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 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2692                                    struct btrfs_root *root)
2693 {
2694         struct btrfs_block_group_cache *cache;
2695         int err = 0;
2696         struct btrfs_path *path;
2697         u64 last = 0;
2698
2699         path = btrfs_alloc_path();
2700         if (!path)
2701                 return -ENOMEM;
2702
2703         while (1) {
2704                 if (last == 0) {
2705                         err = btrfs_run_delayed_refs(trans, root,
2706                                                      (unsigned long)-1);
2707                         BUG_ON(err);
2708                 }
2709
2710                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2711                 while (cache) {
2712                         if (cache->dirty)
2713                                 break;
2714                         cache = next_block_group(root, cache);
2715                 }
2716                 if (!cache) {
2717                         if (last == 0)
2718                                 break;
2719                         last = 0;
2720                         continue;
2721                 }
2722
2723                 cache->dirty = 0;
2724                 last = cache->key.objectid + cache->key.offset;
2725
2726                 err = write_one_cache_group(trans, root, path, cache);
2727                 BUG_ON(err);
2728                 btrfs_put_block_group(cache);
2729         }
2730
2731         btrfs_free_path(path);
2732         return 0;
2733 }
2734
2735 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2736 {
2737         struct btrfs_block_group_cache *block_group;
2738         int readonly = 0;
2739
2740         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2741         if (!block_group || block_group->ro)
2742                 readonly = 1;
2743         if (block_group)
2744                 btrfs_put_block_group(block_group);
2745         return readonly;
2746 }
2747
2748 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2749                              u64 total_bytes, u64 bytes_used,
2750                              struct btrfs_space_info **space_info)
2751 {
2752         struct btrfs_space_info *found;
2753         int i;
2754         int factor;
2755
2756         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2757                      BTRFS_BLOCK_GROUP_RAID10))
2758                 factor = 2;
2759         else
2760                 factor = 1;
2761
2762         found = __find_space_info(info, flags);
2763         if (found) {
2764                 spin_lock(&found->lock);
2765                 found->total_bytes += total_bytes;
2766                 found->disk_total += total_bytes * factor;
2767                 found->bytes_used += bytes_used;
2768                 found->disk_used += bytes_used * factor;
2769                 found->full = 0;
2770                 spin_unlock(&found->lock);
2771                 *space_info = found;
2772                 return 0;
2773         }
2774         found = kzalloc(sizeof(*found), GFP_NOFS);
2775         if (!found)
2776                 return -ENOMEM;
2777
2778         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2779                 INIT_LIST_HEAD(&found->block_groups[i]);
2780         init_rwsem(&found->groups_sem);
2781         spin_lock_init(&found->lock);
2782         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2783                                 BTRFS_BLOCK_GROUP_SYSTEM |
2784                                 BTRFS_BLOCK_GROUP_METADATA);
2785         found->total_bytes = total_bytes;
2786         found->disk_total = total_bytes * factor;
2787         found->bytes_used = bytes_used;
2788         found->disk_used = bytes_used * factor;
2789         found->bytes_pinned = 0;
2790         found->bytes_reserved = 0;
2791         found->bytes_readonly = 0;
2792         found->bytes_may_use = 0;
2793         found->full = 0;
2794         found->force_alloc = 0;
2795         *space_info = found;
2796         list_add_rcu(&found->list, &info->space_info);
2797         atomic_set(&found->caching_threads, 0);
2798         return 0;
2799 }
2800
2801 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2802 {
2803         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2804                                    BTRFS_BLOCK_GROUP_RAID1 |
2805                                    BTRFS_BLOCK_GROUP_RAID10 |
2806                                    BTRFS_BLOCK_GROUP_DUP);
2807         if (extra_flags) {
2808                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2809                         fs_info->avail_data_alloc_bits |= extra_flags;
2810                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2811                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2812                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2813                         fs_info->avail_system_alloc_bits |= extra_flags;
