
3971                 old_val += num_bytes;
3972         else
3973                 old_val -= num_bytes;
3974         btrfs_set_super_bytes_used(&info->super_copy, old_val);
3975         spin_unlock(&info->delalloc_lock);
3976
3977         while (total) {
3978                 cache = btrfs_lookup_block_group(info, bytenr);
3979                 if (!cache)
3980                         return -1;
3981                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
3982                                     BTRFS_BLOCK_GROUP_RAID1 |
3983                                     BTRFS_BLOCK_GROUP_RAID10))
3984                         factor = 2;
3985                 else
3986                         factor = 1;
3987
3988                 byte_in_group = bytenr - cache->key.objectid;
3989                 WARN_ON(byte_in_group > cache->key.offset);
3990
3991                 spin_lock(&cache->space_info->lock);
3992                 spin_lock(&cache->lock);
3993
3994                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
3995                     cache->disk_cache_state < BTRFS_DC_CLEAR)
3996                         cache->disk_cache_state = BTRFS_DC_CLEAR;
3997
3998                 cache->dirty = 1;
3999                 old_val = btrfs_block_group_used(&cache->item);
4000                 num_bytes = min(total, cache->key.offset - byte_in_group);
4001                 if (alloc) {
4002                         old_val += num_bytes;
4003                         btrfs_set_block_group_used(&cache->item, old_val);
4004                         cache->reserved -= num_bytes;
4005                         cache->space_info->bytes_reserved -= num_bytes;
4006                         cache->space_info->bytes_used += num_bytes;
4007                         cache->space_info->disk_used += num_bytes * factor;
4008                         spin_unlock(&cache->lock);
4009                         spin_unlock(&cache->space_info->lock);
4010                 } else {
4011                         old_val -= num_bytes;
4012                         btrfs_set_block_group_used(&cache->item, old_val);
4013                         cache->pinned += num_bytes;
4014                         cache->space_info->bytes_pinned += num_bytes;
4015                         cache->space_info->bytes_used -= num_bytes;
4016                         cache->space_info->disk_used -= num_bytes * factor;
4017                         spin_unlock(&cache->lock);
4018                         spin_unlock(&cache->space_info->lock);
4019
4020                         set_extent_dirty(info->pinned_extents,
4021                                          bytenr, bytenr + num_bytes - 1,
4022                                          GFP_NOFS | __GFP_NOFAIL);
4023                 }
4024                 btrfs_put_block_group(cache);
4025                 total -= num_bytes;
4026                 bytenr += num_bytes;
4027         }
4028         return 0;
4029 }
4030
4031 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4032 {
4033         struct btrfs_block_group_cache *cache;
4034         u64 bytenr;
4035
4036         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4037         if (!cache)
4038                 return 0;
4039
4040         bytenr = cache->key.objectid;
4041         btrfs_put_block_group(cache);
4042
4043         return bytenr;
4044 }
4045
4046 static int pin_down_extent(struct btrfs_root *root,
4047                            struct btrfs_block_group_cache *cache,
4048                            u64 bytenr, u64 num_bytes, int reserved)
4049 {
4050         spin_lock(&cache->space_info->lock);
4051         spin_lock(&cache->lock);
4052         cache->pinned += num_bytes;
4053         cache->space_info->bytes_pinned += num_bytes;
4054         if (reserved) {
4055                 cache->reserved -= num_bytes;
4056                 cache->space_info->bytes_reserved -= num_bytes;
4057         }
4058         spin_unlock(&cache->lock);
4059         spin_unlock(&cache->space_info->lock);
4060
4061         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4062                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4063         return 0;
4064 }
4065
4066 /*
4067  * this function must be called within transaction
4068  */
4069 int btrfs_pin_extent(struct btrfs_root *root,
4070                      u64 bytenr, u64 num_bytes, int reserved)
4071 {
4072         struct btrfs_block_group_cache *cache;
4073
4074         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4075         BUG_ON(!cache);
4076
4077         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4078
4079         btrfs_put_block_group(cache);
4080         return 0;
4081 }
4082
4083 /*
4084  * update size of reserved extents. this function may return -EAGAIN
4085  * if 'reserve' is true or 'sinfo' is false.
4086  */
4087 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4088                                  u64 num_bytes, int reserve, int sinfo)
4089 {
4090         int ret = 0;
4091         if (sinfo) {
4092                 struct btrfs_space_info *space_info = cache->space_info;
4093                 spin_lock(&space_info->lock);
4094                 spin_lock(&cache->lock);
4095                 if (reserve) {
4096                         if (cache->ro) {
4097                                 ret = -EAGAIN;
4098                         } else {
4099                                 cache->reserved += num_bytes;
4100                                 space_info->bytes_reserved += num_bytes;
4101                         }
4102                 } else {
4103                         if (cache->ro)
4104                                 space_info->bytes_readonly += num_bytes;
4105                         cache->reserved -= num_bytes;
4106                         space_info->bytes_reserved -= num_bytes;
4107                 }
4108                 spin_unlock(&cache->lock);
4109                 spin_unlock(&space_info->lock);
4110         } else {
4111                 spin_lock(&cache->lock);
4112                 if (cache->ro) {
4113                         ret = -EAGAIN;
4114                 } else {
4115                         if (reserve)
4116                                 cache->reserved += num_bytes;
4117                         else
4118                                 cache->reserved -= num_bytes;
4119                 }
4120                 spin_unlock(&cache->lock);
4121         }
4122         return ret;
4123 }
4124
4125 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4126                                 struct btrfs_root *root)
4127 {
4128         struct btrfs_fs_info *fs_info = root->fs_info;
4129         struct btrfs_caching_control *next;
4130         struct btrfs_caching_control *caching_ctl;
4131         struct btrfs_block_group_cache *cache;
4132
4133         down_write(&fs_info->extent_commit_sem);
4134
4135         list_for_each_entry_safe(caching_ctl, next,
4136                                  &fs_info->caching_block_groups, list) {
4137                 cache = caching_ctl->block_group;
4138                 if (block_group_cache_done(cache)) {
4139                         cache->last_byte_to_unpin = (u64)-1;
4140                         list_del_init(&caching_ctl->list);
4141                         put_caching_control(caching_ctl);
4142                 } else {
4143                         cache->last_byte_to_unpin = caching_ctl->progress;
4144                 }
4145         }
4146
4147         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4148                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4149         else
4150                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4151
4152         up_write(&fs_info->extent_commit_sem);
4153
4154         update_global_block_rsv(fs_info);
4155         return 0;
4156 }
4157
4158 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4159 {
4160         struct btrfs_fs_info *fs_info = root->fs_info;
4161         struct btrfs_block_group_cache *cache = NULL;
4162         u64 len;
4163
4164         while (start <= end) {
4165                 if (!cache ||
4166                     start >= cache->key.objectid + cache->key.offset) {
4167                         if (cache)
4168                                 btrfs_put_block_group(cache);
4169                         cache = btrfs_lookup_block_group(fs_info, start);
4170                         BUG_ON(!cache);
4171                 }
4172
4173                 len = cache->key.objectid + cache->key.offset - start;
4174                 len = min(len, end + 1 - start);
4175
4176                 if (start < cache->last_byte_to_unpin) {
4177                         len = min(len, cache->last_byte_to_unpin - start);
4178                         btrfs_add_free_space(cache, start, len);
4179                 }
4180
4181                 start += len;
4182
4183                 spin_lock(&cache->space_info->lock);
4184                 spin_lock(&cache->lock);
4185                 cache->pinned -= len;
4186                 cache->space_info->bytes_pinned -= len;
4187                 if (cache->ro) {
4188                         cache->space_info->bytes_readonly += len;
4189                 } else if (cache->reserved_pinned > 0) {
4190                         len = min(len, cache->reserved_pinned);
4191                         cache->reserved_pinned -= len;
4192                         cache->space_info->bytes_reserved += len;
4193                 }
4194                 spin_unlock(&cache->lock);
4195                 spin_unlock(&cache->space_info->lock);
4196         }
4197
4198         if (cache)
4199                 btrfs_put_block_group(cache);
4200         return 0;
4201 }
4202
4203 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4204                                struct btrfs_root *root)
4205 {
4206         struct btrfs_fs_info *fs_info = root->fs_info;
4207         struct extent_io_tree *unpin;
4208         struct btrfs_block_rsv *block_rsv;
4209         struct btrfs_block_rsv *next_rsv;
4210         u64 start;
4211         u64 end;
4212         int idx;
4213         int ret;
4214
4215         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4216                 unpin = &fs_info->freed_extents[1];
4217         else
4218                 unpin = &fs_info->freed_extents[0];
4219
4220         while (1) {
4221                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4222                                             EXTENT_DIRTY);
4223                 if (ret)
4224                         break;
4225
4226                 ret = btrfs_discard_extent(root, start, end + 1 - start);
4227
4228                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4229                 unpin_extent_range(root, start, end);
4230                 cond_resched();
4231         }
4232
4233         mutex_lock(&fs_info->durable_block_rsv_mutex);
4234         list_for_each_entry_safe(block_rsv, next_rsv,
4235                                  &fs_info->durable_block_rsv_list, list) {
4236
4237                 idx = trans->transid & 0x1;
4238                 if (block_rsv->freed[idx] > 0) {
4239                         block_rsv_add_bytes(block_rsv,
4240                                             block_rsv->freed[idx], 0);
4241                         block_rsv->freed[idx] = 0;
4242                 }
4243                 if (atomic_read(&block_rsv->usage) == 0) {
4244                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4245
4246                         if (block_rsv->freed[0] == 0 &&
4247                             block_rsv->freed[1] == 0) {
4248                                 list_del_init(&block_rsv->list);
4249                                 kfree(block_rsv);
4250                         }
4251                 } else {
4252                         btrfs_block_rsv_release(root, block_rsv, 0);
4253                 }
4254         }
4255         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4256
4257         return 0;
4258 }
4259
4260 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4261                                 struct btrfs_root *root,
4262                                 u64 bytenr, u64 num_bytes, u64 parent,
4263                                 u64 root_objectid, u64 owner_objectid,
4264                                 u64 owner_offset, int refs_to_drop,
4265                                 struct btrfs_delayed_extent_op *extent_op)
4266 {
4267         struct btrfs_key key;
4268         struct btrfs_path *path;
4269         struct btrfs_fs_info *info = root->fs_info;
4270         struct btrfs_root *extent_root = info->extent_root;
4271         struct extent_buffer *leaf;
4272         struct btrfs_extent_item *ei;
4273         struct btrfs_extent_inline_ref *iref;
4274         int ret;
4275         int is_data;
4276         int extent_slot = 0;
4277         int found_extent = 0;
4278         int num_to_del = 1;
4279         u32 item_size;
4280         u64 refs;
4281
4282         path = btrfs_alloc_path();
4283         if (!path)
4284                 return -ENOMEM;
4285
4286         path->reada = 1;
4287         path->leave_spinning = 1;
4288
4289         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4290         BUG_ON(!is_data && refs_to_drop != 1);
4291
4292         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4293                                     bytenr, num_bytes, parent,
4294                                     root_objectid, owner_objectid,
4295                                     owner_offset);
4296         if (ret == 0) {
4297                 extent_slot = path->slots[0];
4298                 while (extent_slot >= 0) {
4299                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4300                                               extent_slot);
4301                         if (key.objectid != bytenr)
4302                                 break;
4303                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4304                             key.offset == num_bytes) {
4305                                 found_extent = 1;
4306                                 break;
4307                         }
4308                         if (path->slots[0] - extent_slot > 5)
4309                                 break;
4310                         extent_slot--;
4311                 }
4312 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4313                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4314                 if (found_extent && item_size < sizeof(*ei))
4315                         found_extent = 0;
4316 #endif
4317                 if (!found_extent) {
4318                         BUG_ON(iref);
4319                         ret = remove_extent_backref(trans, extent_root, path,
4320                                                     NULL, refs_to_drop,
4321                                                     is_data);
4322                         BUG_ON(ret);
4323                         btrfs_release_path(extent_root, path);
4324                         path->leave_spinning = 1;
4325
4326                         key.objectid = bytenr;
4327                         key.type = BTRFS_EXTENT_ITEM_KEY;
4328                         key.offset = num_bytes;
4329
4330                         ret = btrfs_search_slot(trans, extent_root,
4331                                                 &key, path, -1, 1);
4332                         if (ret) {
4333                                 printk(KERN_ERR "umm, got %d back from search"
4334                                        ", was looking for %llu\n", ret,
4335                                        (unsigned long long)bytenr);
4336                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4337                         }
4338                         BUG_ON(ret);
4339                         extent_slot = path->slots[0];
4340                 }
4341         } else {
4342                 btrfs_print_leaf(extent_root, path->nodes[0]);
4343                 WARN_ON(1);
4344                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4345                        "parent %llu root %llu  owner %llu offset %llu\n",
4346                        (unsigned long long)bytenr,
4347                        (unsigned long long)parent,
4348                        (unsigned long long)root_objectid,
4349                        (unsigned long long)owner_objectid,
4350                        (unsigned long long)owner_offset);
4351         }
4352
4353         leaf = path->nodes[0];
4354         item_size = btrfs_item_size_nr(leaf, extent_slot);
4355 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4356         if (item_size < sizeof(*ei)) {
4357                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4358                 ret = convert_extent_item_v0(trans, extent_root, path,
4359                                              owner_objectid, 0);
4360                 BUG_ON(ret < 0);
4361
4362                 btrfs_release_path(extent_root, path);
4363                 path->leave_spinning = 1;
4364
4365                 key.objectid = bytenr;
4366                 key.type = BTRFS_EXTENT_ITEM_KEY;
4367                 key.offset = num_bytes;
4368
4369                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4370                                         -1, 1);
4371                 if (ret) {
4372                         printk(KERN_ERR "umm, got %d back from search"
4373                                ", was looking for %llu\n", ret,
4374                                (unsigned long long)bytenr);
4375                         btrfs_print_leaf(extent_root, path->nodes[0]);
4376                 }
4377                 BUG_ON(ret);
4378                 extent_slot = path->slots[0];
4379                 leaf = path->nodes[0];
4380                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4381         }
4382 #endif
4383         BUG_ON(item_size < sizeof(*ei));
4384         ei = btrfs_item_ptr(leaf, extent_slot,
4385                             struct btrfs_extent_item);
4386         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4387                 struct btrfs_tree_block_info *bi;
4388                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4389                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4390                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4391         }
4392
4393         refs = btrfs_extent_refs(leaf, ei);
4394         BUG_ON(refs < refs_to_drop);
4395         refs -= refs_to_drop;
4396
4397         if (refs > 0) {
4398                 if (extent_op)
4399                         __run_delayed_extent_op(extent_op, leaf, ei);
4400                 /*
4401                  * In the case of inline back ref, reference count will
4402                  * be updated by remove_extent_backref
4403                  */
4404                 if (iref) {
4405                         BUG_ON(!found_extent);
4406                 } else {
4407                         btrfs_set_extent_refs(leaf, ei, refs);
4408                         btrfs_mark_buffer_dirty(leaf);
4409                 }
4410                 if (found_extent) {
4411                         ret = remove_extent_backref(trans, extent_root, path,
4412                                                     iref, refs_to_drop,
4413                                                     is_data);
4414                         BUG_ON(ret);
4415                 }
4416         } else {
4417                 if (found_extent) {
4418                         BUG_ON(is_data && refs_to_drop !=
4419                                extent_data_ref_count(root, path, iref));
4420                         if (iref) {
4421                                 BUG_ON(path->slots[0] != extent_slot);
4422                         } else {
4423                                 BUG_ON(path->slots[0] != extent_slot + 1);
4424                                 path->slots[0] = extent_slot;
4425                                 num_to_del = 2;
4426                         }
4427                 }
4428
4429                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4430                                       num_to_del);
4431                 BUG_ON(ret);
4432                 btrfs_release_path(extent_root, path);
4433
4434                 if (is_data) {
4435                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4436                         BUG_ON(ret);
4437                 } else {
4438                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4439                              bytenr >> PAGE_CACHE_SHIFT,
4440                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4441                 }
4442
4443                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4444                 BUG_ON(ret);
4445         }
4446         btrfs_free_path(path);
4447         return ret;
4448 }
4449
4450 /*
4451  * when we free an block, it is possible (and likely) that we free the last
4452  * delayed ref for that extent as well.  This searches the delayed ref tree for
4453  * a given extent, and if there are no other delayed refs to be processed, it
4454  * removes it from the tree.
4455  */
4456 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4457                                       struct btrfs_root *root, u64 bytenr)
4458 {
4459         struct btrfs_delayed_ref_head *head;
4460         struct btrfs_delayed_ref_root *delayed_refs;
4461         struct btrfs_delayed_ref_node *ref;
4462         struct rb_node *node;
4463         int ret = 0;
4464
4465         delayed_refs = &trans->transaction->delayed_refs;
4466         spin_lock(&delayed_refs->lock);
4467         head = btrfs_find_delayed_ref_head(trans, bytenr);
4468         if (!head)
4469                 goto out;
4470
4471         node = rb_prev(&head->node.rb_node);
4472         if (!node)
4473                 goto out;
4474
4475         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4476
4477         /* there are still entries for this ref, we can't drop it */
4478         if (ref->bytenr == bytenr)
4479                 goto out;
4480
4481         if (head->extent_op) {
4482                 if (!head->must_insert_reserved)
4483                         goto out;
4484                 kfree(head->extent_op);
4485                 head->extent_op = NULL;
4486         }
4487
4488         /*
4489          * waiting for the lock here would deadlock.  If someone else has it
4490          * locked they are already in the process of dropping it anyway
4491          */
4492         if (!mutex_trylock(&head->mutex))
4493                 goto out;
4494
4495         /*
4496          * at this point we have a head with no other entries.  Go
4497          * ahead and process it.