2814         }
2815 }
2816
2817 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2818 {
2819         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2820
2821         if (num_devices == 1)
2822                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2823         if (num_devices < 4)
2824                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2825
2826         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2827             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2828                       BTRFS_BLOCK_GROUP_RAID10))) {
2829                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2830         }
2831
2832         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2833             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2834                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2835         }
2836
2837         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2838             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2839              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2840              (flags & BTRFS_BLOCK_GROUP_DUP)))
2841                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2842         return flags;
2843 }
2844
2845 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
2846 {
2847         if (flags & BTRFS_BLOCK_GROUP_DATA)
2848                 flags |= root->fs_info->avail_data_alloc_bits &
2849                          root->fs_info->data_alloc_profile;
2850         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2851                 flags |= root->fs_info->avail_system_alloc_bits &
2852                          root->fs_info->system_alloc_profile;
2853         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
2854                 flags |= root->fs_info->avail_metadata_alloc_bits &
2855                          root->fs_info->metadata_alloc_profile;
2856         return btrfs_reduce_alloc_profile(root, flags);
2857 }
2858
2859 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
2860 {
2861         u64 flags;
2862
2863         if (data)
2864                 flags = BTRFS_BLOCK_GROUP_DATA;
2865         else if (root == root->fs_info->chunk_root)
2866                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2867         else
2868                 flags = BTRFS_BLOCK_GROUP_METADATA;
2869
2870         return get_alloc_profile(root, flags);
2871 }
2872
2873 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2874 {
2875         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2876                                                        BTRFS_BLOCK_GROUP_DATA);
2877 }
2878
2879 /*
2880  * This will check the space that the inode allocates from to make sure we have
2881  * enough space for bytes.
2882  */
2883 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
2884 {
2885         struct btrfs_space_info *data_sinfo;
2886         struct btrfs_root *root = BTRFS_I(inode)->root;
2887         u64 used;
2888         int ret = 0, committed = 0;
2889
2890         /* make sure bytes are sectorsize aligned */
2891         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2892
2893         data_sinfo = BTRFS_I(inode)->space_info;
2894         if (!data_sinfo)
2895                 goto alloc;
2896
2897 again:
2898         /* make sure we have enough space to handle the data first */
2899         spin_lock(&data_sinfo->lock);
2900         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
2901                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
2902                 data_sinfo->bytes_may_use;
2903
2904         if (used + bytes > data_sinfo->total_bytes) {
2905                 struct btrfs_trans_handle *trans;
2906
2907                 /*
2908                  * if we don't have enough free bytes in this space then we need
2909                  * to alloc a new chunk.
2910                  */
2911                 if (!data_sinfo->full) {
2912                         u64 alloc_target;
2913
2914                         data_sinfo->force_alloc = 1;
2915                         spin_unlock(&data_sinfo->lock);
2916 alloc:
2917                         alloc_target = btrfs_get_alloc_profile(root, 1);
2918                         trans = btrfs_join_transaction(root, 1);
2919                         if (IS_ERR(trans))
2920                                 return PTR_ERR(trans);
2921
2922                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2923                                              bytes + 2 * 1024 * 1024,
2924                                              alloc_target, 0);
2925                         btrfs_end_transaction(trans, root);
2926                         if (ret < 0)
2927                                 return ret;
2928
2929                         if (!data_sinfo) {
2930                                 btrfs_set_inode_space_info(root, inode);
2931                                 data_sinfo = BTRFS_I(inode)->space_info;
2932                         }
2933                         goto again;
2934                 }
2935                 spin_unlock(&data_sinfo->lock);
2936
2937                 /* commit the current transaction and try again */
2938                 if (!committed && !root->fs_info->open_ioctl_trans) {
2939                         committed = 1;
2940                         trans = btrfs_join_transaction(root, 1);
2941                         if (IS_ERR(trans))
2942                                 return PTR_ERR(trans);
2943                         ret = btrfs_commit_transaction(trans, root);
2944                         if (ret)
2945                                 return ret;
2946                         goto again;
2947                 }
2948
2949 #if 0 /* I hope we never need this code again, just in case */
2950                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
2951                        "%llu bytes_reserved, " "%llu bytes_pinned, "
2952                        "%llu bytes_readonly, %llu may use %llu total\n",
2953                        (unsigned long long)bytes,
2954                        (unsigned long long)data_sinfo->bytes_used,
2955                        (unsigned long long)data_sinfo->bytes_reserved,
2956                        (unsigned long long)data_sinfo->bytes_pinned,
2957                        (unsigned long long)data_sinfo->bytes_readonly,
2958                        (unsigned long long)data_sinfo->bytes_may_use,
2959                        (unsigned long long)data_sinfo->total_bytes);
2960 #endif
2961                 return -ENOSPC;
2962         }
2963         data_sinfo->bytes_may_use += bytes;
2964         BTRFS_I(inode)->reserved_bytes += bytes;
2965         spin_unlock(&data_sinfo->lock);
2966
2967         return 0;
2968 }
2969
2970 /*
2971  * called when we are clearing an delalloc extent from the
2972  * inode's io_tree or there was an error for whatever reason
2973  * after calling btrfs_check_data_free_space
2974  */
2975 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
2976 {
2977         struct btrfs_root *root = BTRFS_I(inode)->root;
2978         struct btrfs_space_info *data_sinfo;
2979
2980         /* make sure bytes are sectorsize aligned */
2981         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2982
2983         data_sinfo = BTRFS_I(inode)->space_info;
2984         spin_lock(&data_sinfo->lock);
2985         data_sinfo->bytes_may_use -= bytes;
2986         BTRFS_I(inode)->reserved_bytes -= bytes;
2987         spin_unlock(&data_sinfo->lock);
2988 }
2989
2990 static void force_metadata_allocation(struct btrfs_fs_info *info)
2991 {
2992         struct list_head *head = &info->space_info;
2993         struct btrfs_space_info *found;
2994
2995         rcu_read_lock();
2996         list_for_each_entry_rcu(found, head, list) {
2997                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2998                         found->force_alloc = 1;
2999         }
3000         rcu_read_unlock();
3001 }
3002
3003 static int should_alloc_chunk(struct btrfs_space_info *sinfo,
3004                               u64 alloc_bytes)
3005 {
3006         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3007
3008         if (sinfo->bytes_used + sinfo->bytes_reserved +
3009             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3010                 return 0;
3011
3012         if (sinfo->bytes_used + sinfo->bytes_reserved +
3013             alloc_bytes < div_factor(num_bytes, 8))
3014                 return 0;
3015
3016         return 1;
3017 }
3018
3019 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3020                           struct btrfs_root *extent_root, u64 alloc_bytes,
3021                           u64 flags, int force)
3022 {
3023         struct btrfs_space_info *space_info;
3024         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3025         int ret = 0;
3026
3027         mutex_lock(&fs_info->chunk_mutex);
3028
3029         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3030
3031         space_info = __find_space_info(extent_root->fs_info, flags);
3032         if (!space_info) {
3033                 ret = update_space_info(extent_root->fs_info, flags,
3034                                         0, 0, &space_info);
3035                 BUG_ON(ret);
3036         }
3037         BUG_ON(!space_info);
3038
3039         spin_lock(&space_info->lock);
3040         if (space_info->force_alloc)
3041                 force = 1;
3042         if (space_info->full) {
3043                 spin_unlock(&space_info->lock);
3044                 goto out;
3045         }
3046
3047         if (!force && !should_alloc_chunk(space_info, alloc_bytes)) {
3048                 spin_unlock(&space_info->lock);
3049                 goto out;
3050         }
3051         spin_unlock(&space_info->lock);
3052
3053         /*
3054          * if we're doing a data chunk, go ahead and make sure that
3055          * we keep a reasonable number of metadata chunks allocated in the
3056          * FS as well.