4498          */
4499         head->node.in_tree = 0;
4500         rb_erase(&head->node.rb_node, &delayed_refs->root);
4501
4502         delayed_refs->num_entries--;
4503
4504         /*
4505          * we don't take a ref on the node because we're removing it from the
4506          * tree, so we just steal the ref the tree was holding.
4507          */
4508         delayed_refs->num_heads--;
4509         if (list_empty(&head->cluster))
4510                 delayed_refs->num_heads_ready--;
4511
4512         list_del_init(&head->cluster);
4513         spin_unlock(&delayed_refs->lock);
4514
4515         BUG_ON(head->extent_op);
4516         if (head->must_insert_reserved)
4517                 ret = 1;
4518
4519         mutex_unlock(&head->mutex);
4520         btrfs_put_delayed_ref(&head->node);
4521         return ret;
4522 out:
4523         spin_unlock(&delayed_refs->lock);
4524         return 0;
4525 }
4526
4527 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4528                            struct btrfs_root *root,
4529                            struct extent_buffer *buf,
4530                            u64 parent, int last_ref)
4531 {
4532         struct btrfs_block_rsv *block_rsv;
4533         struct btrfs_block_group_cache *cache = NULL;
4534         int ret;
4535
4536         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4537                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4538                                                 parent, root->root_key.objectid,
4539                                                 btrfs_header_level(buf),
4540                                                 BTRFS_DROP_DELAYED_REF, NULL);
4541                 BUG_ON(ret);
4542         }
4543
4544         if (!last_ref)
4545                 return;
4546
4547         block_rsv = get_block_rsv(trans, root);
4548         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4549         if (block_rsv->space_info != cache->space_info)
4550                 goto out;
4551
4552         if (btrfs_header_generation(buf) == trans->transid) {
4553                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4554                         ret = check_ref_cleanup(trans, root, buf->start);
4555                         if (!ret)
4556                                 goto pin;
4557                 }
4558
4559                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4560                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4561                         goto pin;
4562                 }
4563
4564                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4565
4566                 btrfs_add_free_space(cache, buf->start, buf->len);
4567                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4568                 if (ret == -EAGAIN) {
4569                         /* block group became read-only */
4570                         update_reserved_bytes(cache, buf->len, 0, 1);
4571                         goto out;
4572                 }
4573
4574                 ret = 1;
4575                 spin_lock(&block_rsv->lock);
4576                 if (block_rsv->reserved < block_rsv->size) {
4577                         block_rsv->reserved += buf->len;
4578                         ret = 0;
4579                 }
4580                 spin_unlock(&block_rsv->lock);
4581
4582                 if (ret) {
4583                         spin_lock(&cache->space_info->lock);
4584                         cache->space_info->bytes_reserved -= buf->len;
4585                         spin_unlock(&cache->space_info->lock);
4586                 }
4587                 goto out;
4588         }
4589 pin:
4590         if (block_rsv->durable && !cache->ro) {
4591                 ret = 0;
4592                 spin_lock(&cache->lock);
4593                 if (!cache->ro) {
4594                         cache->reserved_pinned += buf->len;
4595                         ret = 1;
4596                 }
4597                 spin_unlock(&cache->lock);
4598
4599                 if (ret) {
4600                         spin_lock(&block_rsv->lock);
4601                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4602                         spin_unlock(&block_rsv->lock);
4603                 }
4604         }
4605 out:
4606         btrfs_put_block_group(cache);
4607 }
4608
4609 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4610                       struct btrfs_root *root,
4611                       u64 bytenr, u64 num_bytes, u64 parent,
4612                       u64 root_objectid, u64 owner, u64 offset)
4613 {
4614         int ret;
4615
4616         /*
4617          * tree log blocks never actually go into the extent allocation
4618          * tree, just update pinning info and exit early.
4619          */
4620         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4621                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4622                 /* unlocks the pinned mutex */
4623                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4624                 ret = 0;
4625         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4626                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4627                                         parent, root_objectid, (int)owner,
4628                                         BTRFS_DROP_DELAYED_REF, NULL);
4629                 BUG_ON(ret);
4630         } else {
4631                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4632                                         parent, root_objectid, owner,
4633                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4634                 BUG_ON(ret);
4635         }
4636         return ret;
4637 }
4638
4639 static u64 stripe_align(struct btrfs_root *root, u64 val)
4640 {
4641         u64 mask = ((u64)root->stripesize - 1);
4642         u64 ret = (val + mask) & ~mask;
4643         return ret;
4644 }
4645
4646 /*
4647  * when we wait for progress in the block group caching, its because
4648  * our allocation attempt failed at least once.  So, we must sleep
4649  * and let some progress happen before we try again.
4650  *
4651  * This function will sleep at least once waiting for new free space to
4652  * show up, and then it will check the block group free space numbers
4653  * for our min num_bytes.  Another option is to have it go ahead
4654  * and look in the rbtree for a free extent of a given size, but this
4655  * is a good start.
4656  */
4657 static noinline int
4658 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4659                                 u64 num_bytes)
4660 {
4661         struct btrfs_caching_control *caching_ctl;
4662         DEFINE_WAIT(wait);
4663
4664         caching_ctl = get_caching_control(cache);
4665         if (!caching_ctl)
4666                 return 0;
4667
4668         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4669                    (cache->free_space >= num_bytes));
4670
4671         put_caching_control(caching_ctl);
4672         return 0;
4673 }
4674
4675 static noinline int
4676 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4677 {
4678         struct btrfs_caching_control *caching_ctl;
4679         DEFINE_WAIT(wait);
4680
4681         caching_ctl = get_caching_control(cache);
4682         if (!caching_ctl)
4683                 return 0;
4684
4685         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4686
4687         put_caching_control(caching_ctl);
4688         return 0;
4689 }
4690
4691 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4692 {
4693         int index;
4694         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4695                 index = 0;
4696         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4697                 index = 1;
4698         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4699                 index = 2;
4700         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4701                 index = 3;
4702         else
4703                 index = 4;
4704         return index;
4705 }
4706
4707 enum btrfs_loop_type {
4708         LOOP_FIND_IDEAL = 0,
4709         LOOP_CACHING_NOWAIT = 1,
4710         LOOP_CACHING_WAIT = 2,
4711         LOOP_ALLOC_CHUNK = 3,
4712         LOOP_NO_EMPTY_SIZE = 4,
4713 };
4714
4715 /*
4716  * walks the btree of allocated extents and find a hole of a given size.
4717  * The key ins is changed to record the hole:
4718  * ins->objectid == block start
4719  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4720  * ins->offset == number of blocks
4721  * Any available blocks before search_start are skipped.
4722  */
4723 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4724                                      struct btrfs_root *orig_root,
4725                                      u64 num_bytes, u64 empty_size,
4726                                      u64 search_start, u64 search_end,
4727                                      u64 hint_byte, struct btrfs_key *ins,
4728                                      int data)
4729 {
4730         int ret = 0;
4731         struct btrfs_root *root = orig_root->fs_info->extent_root;
4732         struct btrfs_free_cluster *last_ptr = NULL;
4733         struct btrfs_block_group_cache *block_group = NULL;
4734         int empty_cluster = 2 * 1024 * 1024;
4735         int allowed_chunk_alloc = 0;
4736         int done_chunk_alloc = 0;
4737         struct btrfs_space_info *space_info;
4738         int last_ptr_loop = 0;
4739         int loop = 0;
4740         int index = 0;
4741         bool found_uncached_bg = false;
4742         bool failed_cluster_refill = false;
4743         bool failed_alloc = false;
4744         u64 ideal_cache_percent = 0;
4745         u64 ideal_cache_offset = 0;
4746
4747         WARN_ON(num_bytes < root->sectorsize);
4748         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4749         ins->objectid = 0;
4750         ins->offset = 0;
4751
4752         space_info = __find_space_info(root->fs_info, data);
4753         if (!space_info) {
4754                 printk(KERN_ERR "No space info for %d\n", data);
4755                 return -ENOSPC;
4756         }
4757
4758         if (orig_root->ref_cows || empty_size)
4759                 allowed_chunk_alloc = 1;
4760
4761         if (data & BTRFS_BLOCK_GROUP_METADATA) {
4762                 last_ptr = &root->fs_info->meta_alloc_cluster;
4763                 if (!btrfs_test_opt(root, SSD))
4764                         empty_cluster = 64 * 1024;
4765         }
4766
4767         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4768                 last_ptr = &root->fs_info->data_alloc_cluster;
4769         }
4770
4771         if (last_ptr) {
4772                 spin_lock(&last_ptr->lock);
4773                 if (last_ptr->block_group)
4774                         hint_byte = last_ptr->window_start;
4775                 spin_unlock(&last_ptr->lock);
4776         }
4777
4778         search_start = max(search_start, first_logical_byte(root, 0));
4779         search_start = max(search_start, hint_byte);
4780
4781         if (!last_ptr)
4782                 empty_cluster = 0;
4783
4784         if (search_start == hint_byte) {
4785 ideal_cache:
4786                 block_group = btrfs_lookup_block_group(root->fs_info,
4787                                                        search_start);
4788                 /*
4789                  * we don't want to use the block group if it doesn't match our
4790                  * allocation bits, or if its not cached.
4791                  *
4792                  * However if we are re-searching with an ideal block group
4793                  * picked out then we don't care that the block group is cached.
4794                  */
4795                 if (block_group && block_group_bits(block_group, data) &&
4796                     (block_group->cached != BTRFS_CACHE_NO ||
4797                      search_start == ideal_cache_offset)) {
4798                         down_read(&space_info->groups_sem);
4799                         if (list_empty(&block_group->list) ||
4800                             block_group->ro) {
4801                                 /*
4802                                  * someone is removing this block group,
4803                                  * we can't jump into the have_block_group
4804                                  * target because our list pointers are not
4805                                  * valid
4806                                  */
4807                                 btrfs_put_block_group(block_group);
4808                                 up_read(&space_info->groups_sem);
4809                         } else {
4810                                 index = get_block_group_index(block_group);
4811                                 goto have_block_group;
4812                         }
4813                 } else if (block_group) {
4814                         btrfs_put_block_group(block_group);
4815                 }
4816         }
4817 search:
4818         down_read(&space_info->groups_sem);
4819         list_for_each_entry(block_group, &space_info->block_groups[index],
4820                             list) {
4821                 u64 offset;
4822                 int cached;
4823
4824                 btrfs_get_block_group(block_group);
4825                 search_start = block_group->key.objectid;
4826
4827 have_block_group:
4828                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4829                         u64 free_percent;
4830
4831                         free_percent = btrfs_block_group_used(&block_group->item);
4832                         free_percent *= 100;
4833                         free_percent = div64_u64(free_percent,
4834                                                  block_group->key.offset);
4835                         free_percent = 100 - free_percent;
4836                         if (free_percent > ideal_cache_percent &&
4837                             likely(!block_group->ro)) {
4838                                 ideal_cache_offset = block_group->key.objectid;
4839                                 ideal_cache_percent = free_percent;
4840                         }
4841
4842                         /*
4843                          * We only want to start kthread caching if we are at
4844                          * the point where we will wait for caching to make
4845                          * progress, or if our ideal search is over and we've
4846                          * found somebody to start caching.
4847                          */
4848                         if (loop > LOOP_CACHING_NOWAIT ||
4849                             (loop > LOOP_FIND_IDEAL &&
4850                              atomic_read(&space_info->caching_threads) < 2)) {
4851                                 ret = cache_block_group(block_group);
4852                                 BUG_ON(ret);
4853                         }
4854                         found_uncached_bg = true;
4855
4856                         /*
4857                          * If loop is set for cached only, try the next block
4858                          * group.
4859                          */
4860                         if (loop == LOOP_FIND_IDEAL)
4861                                 goto loop;
4862                 }
4863
4864                 cached = block_group_cache_done(block_group);
4865                 if (unlikely(!cached))
4866                         found_uncached_bg = true;
4867
4868                 if (unlikely(block_group->ro))
4869                         goto loop;
4870
4871                 /*
4872                  * Ok we want to try and use the cluster allocator, so lets look
4873                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4874                  * have tried the cluster allocator plenty of times at this
4875                  * point and not have found anything, so we are likely way too
4876                  * fragmented for the clustering stuff to find anything, so lets
4877                  * just skip it and let the allocator find whatever block it can
4878                  * find
4879                  */
4880                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4881                         /*
4882                          * the refill lock keeps out other
4883                          * people trying to start a new cluster
4884                          */
4885                         spin_lock(&last_ptr->refill_lock);
4886                         if (last_ptr->block_group &&
4887                             (last_ptr->block_group->ro ||
4888                             !block_group_bits(last_ptr->block_group, data))) {
4889                                 offset = 0;
4890                                 goto refill_cluster;
4891                         }
4892
4893                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4894                                                  num_bytes, search_start);
4895                         if (offset) {
4896                                 /* we have a block, we're done */
4897                                 spin_unlock(&last_ptr->refill_lock);
4898                                 goto checks;
4899                         }
4900
4901                         spin_lock(&last_ptr->lock);
4902                         /*
4903                          * whoops, this cluster doesn't actually point to
4904                          * this block group.  Get a ref on the block
4905                          * group is does point to and try again
4906                          */
4907                         if (!last_ptr_loop && last_ptr->block_group &&
4908                             last_ptr->block_group != block_group) {
4909
4910                                 btrfs_put_block_group(block_group);
4911                                 block_group = last_ptr->block_group;
4912                                 btrfs_get_block_group(block_group);
4913                                 spin_unlock(&last_ptr->lock);
4914                                 spin_unlock(&last_ptr->refill_lock);
4915
4916                                 last_ptr_loop = 1;
4917                                 search_start = block_group->key.objectid;
4918                                 /*
4919                                  * we know this block group is properly
4920                                  * in the list because
4921                                  * btrfs_remove_block_group, drops the
4922                                  * cluster before it removes the block
4923                                  * group from the list
4924                                  */
4925                                 goto have_block_group;
4926                         }
4927                         spin_unlock(&last_ptr->lock);
4928 refill_cluster:
4929                         /*
4930                          * this cluster didn't work out, free it and
4931                          * start over
4932                          */
4933                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4934
4935                         last_ptr_loop = 0;
4936
4937                         /* allocate a cluster in this block group */
4938                         ret = btrfs_find_space_cluster(trans, root,
4939                                                block_group, last_ptr,
4940                                                offset, num_bytes,
4941                                                empty_cluster + empty_size);
4942                         if (ret == 0) {
4943                                 /*
4944                                  * now pull our allocation out of this
4945                                  * cluster
4946                                  */
4947                                 offset = btrfs_alloc_from_cluster(block_group,
4948                                                   last_ptr, num_bytes,
4949                                                   search_start);
4950                                 if (offset) {
4951                                         /* we found one, proceed */
4952                                         spin_unlock(&last_ptr->refill_lock);
4953                                         goto checks;
4954                                 }
4955                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
4956                                    && !failed_cluster_refill) {
4957                                 spin_unlock(&last_ptr->refill_lock);
4958
4959                                 failed_cluster_refill = true;
4960                                 wait_block_group_cache_progress(block_group,
4961                                        num_bytes + empty_cluster + empty_size);
4962                                 goto have_block_group;
4963                         }
4964
4965                         /*
4966                          * at this point we either didn't find a cluster
4967                          * or we weren't able to allocate a block from our
4968                          * cluster.  Free the cluster we've been trying
4969                          * to use, and go to the next block group
4970                          */
4971                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4972                         spin_unlock(&last_ptr->refill_lock);
4973                         goto loop;
4974                 }
4975
4976                 offset = btrfs_find_space_for_alloc(block_group, search_start,
4977                                                     num_bytes, empty_size);
4978                 /*
4979                  * If we didn't find a chunk, and we haven't failed on this
4980                  * block group before, and this block group is in the middle of
4981                  * caching and we are ok with waiting, then go ahead and wait
4982                  * for progress to be made, and set failed_alloc to true.
4983                  *
4984                  * If failed_alloc is true then we've already waited on this
4985                  * block group once and should move on to the next block group.
4986                  */
4987                 if (!offset && !failed_alloc && !cached &&
4988                     loop > LOOP_CACHING_NOWAIT) {
4989                         wait_block_group_cache_progress(block_group,
4990                                                 num_bytes + empty_size);
4991                         failed_alloc = true;
4992                         goto have_block_group;
4993                 } else if (!offset) {
4994                         goto loop;
4995                 }
4996 checks:
4997                 search_start = stripe_align(root, offset);
4998                 /* move on to the next group */
4999                 if (search_start + num_bytes >= search_end) {
5000                         btrfs_add_free_space(block_group, offset, num_bytes);
5001                         goto loop;
5002                 }
5003
5004                 /* move on to the next group */
5005                 if (search_start + num_bytes >
5006                     block_group->key.objectid + block_group->key.offset) {
5007                         btrfs_add_free_space(block_group, offset, num_bytes);
5008                         goto loop;
5009                 }
5010
5011                 ins->objectid = search_start;
5012                 ins->offset = num_bytes;
5013
5014                 if (offset < search_start)
5015                         btrfs_add_free_space(block_group, offset,
5016                                              search_start - offset);
5017                 BUG_ON(offset > search_start);
5018
5019                 ret = update_reserved_bytes(block_group, num_bytes, 1,
5020                                             (data & BTRFS_BLOCK_GROUP_DATA));
5021                 if (ret == -EAGAIN) {
5022                         btrfs_add_free_space(block_group, offset, num_bytes);
5023                         goto loop;
5024                 }
5025
5026                 /* we are all good, lets return */
5027                 ins->objectid = search_start;
5028                 ins->offset = num_bytes;
5029
5030                 if (offset < search_start)
5031                         btrfs_add_free_space(block_group, offset,
5032                                              search_start - offset);
5033                 BUG_ON(offset > search_start);
5034                 break;
5035 loop:
5036                 failed_cluster_refill = false;
5037                 failed_alloc = false;
5038                 BUG_ON(index != get_block_group_index(block_group));
5039                 btrfs_put_block_group(block_group);
5040         }
5041         up_read(&space_info->groups_sem);
5042
5043         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5044                 goto search;
5045
5046         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5047          *                      for them to make caching progress.  Also
5048          *                      determine the best possible bg to cache
5049          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5050          *                      caching kthreads as we move along
5051          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5052          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5053          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5054          *                      again
5055          */
5056         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5057             (found_uncached_bg || empty_size || empty_cluster ||
5058              allowed_chunk_alloc)) {
5059                 index = 0;
5060                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5061                         found_uncached_bg = false;
5062                         loop++;
5063                         if (!ideal_cache_percent &&
5064                             atomic_read(&space_info->caching_threads))
5065                                 goto search;
5066
5067                         /*
5068                          * 1 of the following 2 things have happened so far
5069                          *
5070                          * 1) We found an ideal block group for caching that
5071                          * is mostly full and will cache quickly, so we might
5072                          * as well wait for it.