3057          */
3058         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3059                 fs_info->data_chunk_allocations++;
3060                 if (!(fs_info->data_chunk_allocations %
3061                       fs_info->metadata_ratio))
3062                         force_metadata_allocation(fs_info);
3063         }
3064
3065         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3066         spin_lock(&space_info->lock);
3067         if (ret)
3068                 space_info->full = 1;
3069         else
3070                 ret = 1;
3071         space_info->force_alloc = 0;
3072         spin_unlock(&space_info->lock);
3073 out:
3074         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3075         return ret;
3076 }
3077
3078 static int maybe_allocate_chunk(struct btrfs_trans_handle *trans,
3079                                 struct btrfs_root *root,
3080                                 struct btrfs_space_info *sinfo, u64 num_bytes)
3081 {
3082         int ret;
3083         int end_trans = 0;
3084
3085         if (sinfo->full)
3086                 return 0;
3087
3088         spin_lock(&sinfo->lock);
3089         ret = should_alloc_chunk(sinfo, num_bytes + 2 * 1024 * 1024);
3090         spin_unlock(&sinfo->lock);
3091         if (!ret)
3092                 return 0;
3093
3094         if (!trans) {
3095                 trans = btrfs_join_transaction(root, 1);
3096                 BUG_ON(IS_ERR(trans));
3097                 end_trans = 1;
3098         }
3099
3100         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3101                              num_bytes + 2 * 1024 * 1024,
3102                              get_alloc_profile(root, sinfo->flags), 0);
3103
3104         if (end_trans)
3105                 btrfs_end_transaction(trans, root);
3106
3107         return ret == 1 ? 1 : 0;
3108 }
3109
3110 /*
3111  * shrink metadata reservation for delalloc
3112  */
3113 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3114                            struct btrfs_root *root, u64 to_reclaim)
3115 {
3116         struct btrfs_block_rsv *block_rsv;
3117         u64 reserved;
3118         u64 max_reclaim;
3119         u64 reclaimed = 0;
3120         int no_reclaim = 0;
3121         int pause = 1;
3122         int ret;
3123
3124         block_rsv = &root->fs_info->delalloc_block_rsv;
3125         spin_lock(&block_rsv->lock);
3126         reserved = block_rsv->reserved;
3127         spin_unlock(&block_rsv->lock);
3128
3129         if (reserved == 0)
3130                 return 0;
3131
3132         max_reclaim = min(reserved, to_reclaim);
3133
3134         while (1) {
3135                 ret = btrfs_start_one_delalloc_inode(root, trans ? 1 : 0);
3136                 if (!ret) {
3137                         if (no_reclaim > 2)
3138                                 break;
3139                         no_reclaim++;
3140                         __set_current_state(TASK_INTERRUPTIBLE);
3141                         schedule_timeout(pause);
3142                         pause <<= 1;
3143                         if (pause > HZ / 10)
3144                                 pause = HZ / 10;
3145                 } else {
3146                         no_reclaim = 0;
3147                         pause = 1;
3148                 }
3149
3150                 spin_lock(&block_rsv->lock);
3151                 if (reserved > block_rsv->reserved)
3152                         reclaimed = reserved - block_rsv->reserved;
3153                 reserved = block_rsv->reserved;
3154                 spin_unlock(&block_rsv->lock);
3155
3156                 if (reserved == 0 || reclaimed >= max_reclaim)
3157                         break;
3158
3159                 if (trans && trans->transaction->blocked)
3160                         return -EAGAIN;
3161         }
3162         return reclaimed >= to_reclaim;
3163 }
3164
3165 static int should_retry_reserve(struct btrfs_trans_handle *trans,
3166                                 struct btrfs_root *root,
3167                                 struct btrfs_block_rsv *block_rsv,
3168                                 u64 num_bytes, int *retries)
3169 {
3170         struct btrfs_space_info *space_info = block_rsv->space_info;
3171         int ret;
3172
3173         if ((*retries) > 2)
3174                 return -ENOSPC;
3175
3176         ret = maybe_allocate_chunk(trans, root, space_info, num_bytes);
3177         if (ret)
3178                 return 1;
3179
3180         if (trans && trans->transaction->in_commit)
3181                 return -ENOSPC;
3182
3183         ret = shrink_delalloc(trans, root, num_bytes);
3184         if (ret)
3185                 return ret;
3186
3187         spin_lock(&space_info->lock);
3188         if (space_info->bytes_pinned < num_bytes)
3189                 ret = 1;
3190         spin_unlock(&space_info->lock);
3191         if (ret)
3192                 return -ENOSPC;
3193
3194         (*retries)++;
3195
3196         if (trans)
3197                 return -EAGAIN;
3198
3199         trans = btrfs_join_transaction(root, 1);
3200         BUG_ON(IS_ERR(trans));
3201         ret = btrfs_commit_transaction(trans, root);
3202         BUG_ON(ret);
3203
3204         return 1;
3205 }
3206
3207 static int reserve_metadata_bytes(struct btrfs_block_rsv *block_rsv,
3208                                   u64 num_bytes)
3209 {
3210         struct btrfs_space_info *space_info = block_rsv->space_info;
3211         u64 unused;
3212         int ret = -ENOSPC;