5073                          *
5074                          * 2) We searched for cached only and we didn't find
5075                          * anything, and we didn't start any caching kthreads
5076                          * either, so chances are we will loop through and
5077                          * start a couple caching kthreads, and then come back
5078                          * around and just wait for them.  This will be slower
5079                          * because we will have 2 caching kthreads reading at
5080                          * the same time when we could have just started one
5081                          * and waited for it to get far enough to give us an
5082                          * allocation, so go ahead and go to the wait caching
5083                          * loop.
5084                          */
5085                         loop = LOOP_CACHING_WAIT;
5086                         search_start = ideal_cache_offset;
5087                         ideal_cache_percent = 0;
5088                         goto ideal_cache;
5089                 } else if (loop == LOOP_FIND_IDEAL) {
5090                         /*
5091                          * Didn't find a uncached bg, wait on anything we find
5092                          * next.
5093                          */
5094                         loop = LOOP_CACHING_WAIT;
5095                         goto search;
5096                 }
5097
5098                 if (loop < LOOP_CACHING_WAIT) {
5099                         loop++;
5100                         goto search;
5101                 }
5102
5103                 if (loop == LOOP_ALLOC_CHUNK) {
5104                         empty_size = 0;
5105                         empty_cluster = 0;
5106                 }
5107
5108                 if (allowed_chunk_alloc) {
5109                         ret = do_chunk_alloc(trans, root, num_bytes +
5110                                              2 * 1024 * 1024, data, 1);
5111                         allowed_chunk_alloc = 0;
5112                         done_chunk_alloc = 1;
5113                 } else if (!done_chunk_alloc) {
5114                         space_info->force_alloc = 1;
5115                 }
5116
5117                 if (loop < LOOP_NO_EMPTY_SIZE) {
5118                         loop++;
5119                         goto search;
5120                 }
5121                 ret = -ENOSPC;
5122         } else if (!ins->objectid) {
5123                 ret = -ENOSPC;
5124         }
5125
5126         /* we found what we needed */
5127         if (ins->objectid) {
5128                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5129                         trans->block_group = block_group->key.objectid;
5130
5131                 btrfs_put_block_group(block_group);
5132                 ret = 0;
5133         }
5134
5135         return ret;
5136 }
5137
5138 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5139                             int dump_block_groups)
5140 {
5141         struct btrfs_block_group_cache *cache;
5142         int index = 0;
5143
5144         spin_lock(&info->lock);
5145         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5146                (unsigned long long)(info->total_bytes - info->bytes_used -
5147                                     info->bytes_pinned - info->bytes_reserved -
5148                                     info->bytes_readonly),
5149                (info->full) ? "" : "not ");
5150         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5151                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5152                (unsigned long long)info->total_bytes,
5153                (unsigned long long)info->bytes_used,
5154                (unsigned long long)info->bytes_pinned,
5155                (unsigned long long)info->bytes_reserved,
5156                (unsigned long long)info->bytes_may_use,
5157                (unsigned long long)info->bytes_readonly);
5158         spin_unlock(&info->lock);
5159
5160         if (!dump_block_groups)
5161                 return;
5162
5163         down_read(&info->groups_sem);
5164 again:
5165         list_for_each_entry(cache, &info->block_groups[index], list) {
5166                 spin_lock(&cache->lock);
5167                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5168                        "%llu pinned %llu reserved\n",
5169                        (unsigned long long)cache->key.objectid,
5170                        (unsigned long long)cache->key.offset,
5171                        (unsigned long long)btrfs_block_group_used(&cache->item),
5172                        (unsigned long long)cache->pinned,
5173                        (unsigned long long)cache->reserved);
5174                 btrfs_dump_free_space(cache, bytes);
5175                 spin_unlock(&cache->lock);
5176         }
5177         if (++index < BTRFS_NR_RAID_TYPES)
5178                 goto again;
5179         up_read(&info->groups_sem);
5180 }
5181
5182 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5183                          struct btrfs_root *root,
5184                          u64 num_bytes, u64 min_alloc_size,
5185                          u64 empty_size, u64 hint_byte,
5186                          u64 search_end, struct btrfs_key *ins,
5187                          u64 data)
5188 {
5189         int ret;
5190         u64 search_start = 0;
5191
5192         data = btrfs_get_alloc_profile(root, data);
5193 again:
5194         /*
5195          * the only place that sets empty_size is btrfs_realloc_node, which
5196          * is not called recursively on allocations
5197          */
5198         if (empty_size || root->ref_cows)
5199                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5200                                      num_bytes + 2 * 1024 * 1024, data, 0);
5201
5202         WARN_ON(num_bytes < root->sectorsize);
5203         ret = find_free_extent(trans, root, num_bytes, empty_size,
5204                                search_start, search_end, hint_byte,
5205                                ins, data);
5206
5207         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5208                 num_bytes = num_bytes >> 1;
5209                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5210                 num_bytes = max(num_bytes, min_alloc_size);
5211                 do_chunk_alloc(trans, root->fs_info->extent_root,
5212                                num_bytes, data, 1);
5213                 goto again;
5214         }
5215         if (ret == -ENOSPC) {
5216                 struct btrfs_space_info *sinfo;
5217
5218                 sinfo = __find_space_info(root->fs_info, data);
5219                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5220                        "wanted %llu\n", (unsigned long long)data,
5221                        (unsigned long long)num_bytes);
5222                 dump_space_info(sinfo, num_bytes, 1);
5223         }
5224
5225         return ret;
5226 }
5227
5228 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5229 {
5230         struct btrfs_block_group_cache *cache;
5231         int ret = 0;
5232
5233         cache = btrfs_lookup_block_group(root->fs_info, start);
5234         if (!cache) {
5235                 printk(KERN_ERR "Unable to find block group for %llu\n",
5236                        (unsigned long long)start);
5237                 return -ENOSPC;
5238         }
5239
5240         ret = btrfs_discard_extent(root, start, len);
5241
5242         btrfs_add_free_space(cache, start, len);
5243         update_reserved_bytes(cache, len, 0, 1);
5244         btrfs_put_block_group(cache);
5245
5246         return ret;
5247 }
5248
5249 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5250                                       struct btrfs_root *root,
5251                                       u64 parent, u64 root_objectid,
5252                                       u64 flags, u64 owner, u64 offset,
5253                                       struct btrfs_key *ins, int ref_mod)
5254 {
5255         int ret;
5256         struct btrfs_fs_info *fs_info = root->fs_info;
5257         struct btrfs_extent_item *extent_item;
5258         struct btrfs_extent_inline_ref *iref;
5259         struct btrfs_path *path;
5260         struct extent_buffer *leaf;
5261         int type;
5262         u32 size;
5263
5264         if (parent > 0)
5265                 type = BTRFS_SHARED_DATA_REF_KEY;
5266         else
5267                 type = BTRFS_EXTENT_DATA_REF_KEY;
5268
5269         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5270
5271         path = btrfs_alloc_path();
5272         BUG_ON(!path);
5273
5274         path->leave_spinning = 1;
5275         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5276                                       ins, size);
5277         BUG_ON(ret);
5278
5279         leaf = path->nodes[0];
5280         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5281                                      struct btrfs_extent_item);
5282         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5283         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5284         btrfs_set_extent_flags(leaf, extent_item,
5285                                flags | BTRFS_EXTENT_FLAG_DATA);
5286
5287         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5288         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5289         if (parent > 0) {
5290                 struct btrfs_shared_data_ref *ref;
5291                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5292                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5293                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5294         } else {
5295                 struct btrfs_extent_data_ref *ref;
5296                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5297                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5298                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5299                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5300                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5301         }
5302
5303         btrfs_mark_buffer_dirty(path->nodes[0]);
5304         btrfs_free_path(path);
5305
5306         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5307         if (ret) {
5308                 printk(KERN_ERR "btrfs update block group failed for %llu "
5309                        "%llu\n", (unsigned long long)ins->objectid,
5310                        (unsigned long long)ins->offset);
5311                 BUG();
5312         }
5313         return ret;
5314 }
5315
5316 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5317                                      struct btrfs_root *root,
5318                                      u64 parent, u64 root_objectid,
5319                                      u64 flags, struct btrfs_disk_key *key,
5320                                      int level, struct btrfs_key *ins)
5321 {
5322         int ret;
5323         struct btrfs_fs_info *fs_info = root->fs_info;
5324         struct btrfs_extent_item *extent_item;
5325         struct btrfs_tree_block_info *block_info;
5326         struct btrfs_extent_inline_ref *iref;
5327         struct btrfs_path *path;
5328         struct extent_buffer *leaf;
5329         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5330
5331         path = btrfs_alloc_path();
5332         BUG_ON(!path);
5333
5334         path->leave_spinning = 1;
5335         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5336                                       ins, size);
5337         BUG_ON(ret);
5338
5339         leaf = path->nodes[0];
5340         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5341                                      struct btrfs_extent_item);
5342         btrfs_set_extent_refs(leaf, extent_item, 1);
5343         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5344         btrfs_set_extent_flags(leaf, extent_item,
5345                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5346         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5347
5348         btrfs_set_tree_block_key(leaf, block_info, key);
5349         btrfs_set_tree_block_level(leaf, block_info, level);
5350
5351         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5352         if (parent > 0) {
5353                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5354                 btrfs_set_extent_inline_ref_type(leaf, iref,
5355                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5356                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5357         } else {
5358                 btrfs_set_extent_inline_ref_type(leaf, iref,
5359                                                  BTRFS_TREE_BLOCK_REF_KEY);
5360                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5361         }
5362
5363         btrfs_mark_buffer_dirty(leaf);
5364         btrfs_free_path(path);
5365
5366         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5367         if (ret) {
5368                 printk(KERN_ERR "btrfs update block group failed for %llu "
5369                        "%llu\n", (unsigned long long)ins->objectid,
5370                        (unsigned long long)ins->offset);
5371                 BUG();
5372         }
5373         return ret;
5374 }
5375
5376 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5377                                      struct btrfs_root *root,
5378                                      u64 root_objectid, u64 owner,
5379                                      u64 offset, struct btrfs_key *ins)
5380 {
5381         int ret;
5382
5383         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5384
5385         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5386                                          0, root_objectid, owner, offset,
5387                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5388         return ret;
5389 }
5390
5391 /*
5392  * this is used by the tree logging recovery code.  It records that
5393  * an extent has been allocated and makes sure to clear the free
5394  * space cache bits as well
5395  */
5396 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5397                                    struct btrfs_root *root,
5398                                    u64 root_objectid, u64 owner, u64 offset,
5399                                    struct btrfs_key *ins)
5400 {
5401         int ret;
5402         struct btrfs_block_group_cache *block_group;
5403         struct btrfs_caching_control *caching_ctl;
5404         u64 start = ins->objectid;
5405         u64 num_bytes = ins->offset;
5406
5407         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5408         cache_block_group(block_group);
5409         caching_ctl = get_caching_control(block_group);
5410
5411         if (!caching_ctl) {
5412                 BUG_ON(!block_group_cache_done(block_group));
5413                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5414                 BUG_ON(ret);
5415         } else {
5416                 mutex_lock(&caching_ctl->mutex);
5417
5418                 if (start >= caching_ctl->progress) {
5419                         ret = add_excluded_extent(root, start, num_bytes);
5420                         BUG_ON(ret);
5421                 } else if (start + num_bytes <= caching_ctl->progress) {
5422                         ret = btrfs_remove_free_space(block_group,
5423                                                       start, num_bytes);
5424                         BUG_ON(ret);
5425                 } else {
5426                         num_bytes = caching_ctl->progress - start;
5427                         ret = btrfs_remove_free_space(block_group,
5428                                                       start, num_bytes);
5429                         BUG_ON(ret);
5430
5431                         start = caching_ctl->progress;
5432                         num_bytes = ins->objectid + ins->offset -
5433                                     caching_ctl->progress;
5434                         ret = add_excluded_extent(root, start, num_bytes);
5435                         BUG_ON(ret);
5436                 }
5437
5438                 mutex_unlock(&caching_ctl->mutex);
5439                 put_caching_control(caching_ctl);
5440         }
5441
5442         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5443         BUG_ON(ret);
5444         btrfs_put_block_group(block_group);
5445         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5446                                          0, owner, offset, ins, 1);
5447         return ret;
5448 }
5449
5450 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5451                                             struct btrfs_root *root,
5452                                             u64 bytenr, u32 blocksize,
5453                                             int level)
5454 {
5455         struct extent_buffer *buf;
5456
5457         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5458         if (!buf)
5459                 return ERR_PTR(-ENOMEM);
5460         btrfs_set_header_generation(buf, trans->transid);
5461         btrfs_set_buffer_lockdep_class(buf, level);
5462         btrfs_tree_lock(buf);
5463         clean_tree_block(trans, root, buf);
5464
5465         btrfs_set_lock_blocking(buf);
5466         btrfs_set_buffer_uptodate(buf);
5467
5468         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5469                 /*
5470                  * we allow two log transactions at a time, use different
5471                  * EXENT bit to differentiate dirty pages.
5472                  */
5473                 if (root->log_transid % 2 == 0)
5474                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5475                                         buf->start + buf->len - 1, GFP_NOFS);
5476                 else
5477                         set_extent_new(&root->dirty_log_pages, buf->start,
5478                                         buf->start + buf->len - 1, GFP_NOFS);
5479         } else {
5480                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5481                          buf->start + buf->len - 1, GFP_NOFS);
5482         }
5483         trans->blocks_used++;
5484         /* this returns a buffer locked for blocking */
5485         return buf;
5486 }
5487
5488 static struct btrfs_block_rsv *
5489 use_block_rsv(struct btrfs_trans_handle *trans,
5490               struct btrfs_root *root, u32 blocksize)
5491 {
5492         struct btrfs_block_rsv *block_rsv;
5493         int ret;
5494
5495         block_rsv = get_block_rsv(trans, root);
5496
5497         if (block_rsv->size == 0) {
5498                 ret = reserve_metadata_bytes(block_rsv, blocksize);
5499                 if (ret)
5500                         return ERR_PTR(ret);
5501                 return block_rsv;
5502         }
5503
5504         ret = block_rsv_use_bytes(block_rsv, blocksize);
5505         if (!ret)
5506                 return block_rsv;
5507
5508         WARN_ON(1);
5509         printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
5510                 block_rsv->size, block_rsv->reserved,
5511                 block_rsv->freed[0], block_rsv->freed[1]);
5512
5513         return ERR_PTR(-ENOSPC);
5514 }
5515
5516 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5517 {
5518         block_rsv_add_bytes(block_rsv, blocksize, 0);
5519         block_rsv_release_bytes(block_rsv, NULL, 0);
5520 }
5521
5522 /*
5523  * finds a free extent and does all the dirty work required for allocation
5524  * returns the key for the extent through ins, and a tree buffer for
5525  * the first block of the extent through buf.
5526  *
5527  * returns the tree buffer or NULL.
5528  */
5529 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5530                                         struct btrfs_root *root, u32 blocksize,
5531                                         u64 parent, u64 root_objectid,
5532                                         struct btrfs_disk_key *key, int level,
5533                                         u64 hint, u64 empty_size)
5534 {
5535         struct btrfs_key ins;
5536         struct btrfs_block_rsv *block_rsv;
5537         struct extent_buffer *buf;
5538         u64 flags = 0;
5539         int ret;
5540
5541
5542         block_rsv = use_block_rsv(trans, root, blocksize);
5543         if (IS_ERR(block_rsv))
5544                 return ERR_CAST(block_rsv);
5545
5546         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5547                                    empty_size, hint, (u64)-1, &ins, 0);
5548         if (ret) {
5549                 unuse_block_rsv(block_rsv, blocksize);
5550                 return ERR_PTR(ret);
5551         }
5552
5553         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5554                                     blocksize, level);
5555         BUG_ON(IS_ERR(buf));
5556
5557         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5558                 if (parent == 0)
5559                         parent = ins.objectid;
5560                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5561         } else
5562                 BUG_ON(parent > 0);
5563
5564         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5565                 struct btrfs_delayed_extent_op *extent_op;
5566                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5567                 BUG_ON(!extent_op);
5568                 if (key)
5569                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5570                 else
5571                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5572                 extent_op->flags_to_set = flags;
5573                 extent_op->update_key = 1;
5574                 extent_op->update_flags = 1;
5575                 extent_op->is_data = 0;
5576
5577                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5578                                         ins.offset, parent, root_objectid,
5579                                         level, BTRFS_ADD_DELAYED_EXTENT,
5580                                         extent_op);
5581                 BUG_ON(ret);
5582         }
5583         return buf;
5584 }
5585
5586 struct walk_control {
5587         u64 refs[BTRFS_MAX_LEVEL];
5588         u64 flags[BTRFS_MAX_LEVEL];
5589         struct btrfs_key update_progress;
5590         int stage;
5591         int level;
5592         int shared_level;
5593         int update_ref;
5594         int keep_locks;
5595         int reada_slot;
5596         int reada_count;
5597 };
5598
5599 #define DROP_REFERENCE  1
5600 #define UPDATE_BACKREF  2
5601
5602 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5603                                      struct btrfs_root *root,
5604                                      struct walk_control *wc,
5605                                      struct btrfs_path *path)
5606 {
5607         u64 bytenr;
5608         u64 generation;
5609         u64 refs;
5610         u64 flags;
5611         u64 last = 0;
5612         u32 nritems;
5613         u32 blocksize;
5614         struct btrfs_key key;
5615         struct extent_buffer *eb;
5616         int ret;
5617         int slot;
5618         int nread = 0;
5619
5620         if (path->slots[wc->level] < wc->reada_slot) {
5621                 wc->reada_count = wc->reada_count * 2 / 3;
5622                 wc->reada_count = max(wc->reada_count, 2);
5623         } else {
5624                 wc->reada_count = wc->reada_count * 3 / 2;
5625                 wc->reada_count = min_t(int, wc->reada_count,
5626                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5627         }
5628
5629         eb = path->nodes[wc->level];
5630         nritems = btrfs_header_nritems(eb);
5631         blocksize = btrfs_level_size(root, wc->level - 1);
5632
5633         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5634                 if (nread >= wc->reada_count)
5635                         break;
5636
5637                 cond_resched();
5638                 bytenr = btrfs_node_blockptr(eb, slot);
5639                 generation = btrfs_node_ptr_generation(eb, slot);
5640
5641                 if (slot == path->slots[wc->level])
5642                         goto reada;
5643
5644                 if (wc->stage == UPDATE_BACKREF &&
5645                     generation <= root->root_key.offset)
5646                         continue;
5647
5648                 /* We don't lock the tree block, it's OK to be racy here */
5649                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5650                                                &refs, &flags);
5651                 BUG_ON(ret);
5652                 BUG_ON(refs == 0);
5653
5654                 if (wc->stage == DROP_REFERENCE) {
5655                         if (refs == 1)
5656                                 goto reada;
5657
5658                         if (wc->level == 1 &&
5659                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5660                                 continue;
5661                         if (!wc->update_ref ||
5662                             generation <= root->root_key.offset)
5663                                 continue;
5664                         btrfs_node_key_to_cpu(eb, &key, slot);
5665                         ret = btrfs_comp_cpu_keys(&key,
5666                                                   &wc->update_progress);
5667                         if (ret < 0)
5668                                 continue;
5669                 } else {
5670                         if (wc->level == 1 &&
5671                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5672                                 continue;
5673                 }
5674 reada:
5675                 ret = readahead_tree_block(root, bytenr, blocksize,
5676                                            generation);
5677                 if (ret)
5678                         break;
5679                 last = bytenr + blocksize;
5680                 nread++;
5681         }
5682         wc->reada_slot = slot;
5683 }
5684
5685 /*
5686  * hepler to process tree block while walking down the tree.
5687  *
5688  * when wc->stage == UPDATE_BACKREF, this function updates
5689  * back refs for pointers in the block.
5690  *
5691  * NOTE: return value 1 means we should stop walking down.
5692  */
5693 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5694                                    struct btrfs_root *root,
5695                                    struct btrfs_path *path,
5696                                    struct walk_control *wc, int lookup_info)
5697 {
5698         int level = wc->level;
5699         struct extent_buffer *eb = path->nodes[level];
5700         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5701         int ret;
5702
5703         if (wc->stage == UPDATE_BACKREF &&
5704             btrfs_header_owner(eb) != root->root_key.objectid)
5705                 return 1;
5706
5707         /*
5708          * when reference count of tree block is 1, it won't increase
5709          * again. once full backref flag is set, we never clear it.
5710          */
5711         if (lookup_info &&
5712             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5713              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5714                 BUG_ON(!path->locks[level]);
5715                 ret = btrfs_lookup_extent_info(trans, root,
5716                                                eb->start, eb->len,
5717                                                &wc->refs[level],
5718                                                &wc->flags[level]);
5719                 BUG_ON(ret);
5720                 BUG_ON(wc->refs[level] == 0);
5721         }
5722
5723         if (wc->stage == DROP_REFERENCE) {
5724                 if (wc->refs[level] > 1)
5725                         return 1;
5726
5727                 if (path->locks[level] && !wc->keep_locks) {
5728                         btrfs_tree_unlock(eb);
5729                         path->locks[level] = 0;
5730                 }
5731                 return 0;
5732         }
5733
5734         /* wc->stage == UPDATE_BACKREF */
5735         if (!(wc->flags[level] & flag)) {
5736                 BUG_ON(!path->locks[level]);
5737                 ret = btrfs_inc_ref(trans, root, eb, 1);
5738                 BUG_ON(ret);
5739                 ret = btrfs_dec_ref(trans, root, eb, 0);
5740                 BUG_ON(ret);
5741                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5742                                                   eb->len, flag, 0);
5743                 BUG_ON(ret);
5744                 wc->flags[level] |= flag;
5745         }
5746
5747         /*
5748          * the block is shared by multiple trees, so it's not good to
5749          * keep the tree lock
5750          */
5751         if (path->locks[level] && level > 0) {
5752                 btrfs_tree_unlock(eb);
5753                 path->locks[level] = 0;
5754         }
5755         return 0;
5756 }
5757
5758 /*
5759  * hepler to process tree block pointer.
5760  *
5761  * when wc->stage == DROP_REFERENCE, this function checks
5762  * reference count of the block pointed to. if the block
5763  * is shared and we need update back refs for the subtree
5764  * rooted at the block, this function changes wc->stage to
5765  * UPDATE_BACKREF. if the block is shared and there is no
5766  * need to update back, this function drops the reference
5767  * to the block.
5768  *
5769  * NOTE: return value 1 means we should stop walking down.
5770  */
5771 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5772                                  struct btrfs_root *root,
5773                                  struct btrfs_path *path,
5774                                  struct walk_control *wc, int *lookup_info)
5775 {
5776         u64 bytenr;
5777         u64 generation;
5778         u64 parent;
5779         u32 blocksize;
5780         struct btrfs_key key;
5781         struct extent_buffer *next;
5782         int level = wc->level;
5783         int reada = 0;
5784         int ret = 0;
5785
5786         generation = btrfs_node_ptr_generation(path->nodes[level],
5787                                                path->slots[level]);
5788         /*
5789          * if the lower level block was created before the snapshot
5790          * was created, we know there is no need to update back refs
5791          * for the subtree
5792          */
5793         if (wc->stage == UPDATE_BACKREF &&
5794             generation <= root->root_key.offset) {
5795                 *lookup_info = 1;
5796                 return 1;
5797         }
5798
5799         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5800         blocksize = btrfs_level_size(root, level - 1);
5801
5802         next = btrfs_find_tree_block(root, bytenr, blocksize);
5803         if (!next) {
5804                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5805                 if (!next)
5806                         return -ENOMEM;
5807                 reada = 1;
5808         }
5809         btrfs_tree_lock(next);
5810         btrfs_set_lock_blocking(next);
5811
5812         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5813                                        &wc->refs[level - 1],
5814                                        &wc->flags[level - 1]);
5815         BUG_ON(ret);
5816         BUG_ON(wc->refs[level - 1] == 0);
5817         *lookup_info = 0;
5818
5819         if (wc->stage == DROP_REFERENCE) {
5820                 if (wc->refs[level - 1] > 1) {
5821                         if (level == 1 &&
5822                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5823                                 goto skip;
5824
5825                         if (!wc->update_ref ||
5826                             generation <= root->root_key.offset)
5827                                 goto skip;
5828
5829                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5830                                               path->slots[level]);
5831                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5832                         if (ret < 0)
5833                                 goto skip;
5834
5835                         wc->stage = UPDATE_BACKREF;
5836                         wc->shared_level = level - 1;
5837                 }
5838         } else {
5839                 if (level == 1 &&
5840                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5841                         goto skip;
5842         }
5843
5844         if (!btrfs_buffer_uptodate(next, generation)) {
5845                 btrfs_tree_unlock(next);
5846                 free_extent_buffer(next);
5847                 next = NULL;
5848                 *lookup_info = 1;
5849         }
5850
5851         if (!next) {
5852                 if (reada && level == 1)
5853                         reada_walk_down(trans, root, wc, path);
5854                 next = read_tree_block(root, bytenr, blocksize, generation);
5855                 btrfs_tree_lock(next);
5856                 btrfs_set_lock_blocking(next);
5857         }
5858
5859         level--;
5860         BUG_ON(level != btrfs_header_level(next));
5861         path->nodes[level] = next;
5862         path->slots[level] = 0;
5863         path->locks[level] = 1;
5864         wc->level = level;
5865         if (wc->level == 1)
5866                 wc->reada_slot = 0;
5867         return 0;
5868 skip:
5869         wc->refs[level - 1] = 0;
5870         wc->flags[level - 1] = 0;
5871         if (wc->stage == DROP_REFERENCE) {
5872                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5873                         parent = path->nodes[level]->start;
5874                 } else {
5875                         BUG_ON(root->root_key.objectid !=
5876                                btrfs_header_owner(path->nodes[level]));
5877                         parent = 0;
5878                 }
5879
5880                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5881                                         root->root_key.objectid, level - 1, 0);
5882                 BUG_ON(ret);
5883         }
5884         btrfs_tree_unlock(next);
5885         free_extent_buffer(next);
5886         *lookup_info = 1;
5887         return 1;
5888 }
5889
5890 /*
5891  * hepler to process tree block while walking up the tree.
5892  *
5893  * when wc->stage == DROP_REFERENCE, this function drops
5894  * reference count on the block.
5895  *
5896  * when wc->stage == UPDATE_BACKREF, this function changes
5897  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5898  * to UPDATE_BACKREF previously while processing the block.
5899  *
5900  * NOTE: return value 1 means we should stop walking up.
5901  */
5902 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5903                                  struct btrfs_root *root,
5904                                  struct btrfs_path *path,
5905                                  struct walk_control *wc)
5906 {
5907         int ret;
5908         int level = wc->level;
5909         struct extent_buffer *eb = path->nodes[level];
5910         u64 parent = 0;
5911
5912         if (wc->stage == UPDATE_BACKREF) {
5913                 BUG_ON(wc->shared_level < level);
5914                 if (level < wc->shared_level)
5915                         goto out;
5916
5917                 ret = find_next_key(path, level + 1, &wc->update_progress);
5918                 if (ret > 0)
5919                         wc->update_ref = 0;
5920
5921                 wc->stage = DROP_REFERENCE;
5922                 wc->shared_level = -1;
5923                 path->slots[level] = 0;
5924
5925                 /*
5926                  * check reference count again if the block isn't locked.
5927                  * we should start walking down the tree again if reference
5928                  * count is one.
5929                  */
5930                 if (!path->locks[level]) {
5931                         BUG_ON(level == 0);
5932                         btrfs_tree_lock(eb);
5933                         btrfs_set_lock_blocking(eb);
5934                         path->locks[level] = 1;
5935
5936                         ret = btrfs_lookup_extent_info(trans, root,
5937                                                        eb->start, eb->len,
5938                                                        &wc->refs[level],
5939                                                        &wc->flags[level]);
5940                         BUG_ON(ret);
5941                         BUG_ON(wc->refs[level] == 0);
5942                         if (wc->refs[level] == 1) {
5943                                 btrfs_tree_unlock(eb);
5944                                 path->locks[level] = 0;
5945                                 return 1;
5946                         }
5947                 }
5948         }
5949
5950         /* wc->stage == DROP_REFERENCE */
5951         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5952
5953         if (wc->refs[level] == 1) {
5954                 if (level == 0) {
5955                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5956                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5957                         else
5958                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5959                         BUG_ON(ret);
5960                 }
5961                 /* make block locked assertion in clean_tree_block happy */
5962                 if (!path->locks[level] &&
5963                     btrfs_header_generation(eb) == trans->transid) {
5964                         btrfs_tree_lock(eb);
5965                         btrfs_set_lock_blocking(eb);
5966                         path->locks[level] = 1;
5967                 }
5968                 clean_tree_block(trans, root, eb);
5969         }
5970
5971         if (eb == root->node) {
5972                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5973                         parent = eb->start;
5974                 else
5975                         BUG_ON(root->root_key.objectid !=
5976                                btrfs_header_owner(eb));
5977         } else {
5978                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5979                         parent = path->nodes[level + 1]->start;
5980                 else
5981                         BUG_ON(root->root_key.objectid !=
5982                                btrfs_header_owner(path->nodes[level + 1]));
5983         }
5984
5985         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
5986 out:
5987         wc->refs[level] = 0;
5988         wc->flags[level] = 0;
5989         return 0;
5990 }
5991
5992 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5993                                    struct btrfs_root *root,
5994                                    struct btrfs_path *path,
5995                                    struct walk_control *wc)
5996 {
5997         int level = wc->level;
5998         int lookup_info = 1;
5999         int ret;
6000
6001         while (level >= 0) {
6002                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6003                 if (ret > 0)
6004                         break;
6005
6006                 if (level == 0)
6007                         break;
6008
6009                 if (path->slots[level] >=
6010                     btrfs_header_nritems(path->nodes[level]))
6011                         break;
6012
6013                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6014                 if (ret > 0) {
6015                         path->slots[level]++;
6016                         continue;
6017                 } else if (ret < 0)
6018                         return ret;
6019                 level = wc->level;
6020         }
6021         return 0;
6022 }
6023
6024 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6025                                  struct btrfs_root *root,
6026                                  struct btrfs_path *path,
6027                                  struct walk_control *wc, int max_level)
6028 {
6029         int level = wc->level;
6030         int ret;
6031
6032         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6033         while (level < max_level && path->nodes[level]) {
6034                 wc->level = level;
6035                 if (path->slots[level] + 1 <
6036                     btrfs_header_nritems(path->nodes[level])) {
6037                         path->slots[level]++;
6038                         return 0;
6039                 } else {
6040                         ret = walk_up_proc(trans, root, path, wc);
6041                         if (ret > 0)
6042                                 return 0;
6043
6044                         if (path->locks[level]) {
6045                                 btrfs_tree_unlock(path->nodes[level]);
6046                                 path->locks[level] = 0;
6047                         }
6048                         free_extent_buffer(path->nodes[level]);
6049                         path->nodes[level] = NULL;
6050                         level++;
6051                 }
6052         }
6053         return 1;
6054 }
6055
6056 /*
6057  * drop a subvolume tree.
6058  *
6059  * this function traverses the tree freeing any blocks that only
6060  * referenced by the tree.
6061  *
6062  * when a shared tree block is found. this function decreases its
6063  * reference count by one. if update_ref is true, this function
6064  * also make sure backrefs for the shared block and all lower level
6065  * blocks are properly updated.
6066  */
6067 int btrfs_drop_snapshot(struct btrfs_root *root,
6068                         struct btrfs_block_rsv *block_rsv, int update_ref)
6069 {
6070         struct btrfs_path *path;
6071         struct btrfs_trans_handle *trans;
6072         struct btrfs_root *tree_root = root->fs_info->tree_root;
6073         struct btrfs_root_item *root_item = &root->root_item;
6074         struct walk_control *wc;
6075         struct btrfs_key key;
6076         int err = 0;
6077         int ret;
6078         int level;
6079
6080         path = btrfs_alloc_path();
6081         BUG_ON(!path);
6082
6083         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6084         BUG_ON(!wc);
6085
6086         trans = btrfs_start_transaction(tree_root, 0);
6087         if (block_rsv)
6088                 trans->block_rsv = block_rsv;
6089
6090         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6091                 level = btrfs_header_level(root->node);
6092                 path->nodes[level] = btrfs_lock_root_node(root);
6093                 btrfs_set_lock_blocking(path->nodes[level]);
6094                 path->slots[level] = 0;
6095                 path->locks[level] = 1;
6096                 memset(&wc->update_progress, 0,
6097                        sizeof(wc->update_progress));
6098         } else {
6099                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6100                 memcpy(&wc->update_progress, &key,
6101                        sizeof(wc->update_progress));
6102
6103                 level = root_item->drop_level;
6104                 BUG_ON(level == 0);
6105                 path->lowest_level = level;
6106                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6107                 path->lowest_level = 0;
6108                 if (ret < 0) {
6109                         err = ret;
6110                         goto out;
6111                 }
6112                 WARN_ON(ret > 0);
6113
6114                 /*
6115                  * unlock our path, this is safe because only this
6116                  * function is allowed to delete this snapshot
6117                  */
6118                 btrfs_unlock_up_safe(path, 0);
6119
6120                 level = btrfs_header_level(root->node);
6121                 while (1) {
6122                         btrfs_tree_lock(path->nodes[level]);
6123                         btrfs_set_lock_blocking(path->nodes[level]);
6124
6125                         ret = btrfs_lookup_extent_info(trans, root,
6126                                                 path->nodes[level]->start,
6127                                                 path->nodes[level]->len,
6128                                                 &wc->refs[level],
6129                                                 &wc->flags[level]);
6130                         BUG_ON(ret);
6131                         BUG_ON(wc->refs[level] == 0);
6132
6133                         if (level == root_item->drop_level)
6134                                 break;
6135
6136                         btrfs_tree_unlock(path->nodes[level]);
6137                         WARN_ON(wc->refs[level] != 1);
6138                         level--;
6139                 }
6140         }
6141
6142         wc->level = level;
6143         wc->shared_level = -1;
6144         wc->stage = DROP_REFERENCE;
6145         wc->update_ref = update_ref;
6146         wc->keep_locks = 0;
6147         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6148
6149         while (1) {
6150                 ret = walk_down_tree(trans, root, path, wc);
6151                 if (ret < 0) {
6152                         err = ret;
6153                         break;
6154                 }
6155
6156                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6157                 if (ret < 0) {
6158                         err = ret;
6159                         break;
6160                 }
6161
6162                 if (ret > 0) {
6163                         BUG_ON(wc->stage != DROP_REFERENCE);
6164                         break;
6165                 }
6166
6167                 if (wc->stage == DROP_REFERENCE) {
6168                         level = wc->level;
6169                         btrfs_node_key(path->nodes[level],
6170                                        &root_item->drop_progress,
6171                                        path->slots[level]);
6172                         root_item->drop_level = level;
6173                 }
6174
6175                 BUG_ON(wc->level == 0);
6176                 if (btrfs_should_end_transaction(trans, tree_root)) {
6177                         ret = btrfs_update_root(trans, tree_root,
6178                                                 &root->root_key,
6179                                                 root_item);
6180                         BUG_ON(ret);
6181
6182                         btrfs_end_transaction_throttle(trans, tree_root);
6183                         trans = btrfs_start_transaction(tree_root, 0);
6184                         if (block_rsv)
6185                                 trans->block_rsv = block_rsv;
6186                 }
6187         }
6188         btrfs_release_path(root, path);
6189         BUG_ON(err);
6190
6191         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6192         BUG_ON(ret);
6193
6194         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6195                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6196                                            NULL, NULL);
6197                 BUG_ON(ret < 0);
6198                 if (ret > 0) {
6199                         ret = btrfs_del_orphan_item(trans, tree_root,
6200                                                     root->root_key.objectid);
6201                         BUG_ON(ret);
6202                 }
6203         }
6204
6205         if (root->in_radix) {
6206                 btrfs_free_fs_root(tree_root->fs_info, root);
6207         } else {
6208                 free_extent_buffer(root->node);
6209                 free_extent_buffer(root->commit_root);
6210                 kfree(root);
6211         }
6212 out:
6213         btrfs_end_transaction_throttle(trans, tree_root);
6214         kfree(wc);
6215         btrfs_free_path(path);
6216         return err;
6217 }
6218
6219 /*
6220  * drop subtree rooted at tree block 'node'.