3213
3214         spin_lock(&space_info->lock);
3215         unused = space_info->bytes_used + space_info->bytes_reserved +
3216                  space_info->bytes_pinned + space_info->bytes_readonly;
3217
3218         if (unused < space_info->total_bytes)
3219                 unused = space_info->total_bytes - unused;
3220         else
3221                 unused = 0;
3222
3223         if (unused >= num_bytes) {
3224                 if (block_rsv->priority >= 10) {
3225                         space_info->bytes_reserved += num_bytes;
3226                         ret = 0;
3227                 } else {
3228                         if ((unused + block_rsv->reserved) *
3229                             block_rsv->priority >=
3230                             (num_bytes + block_rsv->reserved) * 10) {
3231                                 space_info->bytes_reserved += num_bytes;
3232                                 ret = 0;
3233                         }
3234                 }
3235         }
3236         spin_unlock(&space_info->lock);
3237
3238         return ret;
3239 }
3240
3241 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3242                                              struct btrfs_root *root)
3243 {
3244         struct btrfs_block_rsv *block_rsv;
3245         if (root->ref_cows)
3246                 block_rsv = trans->block_rsv;
3247         else
3248                 block_rsv = root->block_rsv;
3249
3250         if (!block_rsv)
3251                 block_rsv = &root->fs_info->empty_block_rsv;
3252
3253         return block_rsv;
3254 }
3255
3256 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3257                                u64 num_bytes)
3258 {
3259         int ret = -ENOSPC;
3260         spin_lock(&block_rsv->lock);
3261         if (block_rsv->reserved >= num_bytes) {
3262                 block_rsv->reserved -= num_bytes;
3263                 if (block_rsv->reserved < block_rsv->size)
3264                         block_rsv->full = 0;
3265                 ret = 0;
3266         }
3267         spin_unlock(&block_rsv->lock);
3268         return ret;
3269 }
3270
3271 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3272                                 u64 num_bytes, int update_size)
3273 {
3274         spin_lock(&block_rsv->lock);
3275         block_rsv->reserved += num_bytes;
3276         if (update_size)
3277                 block_rsv->size += num_bytes;
3278         else if (block_rsv->reserved >= block_rsv->size)
3279                 block_rsv->full = 1;
3280         spin_unlock(&block_rsv->lock);
3281 }
3282
3283 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3284                              struct btrfs_block_rsv *dest, u64 num_bytes)
3285 {
3286         struct btrfs_space_info *space_info = block_rsv->space_info;
3287
3288         spin_lock(&block_rsv->lock);
3289         if (num_bytes == (u64)-1)
3290                 num_bytes = block_rsv->size;
3291         block_rsv->size -= num_bytes;
3292         if (block_rsv->reserved >= block_rsv->size) {
3293                 num_bytes = block_rsv->reserved - block_rsv->size;
3294                 block_rsv->reserved = block_rsv->size;
3295                 block_rsv->full = 1;
3296         } else {
3297                 num_bytes = 0;
3298         }
3299         spin_unlock(&block_rsv->lock);
3300
3301         if (num_bytes > 0) {
3302                 if (dest) {
3303                         block_rsv_add_bytes(dest, num_bytes, 0);
3304                 } else {
3305                         spin_lock(&space_info->lock);
3306                         space_info->bytes_reserved -= num_bytes;
3307                         spin_unlock(&space_info->lock);
3308                 }
3309         }
3310 }
3311
3312 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3313                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3314 {
3315         int ret;
3316
3317         ret = block_rsv_use_bytes(src, num_bytes);
3318         if (ret)
3319                 return ret;
3320
3321         block_rsv_add_bytes(dst, num_bytes, 1);
3322         return 0;
3323 }
3324
3325 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3326 {
3327         memset(rsv, 0, sizeof(*rsv));
3328         spin_lock_init(&rsv->lock);
3329         atomic_set(&rsv->usage, 1);
3330         rsv->priority = 6;
3331         INIT_LIST_HEAD(&rsv->list);
3332 }
3333
3334 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3335 {
3336         struct btrfs_block_rsv *block_rsv;
3337         struct btrfs_fs_info *fs_info = root->fs_info;
3338         u64 alloc_target;
3339
3340         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3341         if (!block_rsv)
3342                 return NULL;
3343
3344         btrfs_init_block_rsv(block_rsv);
3345
3346         alloc_target = btrfs_get_alloc_profile(root, 0);
3347         block_rsv->space_info = __find_space_info(fs_info,
3348                                                   BTRFS_BLOCK_GROUP_METADATA);
3349
3350         return block_rsv;
3351 }
3352
3353 void btrfs_free_block_rsv(struct btrfs_root *root,
3354                           struct btrfs_block_rsv *rsv)
3355 {
3356         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3357                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3358                 if (!rsv->durable)
3359                         kfree(rsv);
3360         }
3361 }
3362
3363 /*
3364  * make the block_rsv struct be able to capture freed space.