6221  *
6222  * NOTE: this function will unlock and release tree block 'node'
6223  */
6224 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6225                         struct btrfs_root *root,
6226                         struct extent_buffer *node,
6227                         struct extent_buffer *parent)
6228 {
6229         struct btrfs_path *path;
6230         struct walk_control *wc;
6231         int level;
6232         int parent_level;
6233         int ret = 0;
6234         int wret;
6235
6236         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6237
6238         path = btrfs_alloc_path();
6239         BUG_ON(!path);
6240
6241         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6242         BUG_ON(!wc);
6243
6244         btrfs_assert_tree_locked(parent);
6245         parent_level = btrfs_header_level(parent);
6246         extent_buffer_get(parent);
6247         path->nodes[parent_level] = parent;
6248         path->slots[parent_level] = btrfs_header_nritems(parent);
6249
6250         btrfs_assert_tree_locked(node);
6251         level = btrfs_header_level(node);
6252         path->nodes[level] = node;
6253         path->slots[level] = 0;
6254         path->locks[level] = 1;
6255
6256         wc->refs[parent_level] = 1;
6257         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6258         wc->level = level;
6259         wc->shared_level = -1;
6260         wc->stage = DROP_REFERENCE;
6261         wc->update_ref = 0;
6262         wc->keep_locks = 1;
6263         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6264
6265         while (1) {
6266                 wret = walk_down_tree(trans, root, path, wc);
6267                 if (wret < 0) {
6268                         ret = wret;
6269                         break;
6270                 }
6271
6272                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6273                 if (wret < 0)
6274                         ret = wret;
6275                 if (wret != 0)
6276                         break;
6277         }
6278
6279         kfree(wc);
6280         btrfs_free_path(path);
6281         return ret;
6282 }
6283
6284 #if 0
6285 static unsigned long calc_ra(unsigned long start, unsigned long last,
6286                              unsigned long nr)
6287 {
6288         return min(last, start + nr - 1);
6289 }
6290
6291 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6292                                          u64 len)
6293 {
6294         u64 page_start;
6295         u64 page_end;
6296         unsigned long first_index;
6297         unsigned long last_index;
6298         unsigned long i;
6299         struct page *page;
6300         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6301         struct file_ra_state *ra;
6302         struct btrfs_ordered_extent *ordered;
6303         unsigned int total_read = 0;
6304         unsigned int total_dirty = 0;
6305         int ret = 0;
6306
6307         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6308
6309         mutex_lock(&inode->i_mutex);
6310         first_index = start >> PAGE_CACHE_SHIFT;
6311         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6312
6313         /* make sure the dirty trick played by the caller work */
6314         ret = invalidate_inode_pages2_range(inode->i_mapping,
6315                                             first_index, last_index);
6316         if (ret)
6317                 goto out_unlock;
6318
6319         file_ra_state_init(ra, inode->i_mapping);
6320
6321         for (i = first_index ; i <= last_index; i++) {
6322                 if (total_read % ra->ra_pages == 0) {
6323                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6324                                        calc_ra(i, last_index, ra->ra_pages));
6325                 }
6326                 total_read++;
6327 again:
6328                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6329                         BUG_ON(1);
6330                 page = grab_cache_page(inode->i_mapping, i);
6331                 if (!page) {
6332                         ret = -ENOMEM;
6333                         goto out_unlock;
6334                 }
6335                 if (!PageUptodate(page)) {
6336                         btrfs_readpage(NULL, page);
6337                         lock_page(page);
6338                         if (!PageUptodate(page)) {
6339                                 unlock_page(page);
6340                                 page_cache_release(page);
6341                                 ret = -EIO;
6342                                 goto out_unlock;
6343                         }
6344                 }
6345                 wait_on_page_writeback(page);
6346
6347                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6348                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6349                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6350
6351                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6352                 if (ordered) {
6353                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6354                         unlock_page(page);
6355                         page_cache_release(page);
6356                         btrfs_start_ordered_extent(inode, ordered, 1);
6357                         btrfs_put_ordered_extent(ordered);
6358                         goto again;
6359                 }
6360                 set_page_extent_mapped(page);
6361
6362                 if (i == first_index)
6363                         set_extent_bits(io_tree, page_start, page_end,
6364                                         EXTENT_BOUNDARY, GFP_NOFS);
6365                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6366
6367                 set_page_dirty(page);
6368                 total_dirty++;
6369
6370                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6371                 unlock_page(page);
6372                 page_cache_release(page);
6373         }
6374
6375 out_unlock:
6376         kfree(ra);
6377         mutex_unlock(&inode->i_mutex);
6378         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6379         return ret;
6380 }
6381
6382 static noinline int relocate_data_extent(struct inode *reloc_inode,
6383                                          struct btrfs_key *extent_key,
6384                                          u64 offset)
6385 {
6386         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6387         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6388         struct extent_map *em;
6389         u64 start = extent_key->objectid - offset;
6390         u64 end = start + extent_key->offset - 1;
6391
6392         em = alloc_extent_map(GFP_NOFS);
6393         BUG_ON(!em || IS_ERR(em));
6394
6395         em->start = start;
6396         em->len = extent_key->offset;
6397         em->block_len = extent_key->offset;
6398         em->block_start = extent_key->objectid;
6399         em->bdev = root->fs_info->fs_devices->latest_bdev;
6400         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6401
6402         /* setup extent map to cheat btrfs_readpage */
6403         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6404         while (1) {
6405                 int ret;
6406                 write_lock(&em_tree->lock);
6407                 ret = add_extent_mapping(em_tree, em);
6408                 write_unlock(&em_tree->lock);
6409                 if (ret != -EEXIST) {
6410                         free_extent_map(em);
6411                         break;
6412                 }
6413                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6414         }
6415         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6416
6417         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6418 }
6419
6420 struct btrfs_ref_path {
6421         u64 extent_start;
6422         u64 nodes[BTRFS_MAX_LEVEL];
6423         u64 root_objectid;
6424         u64 root_generation;
6425         u64 owner_objectid;
6426         u32 num_refs;
6427         int lowest_level;
6428         int current_level;
6429         int shared_level;
6430
6431         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6432         u64 new_nodes[BTRFS_MAX_LEVEL];
6433 };
6434
6435 struct disk_extent {
6436         u64 ram_bytes;
6437         u64 disk_bytenr;
6438         u64 disk_num_bytes;
6439         u64 offset;
6440         u64 num_bytes;
6441         u8 compression;
6442         u8 encryption;
6443         u16 other_encoding;
6444 };
6445
6446 static int is_cowonly_root(u64 root_objectid)
6447 {
6448         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6449             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6450             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6451             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6452             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6453             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6454                 return 1;
6455         return 0;
6456 }
6457
6458 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6459                                     struct btrfs_root *extent_root,
6460                                     struct btrfs_ref_path *ref_path,
6461                                     int first_time)
6462 {
6463         struct extent_buffer *leaf;
6464         struct btrfs_path *path;
6465         struct btrfs_extent_ref *ref;
6466         struct btrfs_key key;
6467         struct btrfs_key found_key;
6468         u64 bytenr;
6469         u32 nritems;
6470         int level;
6471         int ret = 1;
6472
6473         path = btrfs_alloc_path();
6474         if (!path)
6475                 return -ENOMEM;
6476
6477         if (first_time) {
6478                 ref_path->lowest_level = -1;
6479                 ref_path->current_level = -1;
6480                 ref_path->shared_level = -1;
6481                 goto walk_up;
6482         }
6483 walk_down:
6484         level = ref_path->current_level - 1;
6485         while (level >= -1) {
6486                 u64 parent;
6487                 if (level < ref_path->lowest_level)
6488                         break;
6489
6490                 if (level >= 0)
6491                         bytenr = ref_path->nodes[level];
6492                 else
6493                         bytenr = ref_path->extent_start;
6494                 BUG_ON(bytenr == 0);
6495
6496                 parent = ref_path->nodes[level + 1];
6497                 ref_path->nodes[level + 1] = 0;
6498                 ref_path->current_level = level;
6499                 BUG_ON(parent == 0);
6500
6501                 key.objectid = bytenr;
6502                 key.offset = parent + 1;
6503                 key.type = BTRFS_EXTENT_REF_KEY;
6504
6505                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6506                 if (ret < 0)
6507                         goto out;
6508                 BUG_ON(ret == 0);
6509
6510                 leaf = path->nodes[0];
6511                 nritems = btrfs_header_nritems(leaf);
6512                 if (path->slots[0] >= nritems) {
6513                         ret = btrfs_next_leaf(extent_root, path);
6514                         if (ret < 0)
6515                                 goto out;
6516                         if (ret > 0)
6517                                 goto next;
6518                         leaf = path->nodes[0];
6519                 }
6520
6521                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6522                 if (found_key.objectid == bytenr &&
6523                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6524                         if (level < ref_path->shared_level)
6525                                 ref_path->shared_level = level;
6526                         goto found;
6527                 }
6528 next:
6529                 level--;
6530                 btrfs_release_path(extent_root, path);
6531                 cond_resched();
6532         }
6533         /* reached lowest level */
6534         ret = 1;
6535         goto out;
6536 walk_up:
6537         level = ref_path->current_level;
6538         while (level < BTRFS_MAX_LEVEL - 1) {
6539                 u64 ref_objectid;
6540
6541                 if (level >= 0)
6542                         bytenr = ref_path->nodes[level];
6543                 else
6544                         bytenr = ref_path->extent_start;
6545
6546                 BUG_ON(bytenr == 0);
6547
6548                 key.objectid = bytenr;
6549                 key.offset = 0;
6550                 key.type = BTRFS_EXTENT_REF_KEY;
6551
6552                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6553                 if (ret < 0)
6554                         goto out;
6555
6556                 leaf = path->nodes[0];
6557                 nritems = btrfs_header_nritems(leaf);
6558                 if (path->slots[0] >= nritems) {
6559                         ret = btrfs_next_leaf(extent_root, path);
6560                         if (ret < 0)
6561                                 goto out;
6562                         if (ret > 0) {
6563                                 /* the extent was freed by someone */
6564                                 if (ref_path->lowest_level == level)
6565                                         goto out;
6566                                 btrfs_release_path(extent_root, path);
6567                                 goto walk_down;
6568                         }
6569                         leaf = path->nodes[0];
6570                 }
6571
6572                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6573                 if (found_key.objectid != bytenr ||
6574                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6575                         /* the extent was freed by someone */
6576                         if (ref_path->lowest_level == level) {
6577                                 ret = 1;
6578                                 goto out;
6579                         }
6580                         btrfs_release_path(extent_root, path);
6581                         goto walk_down;
6582                 }
6583 found:
6584                 ref = btrfs_item_ptr(leaf, path->slots[0],
6585                                 struct btrfs_extent_ref);
6586                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6587                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6588                         if (first_time) {
6589                                 level = (int)ref_objectid;
6590                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6591                                 ref_path->lowest_level = level;
6592                                 ref_path->current_level = level;
6593                                 ref_path->nodes[level] = bytenr;
6594                         } else {
6595                                 WARN_ON(ref_objectid != level);
6596                         }
6597                 } else {
6598                         WARN_ON(level != -1);
6599                 }
6600                 first_time = 0;
6601
6602                 if (ref_path->lowest_level == level) {
6603                         ref_path->owner_objectid = ref_objectid;
6604                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6605                 }
6606
6607                 /*
6608                  * the block is tree root or the block isn't in reference
6609                  * counted tree.
6610                  */
6611                 if (found_key.objectid == found_key.offset ||
6612                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6613                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6614                         ref_path->root_generation =
6615                                 btrfs_ref_generation(leaf, ref);
6616                         if (level < 0) {
6617                                 /* special reference from the tree log */
6618                                 ref_path->nodes[0] = found_key.offset;
6619                                 ref_path->current_level = 0;
6620                         }
6621                         ret = 0;
6622                         goto out;
6623                 }
6624
6625                 level++;
6626                 BUG_ON(ref_path->nodes[level] != 0);
6627                 ref_path->nodes[level] = found_key.offset;
6628                 ref_path->current_level = level;
6629
6630                 /*
6631                  * the reference was created in the running transaction,
6632                  * no need to continue walking up.
6633                  */
6634                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6635                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6636                         ref_path->root_generation =
6637                                 btrfs_ref_generation(leaf, ref);
6638                         ret = 0;
6639                         goto out;
6640                 }
6641
6642                 btrfs_release_path(extent_root, path);
6643                 cond_resched();
6644         }
6645         /* reached max tree level, but no tree root found. */
6646         BUG();
6647 out:
6648         btrfs_free_path(path);
6649         return ret;
6650 }
6651
6652 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6653                                 struct btrfs_root *extent_root,
6654                                 struct btrfs_ref_path *ref_path,
6655                                 u64 extent_start)
6656 {
6657         memset(ref_path, 0, sizeof(*ref_path));
6658         ref_path->extent_start = extent_start;
6659
6660         return __next_ref_path(trans, extent_root, ref_path, 1);
6661 }
6662
6663 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6664                                struct btrfs_root *extent_root,
6665                                struct btrfs_ref_path *ref_path)
6666 {
6667         return __next_ref_path(trans, extent_root, ref_path, 0);
6668 }
6669
6670 static noinline int get_new_locations(struct inode *reloc_inode,
6671                                       struct btrfs_key *extent_key,
6672                                       u64 offset, int no_fragment,
6673                                       struct disk_extent **extents,
6674                                       int *nr_extents)
6675 {
6676         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6677         struct btrfs_path *path;
6678         struct btrfs_file_extent_item *fi;
6679         struct extent_buffer *leaf;
6680         struct disk_extent *exts = *extents;
6681         struct btrfs_key found_key;
6682         u64 cur_pos;
6683         u64 last_byte;
6684         u32 nritems;
6685         int nr = 0;
6686         int max = *nr_extents;
6687         int ret;
6688
6689         WARN_ON(!no_fragment && *extents);
6690         if (!exts) {
6691                 max = 1;
6692                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6693                 if (!exts)
6694                         return -ENOMEM;
6695         }
6696
6697         path = btrfs_alloc_path();
6698         BUG_ON(!path);
6699
6700         cur_pos = extent_key->objectid - offset;
6701         last_byte = extent_key->objectid + extent_key->offset;
6702         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6703                                        cur_pos, 0);
6704         if (ret < 0)
6705                 goto out;
6706         if (ret > 0) {
6707                 ret = -ENOENT;
6708                 goto out;
6709         }
6710
6711         while (1) {
6712                 leaf = path->nodes[0];
6713                 nritems = btrfs_header_nritems(leaf);
6714                 if (path->slots[0] >= nritems) {
6715                         ret = btrfs_next_leaf(root, path);
6716                         if (ret < 0)
6717                                 goto out;
6718                         if (ret > 0)
6719                                 break;
6720                         leaf = path->nodes[0];
6721                 }
6722
6723                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6724                 if (found_key.offset != cur_pos ||
6725                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6726                     found_key.objectid != reloc_inode->i_ino)
6727                         break;
6728
6729                 fi = btrfs_item_ptr(leaf, path->slots[0],
6730                                     struct btrfs_file_extent_item);
6731                 if (btrfs_file_extent_type(leaf, fi) !=
6732                     BTRFS_FILE_EXTENT_REG ||
6733                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6734                         break;
6735
6736                 if (nr == max) {
6737                         struct disk_extent *old = exts;
6738                         max *= 2;
6739                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6740                         memcpy(exts, old, sizeof(*exts) * nr);
6741                         if (old != *extents)
6742                                 kfree(old);
6743                 }
6744
6745                 exts[nr].disk_bytenr =
6746                         btrfs_file_extent_disk_bytenr(leaf, fi);
6747                 exts[nr].disk_num_bytes =
6748                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6749                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6750                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6751                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6752                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6753                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6754                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6755                                                                            fi);
6756                 BUG_ON(exts[nr].offset > 0);
6757                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6758                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6759
6760                 cur_pos += exts[nr].num_bytes;
6761                 nr++;
6762
6763                 if (cur_pos + offset >= last_byte)
6764                         break;
6765
6766                 if (no_fragment) {
6767                         ret = 1;
6768                         goto out;
6769                 }
6770                 path->slots[0]++;
6771         }
6772
6773         BUG_ON(cur_pos + offset > last_byte);
6774         if (cur_pos + offset < last_byte) {
6775                 ret = -ENOENT;
6776                 goto out;
6777         }
6778         ret = 0;
6779 out:
6780         btrfs_free_path(path);
6781         if (ret) {
6782                 if (exts != *extents)
6783                         kfree(exts);
6784         } else {
6785                 *extents = exts;
6786                 *nr_extents = nr;
6787         }
6788         return ret;
6789 }
6790
6791 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6792                                         struct btrfs_root *root,
6793                                         struct btrfs_path *path,
6794                                         struct btrfs_key *extent_key,
6795                                         struct btrfs_key *leaf_key,
6796                                         struct btrfs_ref_path *ref_path,
6797                                         struct disk_extent *new_extents,
6798                                         int nr_extents)
6799 {
6800         struct extent_buffer *leaf;
6801         struct btrfs_file_extent_item *fi;
6802         struct inode *inode = NULL;
6803         struct btrfs_key key;
6804         u64 lock_start = 0;
6805         u64 lock_end = 0;
6806         u64 num_bytes;
6807         u64 ext_offset;
6808         u64 search_end = (u64)-1;
6809         u32 nritems;
6810         int nr_scaned = 0;
6811         int extent_locked = 0;
6812         int extent_type;
6813         int ret;
6814
6815         memcpy(&key, leaf_key, sizeof(key));
6816         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6817                 if (key.objectid < ref_path->owner_objectid ||
6818                     (key.objectid == ref_path->owner_objectid &&
6819                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6820                         key.objectid = ref_path->owner_objectid;
6821                         key.type = BTRFS_EXTENT_DATA_KEY;
6822                         key.offset = 0;
6823                 }
6824         }
6825
6826         while (1) {
6827                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6828                 if (ret < 0)
6829                         goto out;
6830
6831                 leaf = path->nodes[0];
6832                 nritems = btrfs_header_nritems(leaf);
6833 next:
6834                 if (extent_locked && ret > 0) {
6835                         /*
6836                          * the file extent item was modified by someone
6837                          * before the extent got locked.
6838                          */
6839                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6840                                       lock_end, GFP_NOFS);
6841                         extent_locked = 0;
6842                 }
6843
6844                 if (path->slots[0] >= nritems) {
6845                         if (++nr_scaned > 2)
6846                                 break;
6847
6848                         BUG_ON(extent_locked);
6849                         ret = btrfs_next_leaf(root, path);
6850                         if (ret < 0)
6851                                 goto out;
6852                         if (ret > 0)
6853                                 break;
6854                         leaf = path->nodes[0];
6855                         nritems = btrfs_header_nritems(leaf);
6856                 }
6857
6858                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6859
6860                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6861                         if ((key.objectid > ref_path->owner_objectid) ||
6862                             (key.objectid == ref_path->owner_objectid &&
6863                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6864                             key.offset >= search_end)
6865                                 break;
6866                 }
6867
6868                 if (inode && key.objectid != inode->i_ino) {
6869                         BUG_ON(extent_locked);
6870                         btrfs_release_path(root, path);
6871                         mutex_unlock(&inode->i_mutex);
6872                         iput(inode);
6873                         inode = NULL;
6874                         continue;
6875                 }
6876
6877                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6878                         path->slots[0]++;
6879                         ret = 1;
6880                         goto next;
6881                 }
6882                 fi = btrfs_item_ptr(leaf, path->slots[0],
6883                                     struct btrfs_file_extent_item);
6884                 extent_type = btrfs_file_extent_type(leaf, fi);
6885                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6886                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6887                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6888                      extent_key->objectid)) {
6889                         path->slots[0]++;
6890                         ret = 1;
6891                         goto next;
6892                 }
6893
6894                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6895                 ext_offset = btrfs_file_extent_offset(leaf, fi);
6896
6897                 if (search_end == (u64)-1) {
6898                         search_end = key.offset - ext_offset +
6899                                 btrfs_file_extent_ram_bytes(leaf, fi);
6900                 }
6901
6902                 if (!extent_locked) {
6903                         lock_start = key.offset;
6904                         lock_end = lock_start + num_bytes - 1;
6905                 } else {
6906                         if (lock_start > key.offset ||
6907                             lock_end + 1 < key.offset + num_bytes) {
6908                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6909                                               lock_start, lock_end, GFP_NOFS);
6910                                 extent_locked = 0;
6911                         }
6912                 }
6913
6914                 if (!inode) {
6915                         btrfs_release_path(root, path);
6916
6917                         inode = btrfs_iget_locked(root->fs_info->sb,
6918                                                   key.objectid, root);
6919                         if (inode->i_state & I_NEW) {
6920                                 BTRFS_I(inode)->root = root;
6921                                 BTRFS_I(inode)->location.objectid =
6922                                         key.objectid;
6923                                 BTRFS_I(inode)->location.type =
6924                                         BTRFS_INODE_ITEM_KEY;
6925                                 BTRFS_I(inode)->location.offset = 0;
6926                                 btrfs_read_locked_inode(inode);
6927                                 unlock_new_inode(inode);
6928                         }
6929                         /*
6930                          * some code call btrfs_commit_transaction while
6931                          * holding the i_mutex, so we can't use mutex_lock
6932                          * here.
6933                          */
6934                         if (is_bad_inode(inode) ||
6935                             !mutex_trylock(&inode->i_mutex)) {
6936                                 iput(inode);
6937                                 inode = NULL;
6938                                 key.offset = (u64)-1;
6939                                 goto skip;
6940                         }
6941                 }
6942
6943                 if (!extent_locked) {
6944                         struct btrfs_ordered_extent *ordered;
6945
6946                         btrfs_release_path(root, path);
6947
6948                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6949                                     lock_end, GFP_NOFS);
6950                         ordered = btrfs_lookup_first_ordered_extent(inode,
6951                                                                     lock_end);
6952                         if (ordered &&
6953                             ordered->file_offset <= lock_end &&
6954                             ordered->file_offset + ordered->len > lock_start) {
6955                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6956                                               lock_start, lock_end, GFP_NOFS);
6957                                 btrfs_start_ordered_extent(inode, ordered, 1);
6958                                 btrfs_put_ordered_extent(ordered);
6959                                 key.offset += num_bytes;
6960                                 goto skip;
6961                         }
6962                         if (ordered)
6963                                 btrfs_put_ordered_extent(ordered);
6964
6965                         extent_locked = 1;
6966                         continue;
6967                 }
6968
6969                 if (nr_extents == 1) {
6970                         /* update extent pointer in place */
6971                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
6972                                                 new_extents[0].disk_bytenr);
6973                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6974                                                 new_extents[0].disk_num_bytes);
6975                         btrfs_mark_buffer_dirty(leaf);
6976
6977                         btrfs_drop_extent_cache(inode, key.offset,
6978                                                 key.offset + num_bytes - 1, 0);
6979
6980                         ret = btrfs_inc_extent_ref(trans, root,
6981                                                 new_extents[0].disk_bytenr,
6982                                                 new_extents[0].disk_num_bytes,
6983                                                 leaf->start,
6984                                                 root->root_key.objectid,
6985                                                 trans->transid,
6986                                                 key.objectid);
6987                         BUG_ON(ret);
6988
6989                         ret = btrfs_free_extent(trans, root,
6990                                                 extent_key->objectid,
6991                                                 extent_key->offset,
6992                                                 leaf->start,
6993                                                 btrfs_header_owner(leaf),
6994                                                 btrfs_header_generation(leaf),
6995                                                 key.objectid, 0);
6996                         BUG_ON(ret);
6997
6998                         btrfs_release_path(root, path);
6999                         key.offset += num_bytes;
7000                 } else {
7001                         BUG_ON(1);
7002 #if 0
7003                         u64 alloc_hint;
7004                         u64 extent_len;
7005                         int i;
7006                         /*
7007                          * drop old extent pointer at first, then insert the
7008                          * new pointers one bye one
7009                          */
7010                         btrfs_release_path(root, path);
7011                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7012                                                  key.offset + num_bytes,
7013                                                  key.offset, &alloc_hint);
7014                         BUG_ON(ret);
7015
7016                         for (i = 0; i < nr_extents; i++) {
7017                                 if (ext_offset >= new_extents[i].num_bytes) {
7018                                         ext_offset -= new_extents[i].num_bytes;
7019                                         continue;
7020                                 }
7021                                 extent_len = min(new_extents[i].num_bytes -
7022                                                  ext_offset, num_bytes);
7023
7024                                 ret = btrfs_insert_empty_item(trans, root,
7025                                                               path, &key,
7026                                                               sizeof(*fi));
7027                                 BUG_ON(ret);
7028
7029                                 leaf = path->nodes[0];
7030                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7031                                                 struct btrfs_file_extent_item);
7032                                 btrfs_set_file_extent_generation(leaf, fi,
7033                                                         trans->transid);
7034                                 btrfs_set_file_extent_type(leaf, fi,
7035                                                         BTRFS_FILE_EXTENT_REG);
7036                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7037                                                 new_extents[i].disk_bytenr);
7038                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7039                                                 new_extents[i].disk_num_bytes);
7040                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7041                                                 new_extents[i].ram_bytes);
7042
7043                                 btrfs_set_file_extent_compression(leaf, fi,
7044                                                 new_extents[i].compression);
7045                                 btrfs_set_file_extent_encryption(leaf, fi,
7046                                                 new_extents[i].encryption);
7047                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7048                                                 new_extents[i].other_encoding);
7049
7050                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7051                                                         extent_len);
7052                                 ext_offset += new_extents[i].offset;
7053                                 btrfs_set_file_extent_offset(leaf, fi,
7054                                                         ext_offset);
7055                                 btrfs_mark_buffer_dirty(leaf);
7056
7057                                 btrfs_drop_extent_cache(inode, key.offset,
7058                                                 key.offset + extent_len - 1, 0);
7059
7060                                 ret = btrfs_inc_extent_ref(trans, root,
7061                                                 new_extents[i].disk_bytenr,
7062                                                 new_extents[i].disk_num_bytes,
7063                                                 leaf->start,
7064                                                 root->root_key.objectid,
7065                                                 trans->transid, key.objectid);
7066                                 BUG_ON(ret);
7067                                 btrfs_release_path(root, path);
7068
7069                                 inode_add_bytes(inode, extent_len);
7070
7071                                 ext_offset = 0;
7072                                 num_bytes -= extent_len;
7073                                 key.offset += extent_len;
7074
7075                                 if (num_bytes == 0)
7076                                         break;
7077                         }
7078                         BUG_ON(i >= nr_extents);
7079 #endif
7080                 }
7081
7082                 if (extent_locked) {
7083                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7084                                       lock_end, GFP_NOFS);
7085                         extent_locked = 0;
7086                 }
7087 skip:
7088                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7089                     key.offset >= search_end)
7090                         break;
7091
7092                 cond_resched();
7093         }
7094         ret = 0;
7095 out:
7096         btrfs_release_path(root, path);
7097         if (inode) {
7098                 mutex_unlock(&inode->i_mutex);
7099                 if (extent_locked) {
7100                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7101                                       lock_end, GFP_NOFS);
7102                 }
7103                 iput(inode);
7104         }
7105         return ret;
7106 }
7107
7108 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7109                                struct btrfs_root *root,
7110                                struct extent_buffer *buf, u64 orig_start)
7111 {
7112         int level;
7113         int ret;
7114
7115         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7116         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7117
7118         level = btrfs_header_level(buf);
7119         if (level == 0) {
7120                 struct btrfs_leaf_ref *ref;
7121                 struct btrfs_leaf_ref *orig_ref;
7122
7123                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7124                 if (!orig_ref)
7125                         return -ENOENT;
7126
7127                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7128                 if (!ref) {
7129                         btrfs_free_leaf_ref(root, orig_ref);
7130                         return -ENOMEM;
7131                 }
7132
7133                 ref->nritems = orig_ref->nritems;
7134                 memcpy(ref->extents, orig_ref->extents,
7135                         sizeof(ref->extents[0]) * ref->nritems);
7136
7137                 btrfs_free_leaf_ref(root, orig_ref);
7138
7139                 ref->root_gen = trans->transid;
7140                 ref->bytenr = buf->start;
7141                 ref->owner = btrfs_header_owner(buf);
7142                 ref->generation = btrfs_header_generation(buf);
7143
7144                 ret = btrfs_add_leaf_ref(root, ref, 0);
7145                 WARN_ON(ret);
7146                 btrfs_free_leaf_ref(root, ref);
7147         }
7148         return 0;
7149 }
7150
7151 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7152                                         struct extent_buffer *leaf,
7153                                         struct btrfs_block_group_cache *group,
7154                                         struct btrfs_root *target_root)
7155 {
7156         struct btrfs_key key;
7157         struct inode *inode = NULL;
7158         struct btrfs_file_extent_item *fi;
7159         struct extent_state *cached_state = NULL;
7160         u64 num_bytes;
7161         u64 skip_objectid = 0;
7162         u32 nritems;
7163         u32 i;
7164
7165         nritems = btrfs_header_nritems(leaf);
7166         for (i = 0; i < nritems; i++) {
7167                 btrfs_item_key_to_cpu(leaf, &key, i);
7168                 if (key.objectid == skip_objectid ||
7169                     key.type != BTRFS_EXTENT_DATA_KEY)
7170                         continue;
7171                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7172                 if (btrfs_file_extent_type(leaf, fi) ==
7173                     BTRFS_FILE_EXTENT_INLINE)
7174                         continue;
7175                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7176                         continue;
7177                 if (!inode || inode->i_ino != key.objectid) {
7178                         iput(inode);
7179                         inode = btrfs_ilookup(target_root->fs_info->sb,
7180                                               key.objectid, target_root, 1);
7181                 }
7182                 if (!inode) {
7183                         skip_objectid = key.objectid;
7184                         continue;
7185                 }
7186                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7187
7188                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7189                                  key.offset + num_bytes - 1, 0, &cached_state,
7190                                  GFP_NOFS);
7191                 btrfs_drop_extent_cache(inode, key.offset,
7192                                         key.offset + num_bytes - 1, 1);
7193                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7194                                      key.offset + num_bytes - 1, &cached_state,
7195                                      GFP_NOFS);
7196                 cond_resched();
7197         }
7198         iput(inode);
7199         return 0;
7200 }
7201
7202 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7203                                         struct btrfs_root *root,
7204                                         struct extent_buffer *leaf,
7205                                         struct btrfs_block_group_cache *group,
7206                                         struct inode *reloc_inode)
7207 {
7208         struct btrfs_key key;
7209         struct btrfs_key extent_key;
7210         struct btrfs_file_extent_item *fi;
7211         struct btrfs_leaf_ref *ref;
7212         struct disk_extent *new_extent;
7213         u64 bytenr;
7214         u64 num_bytes;
7215         u32 nritems;
7216         u32 i;
7217         int ext_index;
7218         int nr_extent;
7219         int ret;
7220
7221         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7222         BUG_ON(!new_extent);
7223
7224         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7225         BUG_ON(!ref);
7226
7227         ext_index = -1;
7228         nritems = btrfs_header_nritems(leaf);
7229         for (i = 0; i < nritems; i++) {
7230                 btrfs_item_key_to_cpu(leaf, &key, i);
7231                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7232                         continue;
7233                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7234                 if (btrfs_file_extent_type(leaf, fi) ==
7235                     BTRFS_FILE_EXTENT_INLINE)
7236                         continue;
7237                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7238                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7239                 if (bytenr == 0)
7240                         continue;
7241
7242                 ext_index++;
7243                 if (bytenr >= group->key.objectid + group->key.offset ||
7244                     bytenr + num_bytes <= group->key.objectid)
7245                         continue;
7246
7247                 extent_key.objectid = bytenr;
7248                 extent_key.offset = num_bytes;
7249                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7250                 nr_extent = 1;
7251                 ret = get_new_locations(reloc_inode, &extent_key,
7252                                         group->key.objectid, 1,
7253                                         &new_extent, &nr_extent);
7254                 if (ret > 0)
7255                         continue;
7256                 BUG_ON(ret < 0);
7257
7258                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7259                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7260                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7261                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7262
7263                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7264                                                 new_extent->disk_bytenr);
7265                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7266                                                 new_extent->disk_num_bytes);
7267                 btrfs_mark_buffer_dirty(leaf);
7268
7269                 ret = btrfs_inc_extent_ref(trans, root,
7270                                         new_extent->disk_bytenr,
7271                                         new_extent->disk_num_bytes,
7272                                         leaf->start,
7273                                         root->root_key.objectid,
7274                                         trans->transid, key.objectid);
7275                 BUG_ON(ret);
7276
7277                 ret = btrfs_free_extent(trans, root,
7278                                         bytenr, num_bytes, leaf->start,
7279                                         btrfs_header_owner(leaf),
7280                                         btrfs_header_generation(leaf),
7281                                         key.objectid, 0);
7282                 BUG_ON(ret);
7283                 cond_resched();
7284         }
7285         kfree(new_extent);
7286         BUG_ON(ext_index + 1 != ref->nritems);
7287         btrfs_free_leaf_ref(root, ref);
7288         return 0;
7289 }
7290
7291 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7292                           struct btrfs_root *root)
7293 {
7294         struct btrfs_root *reloc_root;
7295         int ret;
7296
7297         if (root->reloc_root) {
7298                 reloc_root = root->reloc_root;
7299                 root->reloc_root = NULL;
7300                 list_add(&reloc_root->dead_list,
7301                          &root->fs_info->dead_reloc_roots);
7302
7303                 btrfs_set_root_bytenr(&reloc_root->root_item,
7304                                       reloc_root->node->start);
7305                 btrfs_set_root_level(&root->root_item,
7306                                      btrfs_header_level(reloc_root->node));
7307                 memset(&reloc_root->root_item.drop_progress, 0,
7308                         sizeof(struct btrfs_disk_key));
7309                 reloc_root->root_item.drop_level = 0;
7310
7311                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7312                                         &reloc_root->root_key,
7313                                         &reloc_root->root_item);
7314                 BUG_ON(ret);
7315         }
7316         return 0;
7317 }
7318
7319 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7320 {
7321         struct btrfs_trans_handle *trans;
7322         struct btrfs_root *reloc_root;
7323         struct btrfs_root *prev_root = NULL;
7324         struct list_head dead_roots;
7325         int ret;
7326         unsigned long nr;
7327
7328         INIT_LIST_HEAD(&dead_roots);
7329         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7330
7331         while (!list_empty(&dead_roots)) {
7332                 reloc_root = list_entry(dead_roots.prev,
7333                                         struct btrfs_root, dead_list);
7334                 list_del_init(&reloc_root->dead_list);
7335
7336                 BUG_ON(reloc_root->commit_root != NULL);
7337                 while (1) {
7338                         trans = btrfs_join_transaction(root, 1);
7339                         BUG_ON(!trans);
7340
7341                         mutex_lock(&root->fs_info->drop_mutex);
7342                         ret = btrfs_drop_snapshot(trans, reloc_root);
7343                         if (ret != -EAGAIN)
7344                                 break;
7345                         mutex_unlock(&root->fs_info->drop_mutex);
7346
7347                         nr = trans->blocks_used;
7348                         ret = btrfs_end_transaction(trans, root);
7349                         BUG_ON(ret);
7350                         btrfs_btree_balance_dirty(root, nr);
7351                 }
7352
7353                 free_extent_buffer(reloc_root->node);
7354
7355                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7356                                      &reloc_root->root_key);
7357                 BUG_ON(ret);
7358                 mutex_unlock(&root->fs_info->drop_mutex);
7359
7360                 nr = trans->blocks_used;
7361                 ret = btrfs_end_transaction(trans, root);
7362                 BUG_ON(ret);
7363                 btrfs_btree_balance_dirty(root, nr);
7364
7365                 kfree(prev_root);
7366                 prev_root = reloc_root;
7367         }
7368         if (prev_root) {
7369                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7370                 kfree(prev_root);
7371         }
7372         return 0;
7373 }
7374
7375 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7376 {
7377         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7378         return 0;
7379 }
7380
7381 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7382 {
7383         struct btrfs_root *reloc_root;
7384         struct btrfs_trans_handle *trans;
7385         struct btrfs_key location;
7386         int found;
7387         int ret;
7388
7389         mutex_lock(&root->fs_info->tree_reloc_mutex);
7390         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7391         BUG_ON(ret);
7392         found = !list_empty(&root->fs_info->dead_reloc_roots);
7393         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7394
7395         if (found) {
7396                 trans = btrfs_start_transaction(root, 1);
7397                 BUG_ON(!trans);
7398                 ret = btrfs_commit_transaction(trans, root);
7399                 BUG_ON(ret);
7400         }
7401
7402         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7403         location.offset = (u64)-1;
7404         location.type = BTRFS_ROOT_ITEM_KEY;
7405
7406         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7407         BUG_ON(!reloc_root);
7408         btrfs_orphan_cleanup(reloc_root);
7409         return 0;
7410 }
7411
7412 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7413                                     struct btrfs_root *root)
7414 {
7415         struct btrfs_root *reloc_root;
7416         struct extent_buffer *eb;
7417         struct btrfs_root_item *root_item;
7418         struct btrfs_key root_key;
7419         int ret;
7420
7421         BUG_ON(!root->ref_cows);
7422         if (root->reloc_root)
7423                 return 0;
7424
7425         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7426         BUG_ON(!root_item);
7427
7428         ret = btrfs_copy_root(trans, root, root->commit_root,
7429                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7430         BUG_ON(ret);
7431
7432         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7433         root_key.offset = root->root_key.objectid;
7434         root_key.type = BTRFS_ROOT_ITEM_KEY;
7435
7436         memcpy(root_item, &root->root_item, sizeof(root_item));
7437         btrfs_set_root_refs(root_item, 0);
7438         btrfs_set_root_bytenr(root_item, eb->start);
7439         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7440         btrfs_set_root_generation(root_item, trans->transid);
7441
7442         btrfs_tree_unlock(eb);
7443         free_extent_buffer(eb);
7444
7445         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7446                                 &root_key, root_item);
7447         BUG_ON(ret);
7448         kfree(root_item);
7449
7450         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7451                                                  &root_key);
7452         BUG_ON(!reloc_root);
7453         reloc_root->last_trans = trans->transid;
7454         reloc_root->commit_root = NULL;
7455         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7456
7457         root->reloc_root = reloc_root;
7458         return 0;
7459 }
7460
7461 /*
7462  * Core function of space balance.