3365  * the captured space will re-add to the the block_rsv struct
3366  * after transaction commit
3367  */
3368 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3369                                  struct btrfs_block_rsv *block_rsv)
3370 {
3371         block_rsv->durable = 1;
3372         mutex_lock(&fs_info->durable_block_rsv_mutex);
3373         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3374         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3375 }
3376
3377 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3378                         struct btrfs_root *root,
3379                         struct btrfs_block_rsv *block_rsv,
3380                         u64 num_bytes, int *retries)
3381 {
3382         int ret;
3383
3384         if (num_bytes == 0)
3385                 return 0;
3386 again:
3387         ret = reserve_metadata_bytes(block_rsv, num_bytes);
3388         if (!ret) {
3389                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3390                 return 0;
3391         }
3392
3393         ret = should_retry_reserve(trans, root, block_rsv, num_bytes, retries);
3394         if (ret > 0)
3395                 goto again;
3396
3397         return ret;
3398 }
3399
3400 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3401                           struct btrfs_root *root,
3402                           struct btrfs_block_rsv *block_rsv,
3403                           u64 min_reserved, int min_factor)
3404 {
3405         u64 num_bytes = 0;
3406         int commit_trans = 0;
3407         int ret = -ENOSPC;
3408
3409         if (!block_rsv)
3410                 return 0;
3411
3412         spin_lock(&block_rsv->lock);
3413         if (min_factor > 0)
3414                 num_bytes = div_factor(block_rsv->size, min_factor);
3415         if (min_reserved > num_bytes)
3416                 num_bytes = min_reserved;
3417
3418         if (block_rsv->reserved >= num_bytes) {
3419                 ret = 0;
3420         } else {
3421                 num_bytes -= block_rsv->reserved;
3422                 if (block_rsv->durable &&
3423                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3424                         commit_trans = 1;
3425         }
3426         spin_unlock(&block_rsv->lock);
3427         if (!ret)
3428                 return 0;
3429
3430         if (block_rsv->refill_used) {
3431                 ret = reserve_metadata_bytes(block_rsv, num_bytes);
3432                 if (!ret) {
3433                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3434                         return 0;
3435                 }
3436         }
3437
3438         if (commit_trans) {
3439                 if (trans)
3440                         return -EAGAIN;
3441
3442                 trans = btrfs_join_transaction(root, 1);
3443                 BUG_ON(IS_ERR(trans));
3444                 ret = btrfs_commit_transaction(trans, root);
3445                 return 0;
3446         }
3447
3448         WARN_ON(1);
3449         printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
3450                 block_rsv->size, block_rsv->reserved,
3451                 block_rsv->freed[0], block_rsv->freed[1]);
3452
3453         return -ENOSPC;
3454 }
3455
3456 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3457                             struct btrfs_block_rsv *dst_rsv,
3458                             u64 num_bytes)
3459 {
3460         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3461 }
3462
3463 void btrfs_block_rsv_release(struct btrfs_root *root,
3464                              struct btrfs_block_rsv *block_rsv,
3465                              u64 num_bytes)
3466 {
3467         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3468         if (global_rsv->full || global_rsv == block_rsv ||
3469             block_rsv->space_info != global_rsv->space_info)
3470                 global_rsv = NULL;
3471         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3472 }
3473
3474 /*
3475  * helper to calculate size of global block reservation.
3476  * the desired value is sum of space used by extent tree,
3477  * checksum tree and root tree
3478  */
3479 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3480 {
3481         struct btrfs_space_info *sinfo;
3482         u64 num_bytes;
3483         u64 meta_used;
3484         u64 data_used;
3485         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3486 #if 0
3487         /*
3488          * per tree used space accounting can be inaccuracy, so we
3489          * can't rely on it.