7463  *
7464  * The idea is using reloc trees to relocate tree blocks in reference
7465  * counted roots. There is one reloc tree for each subvol, and all
7466  * reloc trees share same root key objectid. Reloc trees are snapshots
7467  * of the latest committed roots of subvols (root->commit_root).
7468  *
7469  * To relocate a tree block referenced by a subvol, there are two steps.
7470  * COW the block through subvol's reloc tree, then update block pointer
7471  * in the subvol to point to the new block. Since all reloc trees share
7472  * same root key objectid, doing special handing for tree blocks owned
7473  * by them is easy. Once a tree block has been COWed in one reloc tree,
7474  * we can use the resulting new block directly when the same block is
7475  * required to COW again through other reloc trees. By this way, relocated
7476  * tree blocks are shared between reloc trees, so they are also shared
7477  * between subvols.
7478  */
7479 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7480                                       struct btrfs_root *root,
7481                                       struct btrfs_path *path,
7482                                       struct btrfs_key *first_key,
7483                                       struct btrfs_ref_path *ref_path,
7484                                       struct btrfs_block_group_cache *group,
7485                                       struct inode *reloc_inode)
7486 {
7487         struct btrfs_root *reloc_root;
7488         struct extent_buffer *eb = NULL;
7489         struct btrfs_key *keys;
7490         u64 *nodes;
7491         int level;
7492         int shared_level;
7493         int lowest_level = 0;
7494         int ret;
7495
7496         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7497                 lowest_level = ref_path->owner_objectid;
7498
7499         if (!root->ref_cows) {
7500                 path->lowest_level = lowest_level;
7501                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7502                 BUG_ON(ret < 0);
7503                 path->lowest_level = 0;
7504                 btrfs_release_path(root, path);
7505                 return 0;
7506         }
7507
7508         mutex_lock(&root->fs_info->tree_reloc_mutex);
7509         ret = init_reloc_tree(trans, root);
7510         BUG_ON(ret);
7511         reloc_root = root->reloc_root;
7512
7513         shared_level = ref_path->shared_level;
7514         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7515
7516         keys = ref_path->node_keys;
7517         nodes = ref_path->new_nodes;
7518         memset(&keys[shared_level + 1], 0,
7519                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7520         memset(&nodes[shared_level + 1], 0,
7521                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7522
7523         if (nodes[lowest_level] == 0) {
7524                 path->lowest_level = lowest_level;
7525                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7526                                         0, 1);
7527                 BUG_ON(ret);
7528                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7529                         eb = path->nodes[level];
7530                         if (!eb || eb == reloc_root->node)
7531                                 break;
7532                         nodes[level] = eb->start;
7533                         if (level == 0)
7534                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7535                         else
7536                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7537                 }
7538                 if (nodes[0] &&
7539                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7540                         eb = path->nodes[0];
7541                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7542                                                       group, reloc_inode);
7543                         BUG_ON(ret);
7544                 }
7545                 btrfs_release_path(reloc_root, path);
7546         } else {
7547                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7548                                        lowest_level);
7549                 BUG_ON(ret);
7550         }
7551
7552         /*
7553          * replace tree blocks in the fs tree with tree blocks in
7554          * the reloc tree.
7555          */
7556         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7557         BUG_ON(ret < 0);
7558
7559         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7560                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7561                                         0, 0);
7562                 BUG_ON(ret);
7563                 extent_buffer_get(path->nodes[0]);
7564                 eb = path->nodes[0];
7565                 btrfs_release_path(reloc_root, path);
7566                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7567                 BUG_ON(ret);
7568                 free_extent_buffer(eb);
7569         }
7570
7571         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7572         path->lowest_level = 0;
7573         return 0;
7574 }
7575
7576 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7577                                         struct btrfs_root *root,
7578                                         struct btrfs_path *path,
7579                                         struct btrfs_key *first_key,
7580                                         struct btrfs_ref_path *ref_path)
7581 {
7582         int ret;
7583
7584         ret = relocate_one_path(trans, root, path, first_key,
7585                                 ref_path, NULL, NULL);
7586         BUG_ON(ret);
7587
7588         return 0;
7589 }
7590
7591 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7592                                     struct btrfs_root *extent_root,
7593                                     struct btrfs_path *path,
7594                                     struct btrfs_key *extent_key)
7595 {
7596         int ret;
7597
7598         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7599         if (ret)
7600                 goto out;
7601         ret = btrfs_del_item(trans, extent_root, path);
7602 out:
7603         btrfs_release_path(extent_root, path);
7604         return ret;
7605 }
7606
7607 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7608                                                 struct btrfs_ref_path *ref_path)
7609 {
7610         struct btrfs_key root_key;
7611
7612         root_key.objectid = ref_path->root_objectid;
7613         root_key.type = BTRFS_ROOT_ITEM_KEY;
7614         if (is_cowonly_root(ref_path->root_objectid))
7615                 root_key.offset = 0;
7616         else
7617                 root_key.offset = (u64)-1;
7618
7619         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7620 }
7621
7622 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7623                                         struct btrfs_path *path,
7624                                         struct btrfs_key *extent_key,
7625                                         struct btrfs_block_group_cache *group,
7626                                         struct inode *reloc_inode, int pass)
7627 {
7628         struct btrfs_trans_handle *trans;
7629         struct btrfs_root *found_root;
7630         struct btrfs_ref_path *ref_path = NULL;
7631         struct disk_extent *new_extents = NULL;
7632         int nr_extents = 0;
7633         int loops;
7634         int ret;
7635         int level;
7636         struct btrfs_key first_key;
7637         u64 prev_block = 0;
7638
7639
7640         trans = btrfs_start_transaction(extent_root, 1);
7641         BUG_ON(!trans);
7642
7643         if (extent_key->objectid == 0) {
7644                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7645                 goto out;
7646         }
7647
7648         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7649         if (!ref_path) {
7650                 ret = -ENOMEM;
7651                 goto out;
7652         }
7653
7654         for (loops = 0; ; loops++) {
7655                 if (loops == 0) {
7656                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7657                                                    extent_key->objectid);
7658                 } else {
7659                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7660                 }
7661                 if (ret < 0)
7662                         goto out;
7663                 if (ret > 0)
7664                         break;
7665
7666                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7667                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7668                         continue;
7669
7670                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7671                 BUG_ON(!found_root);
7672                 /*
7673                  * for reference counted tree, only process reference paths
7674                  * rooted at the latest committed root.
7675                  */
7676                 if (found_root->ref_cows &&
7677                     ref_path->root_generation != found_root->root_key.offset)
7678                         continue;
7679
7680                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7681                         if (pass == 0) {
7682                                 /*
7683                                  * copy data extents to new locations
7684                                  */
7685                                 u64 group_start = group->key.objectid;
7686                                 ret = relocate_data_extent(reloc_inode,
7687                                                            extent_key,
7688                                                            group_start);
7689                                 if (ret < 0)
7690                                         goto out;
7691                                 break;
7692                         }
7693                         level = 0;
7694                 } else {
7695                         level = ref_path->owner_objectid;
7696                 }
7697
7698                 if (prev_block != ref_path->nodes[level]) {
7699                         struct extent_buffer *eb;
7700                         u64 block_start = ref_path->nodes[level];
7701                         u64 block_size = btrfs_level_size(found_root, level);
7702
7703                         eb = read_tree_block(found_root, block_start,
7704                                              block_size, 0);
7705                         btrfs_tree_lock(eb);
7706                         BUG_ON(level != btrfs_header_level(eb));
7707
7708                         if (level == 0)
7709                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7710                         else
7711                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7712
7713                         btrfs_tree_unlock(eb);
7714                         free_extent_buffer(eb);
7715                         prev_block = block_start;
7716                 }
7717
7718                 mutex_lock(&extent_root->fs_info->trans_mutex);
7719                 btrfs_record_root_in_trans(found_root);
7720                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7721                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7722                         /*
7723                          * try to update data extent references while
7724                          * keeping metadata shared between snapshots.
7725                          */
7726                         if (pass == 1) {
7727                                 ret = relocate_one_path(trans, found_root,
7728                                                 path, &first_key, ref_path,
7729                                                 group, reloc_inode);
7730                                 if (ret < 0)
7731                                         goto out;
7732                                 continue;
7733                         }
7734                         /*
7735                          * use fallback method to process the remaining
7736                          * references.
7737                          */
7738                         if (!new_extents) {
7739                                 u64 group_start = group->key.objectid;
7740                                 new_extents = kmalloc(sizeof(*new_extents),
7741                                                       GFP_NOFS);
7742                                 nr_extents = 1;
7743                                 ret = get_new_locations(reloc_inode,
7744                                                         extent_key,
7745                                                         group_start, 1,
7746                                                         &new_extents,
7747                                                         &nr_extents);
7748                                 if (ret)
7749                                         goto out;
7750                         }
7751                         ret = replace_one_extent(trans, found_root,
7752                                                 path, extent_key,
7753                                                 &first_key, ref_path,
7754                                                 new_extents, nr_extents);
7755                 } else {
7756                         ret = relocate_tree_block(trans, found_root, path,
7757                                                   &first_key, ref_path);
7758                 }
7759                 if (ret < 0)
7760                         goto out;
7761         }
7762         ret = 0;
7763 out:
7764         btrfs_end_transaction(trans, extent_root);
7765         kfree(new_extents);
7766         kfree(ref_path);
7767         return ret;
7768 }
7769 #endif
7770
7771 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7772 {
7773         u64 num_devices;
7774         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7775                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7776
7777         num_devices = root->fs_info->fs_devices->rw_devices;
7778         if (num_devices == 1) {
7779                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7780                 stripped = flags & ~stripped;
7781
7782                 /* turn raid0 into single device chunks */
7783                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7784                         return stripped;
7785
7786                 /* turn mirroring into duplication */
7787                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7788                              BTRFS_BLOCK_GROUP_RAID10))
7789                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7790                 return flags;
7791         } else {
7792                 /* they already had raid on here, just return */
7793                 if (flags & stripped)
7794                         return flags;
7795
7796                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7797                 stripped = flags & ~stripped;
7798
7799                 /* switch duplicated blocks with raid1 */
7800                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7801                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7802
7803                 /* turn single device chunks into raid0 */
7804                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7805         }
7806         return flags;
7807 }
7808
7809 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
7810 {
7811         struct btrfs_space_info *sinfo = cache->space_info;
7812         u64 num_bytes;
7813         int ret = -ENOSPC;
7814
7815         if (cache->ro)
7816                 return 0;
7817
7818         spin_lock(&sinfo->lock);
7819         spin_lock(&cache->lock);
7820         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7821                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7822
7823         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7824             sinfo->bytes_may_use + sinfo->bytes_readonly +
7825             cache->reserved_pinned + num_bytes < sinfo->total_bytes) {
7826                 sinfo->bytes_readonly += num_bytes;
7827                 sinfo->bytes_reserved += cache->reserved_pinned;
7828                 cache->reserved_pinned = 0;
7829                 cache->ro = 1;
7830                 ret = 0;
7831         }
7832         spin_unlock(&cache->lock);
7833         spin_unlock(&sinfo->lock);
7834         return ret;
7835 }
7836
7837 int btrfs_set_block_group_ro(struct btrfs_root *root,
7838                              struct btrfs_block_group_cache *cache)
7839
7840 {
7841         struct btrfs_trans_handle *trans;
7842         u64 alloc_flags;
7843         int ret;
7844
7845         BUG_ON(cache->ro);
7846
7847         trans = btrfs_join_transaction(root, 1);
7848         BUG_ON(IS_ERR(trans));
7849
7850         alloc_flags = update_block_group_flags(root, cache->flags);
7851         if (alloc_flags != cache->flags)
7852                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7853
7854         ret = set_block_group_ro(cache);
7855         if (!ret)
7856                 goto out;
7857         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7858         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7859         if (ret < 0)
7860                 goto out;
7861         ret = set_block_group_ro(cache);
7862 out:
7863         btrfs_end_transaction(trans, root);
7864         return ret;
7865 }
7866
7867 int btrfs_set_block_group_rw(struct btrfs_root *root,
7868                               struct btrfs_block_group_cache *cache)
7869 {
7870         struct btrfs_space_info *sinfo = cache->space_info;
7871         u64 num_bytes;
7872
7873         BUG_ON(!cache->ro);
7874
7875         spin_lock(&sinfo->lock);
7876         spin_lock(&cache->lock);
7877         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7878                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7879         sinfo->bytes_readonly -= num_bytes;
7880         cache->ro = 0;
7881         spin_unlock(&cache->lock);
7882         spin_unlock(&sinfo->lock);
7883         return 0;
7884 }
7885
7886 /*
7887  * checks to see if its even possible to relocate this block group.
7888  *
7889  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7890  * ok to go ahead and try.
7891  */
7892 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7893 {
7894         struct btrfs_block_group_cache *block_group;
7895         struct btrfs_space_info *space_info;
7896         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7897         struct btrfs_device *device;
7898         int full = 0;
7899         int ret = 0;
7900
7901         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7902
7903         /* odd, couldn't find the block group, leave it alone */
7904         if (!block_group)
7905                 return -1;
7906
7907         /* no bytes used, we're good */
7908         if (!btrfs_block_group_used(&block_group->item))
7909                 goto out;
7910
7911         space_info = block_group->space_info;
7912         spin_lock(&space_info->lock);
7913
7914         full = space_info->full;
7915
7916         /*
7917          * if this is the last block group we have in this space, we can't
7918          * relocate it unless we're able to allocate a new chunk below.
7919          *
7920          * Otherwise, we need to make sure we have room in the space to handle
7921          * all of the extents from this block group.  If we can, we're good
7922          */
7923         if ((space_info->total_bytes != block_group->key.offset) &&
7924            (space_info->bytes_used + space_info->bytes_reserved +
7925             space_info->bytes_pinned + space_info->bytes_readonly +
7926             btrfs_block_group_used(&block_group->item) <
7927             space_info->total_bytes)) {
7928                 spin_unlock(&space_info->lock);
7929                 goto out;
7930         }
7931         spin_unlock(&space_info->lock);
7932
7933         /*
7934          * ok we don't have enough space, but maybe we have free space on our
7935          * devices to allocate new chunks for relocation, so loop through our
7936          * alloc devices and guess if we have enough space.  However, if we
7937          * were marked as full, then we know there aren't enough chunks, and we
7938          * can just return.
7939          */
7940         ret = -1;
7941         if (full)
7942                 goto out;
7943
7944         mutex_lock(&root->fs_info->chunk_mutex);
7945         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7946                 u64 min_free = btrfs_block_group_used(&block_group->item);
7947                 u64 dev_offset, max_avail;
7948
7949                 /*
7950                  * check to make sure we can actually find a chunk with enough
7951                  * space to fit our block group in.
7952                  */
7953                 if (device->total_bytes > device->bytes_used + min_free) {
7954                         ret = find_free_dev_extent(NULL, device, min_free,
7955                                                    &dev_offset, &max_avail);
7956                         if (!ret)
7957                                 break;
7958                         ret = -1;
7959                 }
7960         }
7961         mutex_unlock(&root->fs_info->chunk_mutex);
7962 out:
7963         btrfs_put_block_group(block_group);
7964         return ret;
7965 }
7966
7967 static int find_first_block_group(struct btrfs_root *root,
7968                 struct btrfs_path *path, struct btrfs_key *key)
7969 {
7970         int ret = 0;
7971         struct btrfs_key found_key;
7972         struct extent_buffer *leaf;
7973         int slot;
7974
7975         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7976         if (ret < 0)
7977                 goto out;
7978
7979         while (1) {
7980                 slot = path->slots[0];
7981                 leaf = path->nodes[0];
7982                 if (slot >= btrfs_header_nritems(leaf)) {
7983                         ret = btrfs_next_leaf(root, path);
7984                         if (ret == 0)
7985                                 continue;
7986                         if (ret < 0)
7987                                 goto out;
7988                         break;
7989                 }
7990                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7991
7992                 if (found_key.objectid >= key->objectid &&
7993                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7994                         ret = 0;
7995                         goto out;
7996                 }
7997                 path->slots[0]++;
7998         }
7999 out:
8000         return ret;
8001 }
8002
8003 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8004 {
8005         struct btrfs_block_group_cache *block_group;
8006         u64 last = 0;
8007
8008         while (1) {
8009                 struct inode *inode;
8010
8011                 block_group = btrfs_lookup_first_block_group(info, last);
8012                 while (block_group) {
8013                         spin_lock(&block_group->lock);
8014                         if (block_group->iref)
8015                                 break;
8016                         spin_unlock(&block_group->lock);
8017                         block_group = next_block_group(info->tree_root,
8018                                                        block_group);
8019                 }
8020                 if (!block_group) {
8021                         if (last == 0)
8022                                 break;
8023                         last = 0;
8024                         continue;
8025                 }
8026
8027                 inode = block_group->inode;
8028                 block_group->iref = 0;
8029                 block_group->inode = NULL;
8030                 spin_unlock(&block_group->lock);
8031                 iput(inode);
8032                 last = block_group->key.objectid + block_group->key.offset;
8033                 btrfs_put_block_group(block_group);
8034         }
8035 }
8036
8037 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8038 {
8039         struct btrfs_block_group_cache *block_group;
8040         struct btrfs_space_info *space_info;
8041         struct btrfs_caching_control *caching_ctl;
8042         struct rb_node *n;
8043
8044         down_write(&info->extent_commit_sem);
8045         while (!list_empty(&info->caching_block_groups)) {
8046                 caching_ctl = list_entry(info->caching_block_groups.next,
8047                                          struct btrfs_caching_control, list);
8048                 list_del(&caching_ctl->list);
8049                 put_caching_control(caching_ctl);
8050         }
8051         up_write(&info->extent_commit_sem);
8052
8053         spin_lock(&info->block_group_cache_lock);
8054         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8055                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8056                                        cache_node);
8057                 rb_erase(&block_group->cache_node,
8058                          &info->block_group_cache_tree);
8059                 spin_unlock(&info->block_group_cache_lock);
8060
8061                 down_write(&block_group->space_info->groups_sem);
8062                 list_del(&block_group->list);
8063                 up_write(&block_group->space_info->groups_sem);
8064
8065                 if (block_group->cached == BTRFS_CACHE_STARTED)
8066                         wait_block_group_cache_done(block_group);
8067
8068                 btrfs_remove_free_space_cache(block_group);
8069                 btrfs_put_block_group(block_group);
8070
8071                 spin_lock(&info->block_group_cache_lock);
8072         }
8073         spin_unlock(&info->block_group_cache_lock);
8074
8075         /* now that all the block groups are freed, go through and
8076          * free all the space_info structs.  This is only called during
8077          * the final stages of unmount, and so we know nobody is
8078          * using them.  We call synchronize_rcu() once before we start,
8079          * just to be on the safe side.
8080          */
8081         synchronize_rcu();
8082
8083         release_global_block_rsv(info);
8084
8085         while(!list_empty(&info->space_info)) {
8086                 space_info = list_entry(info->space_info.next,
8087                                         struct btrfs_space_info,
8088                                         list);
8089                 if (space_info->bytes_pinned > 0 ||
8090                     space_info->bytes_reserved > 0) {
8091                         WARN_ON(1);
8092                         dump_space_info(space_info, 0, 0);
8093                 }
8094                 list_del(&space_info->list);
8095                 kfree(space_info);
8096         }
8097         return 0;
8098 }
8099
8100 static void __link_block_group(struct btrfs_space_info *space_info,
8101                                struct btrfs_block_group_cache *cache)
8102 {
8103         int index = get_block_group_index(cache);
8104
8105         down_write(&space_info->groups_sem);
8106         list_add_tail(&cache->list, &space_info->block_groups[index]);
8107         up_write(&space_info->groups_sem);
8108 }
8109
8110 int btrfs_read_block_groups(struct btrfs_root *root)
8111 {
8112         struct btrfs_path *path;
8113         int ret;
8114         struct btrfs_block_group_cache *cache;
8115         struct btrfs_fs_info *info = root->fs_info;
8116         struct btrfs_space_info *space_info;
8117         struct btrfs_key key;
8118         struct btrfs_key found_key;
8119         struct extent_buffer *leaf;
8120         int need_clear = 0;
8121         u64 cache_gen;
8122
8123         root = info->extent_root;
8124         key.objectid = 0;
8125         key.offset = 0;
8126         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8127         path = btrfs_alloc_path();
8128         if (!path)
8129                 return -ENOMEM;
8130
8131         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8132         if (cache_gen != 0 &&
8133             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8134                 need_clear = 1;
8135
8136         while (1) {
8137                 ret = find_first_block_group(root, path, &key);
8138                 if (ret > 0)
8139                         break;
8140                 if (ret != 0)
8141                         goto error;
8142
8143                 leaf = path->nodes[0];
8144                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8145                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8146                 if (!cache) {
8147                         ret = -ENOMEM;
8148                         goto error;
8149                 }
8150
8151                 atomic_set(&cache->count, 1);
8152                 spin_lock_init(&cache->lock);
8153                 spin_lock_init(&cache->tree_lock);
8154                 cache->fs_info = info;
8155                 INIT_LIST_HEAD(&cache->list);
8156                 INIT_LIST_HEAD(&cache->cluster_list);
8157
8158                 if (need_clear)
8159                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8160
8161                 /*
8162                  * we only want to have 32k of ram per block group for keeping
8163                  * track of free space, and if we pass 1/2 of that we want to
8164                  * start converting things over to using bitmaps
8165                  */
8166                 cache->extents_thresh = ((1024 * 32) / 2) /
8167                         sizeof(struct btrfs_free_space);
8168
8169                 read_extent_buffer(leaf, &cache->item,
8170                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8171                                    sizeof(cache->item));
8172                 memcpy(&cache->key, &found_key, sizeof(found_key));
8173
8174                 key.objectid = found_key.objectid + found_key.offset;
8175                 btrfs_release_path(root, path);
8176                 cache->flags = btrfs_block_group_flags(&cache->item);
8177                 cache->sectorsize = root->sectorsize;
8178
8179                 /*
8180                  * check for two cases, either we are full, and therefore
8181                  * don't need to bother with the caching work since we won't
8182                  * find any space, or we are empty, and we can just add all
8183                  * the space in and be done with it.  This saves us _alot_ of
8184                  * time, particularly in the full case.
8185                  */
8186                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8187                         exclude_super_stripes(root, cache);
8188                         cache->last_byte_to_unpin = (u64)-1;
8189                         cache->cached = BTRFS_CACHE_FINISHED;
8190                         free_excluded_extents(root, cache);
8191                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8192                         exclude_super_stripes(root, cache);
8193                         cache->last_byte_to_unpin = (u64)-1;
8194                         cache->cached = BTRFS_CACHE_FINISHED;
8195                         add_new_free_space(cache, root->fs_info,
8196                                            found_key.objectid,
8197                                            found_key.objectid +
8198                                            found_key.offset);
8199                         free_excluded_extents(root, cache);
8200                 }
8201
8202                 ret = update_space_info(info, cache->flags, found_key.offset,
8203                                         btrfs_block_group_used(&cache->item),
8204                                         &space_info);
8205                 BUG_ON(ret);
8206                 cache->space_info = space_info;
8207                 spin_lock(&cache->space_info->lock);
8208                 cache->space_info->bytes_readonly += cache->bytes_super;
8209                 spin_unlock(&cache->space_info->lock);
8210
8211                 __link_block_group(space_info, cache);
8212
8213                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8214                 BUG_ON(ret);
8215
8216                 set_avail_alloc_bits(root->fs_info, cache->flags);
8217                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8218                         set_block_group_ro(cache);
8219         }
8220
8221         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8222                 if (!(get_alloc_profile(root, space_info->flags) &
8223                       (BTRFS_BLOCK_GROUP_RAID10 |
8224                        BTRFS_BLOCK_GROUP_RAID1 |
8225                        BTRFS_BLOCK_GROUP_DUP)))
8226                         continue;
8227                 /*
8228                  * avoid allocating from un-mirrored block group if there are
8229                  * mirrored block groups.
8230                  */
8231                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8232                         set_block_group_ro(cache);
8233                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8234                         set_block_group_ro(cache);
8235         }
8236
8237         init_global_block_rsv(info);
8238         ret = 0;
8239 error:
8240         btrfs_free_path(path);
8241         return ret;
8242 }
8243
8244 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8245                            struct btrfs_root *root, u64 bytes_used,
8246                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8247                            u64 size)
8248 {
8249         int ret;
8250         struct btrfs_root *extent_root;
8251         struct btrfs_block_group_cache *cache;
8252
8253         extent_root = root->fs_info->extent_root;
8254
8255         root->fs_info->last_trans_log_full_commit = trans->transid;
8256
8257         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8258         if (!cache)
8259                 return -ENOMEM;
8260
8261         cache->key.objectid = chunk_offset;
8262         cache->key.offset = size;
8263         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8264         cache->sectorsize = root->sectorsize;
8265         cache->fs_info = root->fs_info;
8266
8267         /*
8268          * we only want to have 32k of ram per block group for keeping track
8269          * of free space, and if we pass 1/2 of that we want to start
8270          * converting things over to using bitmaps
8271          */
8272         cache->extents_thresh = ((1024 * 32) / 2) /
8273                 sizeof(struct btrfs_free_space);
8274         atomic_set(&cache->count, 1);
8275         spin_lock_init(&cache->lock);
8276         spin_lock_init(&cache->tree_lock);
8277         INIT_LIST_HEAD(&cache->list);
8278         INIT_LIST_HEAD(&cache->cluster_list);
8279
8280         btrfs_set_block_group_used(&cache->item, bytes_used);
8281         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8282         cache->flags = type;
8283         btrfs_set_block_group_flags(&cache->item, type);
8284
8285         cache->last_byte_to_unpin = (u64)-1;
8286         cache->cached = BTRFS_CACHE_FINISHED;
8287         exclude_super_stripes(root, cache);
8288
8289         add_new_free_space(cache, root->fs_info, chunk_offset,
8290                            chunk_offset + size);
8291
8292         free_excluded_extents(root, cache);
8293
8294         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8295                                 &cache->space_info);
8296         BUG_ON(ret);
8297
8298         spin_lock(&cache->space_info->lock);
8299         cache->space_info->bytes_readonly += cache->bytes_super;
8300         spin_unlock(&cache->space_info->lock);
8301
8302         __link_block_group(cache->space_info, cache);
8303
8304         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8305         BUG_ON(ret);
8306
8307         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8308                                 sizeof(cache->item));
8309         BUG_ON(ret);
8310
8311         set_avail_alloc_bits(extent_root->fs_info, type);
8312
8313         return 0;
8314 }
8315
8316 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8317                              struct btrfs_root *root, u64 group_start)
8318 {
8319         struct btrfs_path *path;
8320         struct btrfs_block_group_cache *block_group;
8321         struct btrfs_free_cluster *cluster;
8322         struct btrfs_root *tree_root = root->fs_info->tree_root;
8323         struct btrfs_key key;
8324         struct inode *inode;
8325         int ret;
8326
8327         root = root->fs_info->extent_root;
8328
8329         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8330         BUG_ON(!block_group);
8331         BUG_ON(!block_group->ro);
8332
8333         /* make sure this block group isn't part of an allocation cluster */
8334         cluster = &root->fs_info->data_alloc_cluster;
8335         spin_lock(&cluster->refill_lock);
8336         btrfs_return_cluster_to_free_space(block_group, cluster);
8337         spin_unlock(&cluster->refill_lock);
8338
8339         /*
8340          * make sure this block group isn't part of a metadata
8341          * allocation cluster
8342          */
8343         cluster = &root->fs_info->meta_alloc_cluster;
8344         spin_lock(&cluster->refill_lock);
8345         btrfs_return_cluster_to_free_space(block_group, cluster);
8346         spin_unlock(&cluster->refill_lock);
8347
8348         path = btrfs_alloc_path();
8349         BUG_ON(!path);
8350
8351         inode = lookup_free_space_inode(root, block_group, path);
8352         if (!IS_ERR(inode)) {
8353                 btrfs_orphan_add(trans, inode);
8354                 clear_nlink(inode);
8355                 /* One for the block groups ref */
8356                 spin_lock(&block_group->lock);
8357                 if (block_group->iref) {
8358                         block_group->iref = 0;
8359                         block_group->inode = NULL;
8360                         spin_unlock(&block_group->lock);
8361                         iput(inode);
8362                 } else {
8363                         spin_unlock(&block_group->lock);
8364                 }
8365                 /* One for our lookup ref */
8366                 iput(inode);
8367         }
8368
8369         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8370         key.offset = block_group->key.objectid;
8371         key.type = 0;
8372
8373         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8374         if (ret < 0)
8375                 goto out;
8376         if (ret > 0)
8377                 btrfs_release_path(tree_root, path);
8378         if (ret == 0) {
8379                 ret = btrfs_del_item(trans, tree_root, path);
8380                 if (ret)
8381                         goto out;
8382                 btrfs_release_path(tree_root, path);
8383         }
8384
8385         spin_lock(&root->fs_info->block_group_cache_lock);
8386         rb_erase(&block_group->cache_node,
8387                  &root->fs_info->block_group_cache_tree);
8388         spin_unlock(&root->fs_info->block_group_cache_lock);
8389
8390         down_write(&block_group->space_info->groups_sem);
8391         /*
8392          * we must use list_del_init so people can check to see if they
8393          * are still on the list after taking the semaphore
8394          */
8395         list_del_init(&block_group->list);
8396         up_write(&block_group->space_info->groups_sem);
8397
8398         if (block_group->cached == BTRFS_CACHE_STARTED)
8399                 wait_block_group_cache_done(block_group);
8400
8401         btrfs_remove_free_space_cache(block_group);
8402
8403         spin_lock(&block_group->space_info->lock);
8404         block_group->space_info->total_bytes -= block_group->key.offset;
8405         block_group->space_info->bytes_readonly -= block_group->key.offset;
8406         spin_unlock(&block_group->space_info->lock);
8407
8408         memcpy(&key, &block_group->key, sizeof(key));
8409
8410         btrfs_clear_space_info_full(root->fs_info);
8411
8412         btrfs_put_block_group(block_group);
8413         btrfs_put_block_group(block_group);
8414
8415         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8416         if (ret > 0)
8417                 ret = -EIO;
8418         if (ret < 0)
8419                 goto out;
8420
8421         ret = btrfs_del_item(trans, root, path);
8422 out:
8423         btrfs_free_path(path);
8424         return ret;
8425 }