3490          */
3491         spin_lock(&fs_info->extent_root->accounting_lock);
3492         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3493         spin_unlock(&fs_info->extent_root->accounting_lock);
3494
3495         spin_lock(&fs_info->csum_root->accounting_lock);
3496         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3497         spin_unlock(&fs_info->csum_root->accounting_lock);
3498
3499         spin_lock(&fs_info->tree_root->accounting_lock);
3500         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3501         spin_unlock(&fs_info->tree_root->accounting_lock);
3502 #endif
3503         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3504         spin_lock(&sinfo->lock);
3505         data_used = sinfo->bytes_used;
3506         spin_unlock(&sinfo->lock);
3507
3508         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3509         spin_lock(&sinfo->lock);
3510         meta_used = sinfo->bytes_used;
3511         spin_unlock(&sinfo->lock);
3512
3513         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3514                     csum_size * 2;
3515         num_bytes += div64_u64(data_used + meta_used, 50);
3516
3517         if (num_bytes * 3 > meta_used)
3518                 num_bytes = div64_u64(meta_used, 3);
3519
3520         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3521 }
3522
3523 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3524 {
3525         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3526         struct btrfs_space_info *sinfo = block_rsv->space_info;
3527         u64 num_bytes;
3528
3529         num_bytes = calc_global_metadata_size(fs_info);
3530
3531         spin_lock(&block_rsv->lock);
3532         spin_lock(&sinfo->lock);
3533
3534         block_rsv->size = num_bytes;
3535
3536         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3537                     sinfo->bytes_reserved + sinfo->bytes_readonly;
3538
3539         if (sinfo->total_bytes > num_bytes) {
3540                 num_bytes = sinfo->total_bytes - num_bytes;
3541                 block_rsv->reserved += num_bytes;
3542                 sinfo->bytes_reserved += num_bytes;
3543         }
3544
3545         if (block_rsv->reserved >= block_rsv->size) {
3546                 num_bytes = block_rsv->reserved - block_rsv->size;
3547                 sinfo->bytes_reserved -= num_bytes;
3548                 block_rsv->reserved = block_rsv->size;
3549                 block_rsv->full = 1;
3550         }
3551 #if 0
3552         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3553                 block_rsv->size, block_rsv->reserved);
3554 #endif
3555         spin_unlock(&sinfo->lock);
3556         spin_unlock(&block_rsv->lock);
3557 }
3558
3559 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3560 {
3561         struct btrfs_space_info *space_info;
3562
3563         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3564         fs_info->chunk_block_rsv.space_info = space_info;
3565         fs_info->chunk_block_rsv.priority = 10;
3566
3567         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3568         fs_info->global_block_rsv.space_info = space_info;
3569         fs_info->global_block_rsv.priority = 10;
3570         fs_info->global_block_rsv.refill_used = 1;
3571         fs_info->delalloc_block_rsv.space_info = space_info;
3572         fs_info->trans_block_rsv.space_info = space_info;
3573         fs_info->empty_block_rsv.space_info = space_info;
3574         fs_info->empty_block_rsv.priority = 10;
3575
3576         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3577         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3578         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3579         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3580         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3581
3582         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3583
3584         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3585
3586         update_global_block_rsv(fs_info);
3587 }
3588
3589 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3590 {
3591         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3592         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3593         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3594         WARN_ON(fs_info->trans_block_rsv.size > 0);
3595         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3596         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3597         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3598 }
3599
3600 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3601 {
3602         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3603                 3 * num_items;
3604 }
3605
3606 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3607                                  struct btrfs_root *root,
3608                                  int num_items, int *retries)
3609 {
3610         u64 num_bytes;
3611         int ret;
3612
3613         if (num_items == 0 || root->fs_info->chunk_root == root)
3614                 return 0;
3615
3616         num_bytes = calc_trans_metadata_size(root, num_items);
3617         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3618                                   num_bytes, retries);
3619         if (!ret) {
3620                 trans->bytes_reserved += num_bytes;
3621                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3622         }
3623         return ret;
3624 }
3625
3626 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3627                                   struct btrfs_root *root)
3628 {
3629         if (!trans->bytes_reserved)
3630                 return;
3631
3632         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3633         btrfs_block_rsv_release(root, trans->block_rsv,
3634                                 trans->bytes_reserved);
3635         trans->bytes_reserved = 0;
3636 }
3637
3638 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3639                                   struct inode *inode)
3640 {
3641         struct btrfs_root *root = BTRFS_I(inode)->root;
3642         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3643         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3644
3645         /*
3646          * one for deleting orphan item, one for updating inode and
3647          * two for calling btrfs_truncate_inode_items.
3648          *
3649          * btrfs_truncate_inode_items is a delete operation, it frees
3650          * more space than it uses in most cases. So two units of
3651          * metadata space should be enough for calling it many times.
3652          * If all of the metadata space is used, we can commit
3653          * transaction and use space it freed.
3654          */
3655         u64 num_bytes = calc_trans_metadata_size(root, 4);
3656         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3657 }
3658
3659 void btrfs_orphan_release_metadata(struct inode *inode)
3660 {
3661         struct btrfs_root *root = BTRFS_I(inode)->root;
3662         u64 num_bytes = calc_trans_metadata_size(root, 4);
3663         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3664 }
3665
3666 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3667                                 struct btrfs_pending_snapshot *pending)
3668 {
3669         struct btrfs_root *root = pending->root;
3670         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3671         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3672         /*
3673          * two for root back/forward refs, two for directory entries
3674          * and one for root of the snapshot.
3675          */
3676         u64 num_bytes = calc_trans_metadata_size(root, 5);
3677         dst_rsv->space_info = src_rsv->space_info;
3678         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3679 }
3680
3681 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3682 {
3683         return num_bytes >>= 3;
3684 }
3685
3686 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3687 {
3688         struct btrfs_root *root = BTRFS_I(inode)->root;
3689         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3690         u64 to_reserve;
3691         int nr_extents;
3692         int retries = 0;
3693         int ret;
3694
3695         if (btrfs_transaction_in_commit(root->fs_info))
3696                 schedule_timeout(1);
3697
3698         num_bytes = ALIGN(num_bytes, root->sectorsize);
3699 again:
3700         spin_lock(&BTRFS_I(inode)->accounting_lock);
3701         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
3702         if (nr_extents > BTRFS_I(inode)->reserved_extents) {
3703                 nr_extents -= BTRFS_I(inode)->reserved_extents;
3704                 to_reserve = calc_trans_metadata_size(root, nr_extents);
3705         } else {
3706                 nr_extents = 0;
3707                 to_reserve = 0;
3708         }
3709
3710         to_reserve += calc_csum_metadata_size(inode, num_bytes);
3711         ret = reserve_metadata_bytes(block_rsv, to_reserve);
3712         if (ret) {
3713                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3714                 ret = should_retry_reserve(NULL, root, block_rsv, to_reserve,
3715                                            &retries);
3716                 if (ret > 0)
3717                         goto again;
3718                 return ret;
3719         }
3720
3721         BTRFS_I(inode)->reserved_extents += nr_extents;
3722         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
3723         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3724
3725         block_rsv_add_bytes(block_rsv, to_reserve, 1);
3726
3727         if (block_rsv->size > 512 * 1024 * 1024)
3728                 shrink_delalloc(NULL, root, to_reserve);
3729
3730         return 0;
3731 }
3732
3733 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
3734 {
3735         struct btrfs_root *root = BTRFS_I(inode)->root;
3736         u64 to_free;
3737         int nr_extents;
3738
3739         num_bytes = ALIGN(num_bytes, root->sectorsize);
3740         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
3741
3742         spin_lock(&BTRFS_I(inode)->accounting_lock);
3743         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
3744         if (nr_extents < BTRFS_I(inode)->reserved_extents) {
3745                 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
3746                 BTRFS_I(inode)->reserved_extents -= nr_extents;
3747         } else {
3748                 nr_extents = 0;
3749         }
3750         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3751
3752         to_free = calc_csum_metadata_size(inode, num_bytes);
3753         if (nr_extents > 0)
3754                 to_free += calc_trans_metadata_size(root, nr_extents);
3755
3756         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
3757                                 to_free);
3758 }
3759
3760 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
3761 {
3762         int ret;
3763
3764         ret = btrfs_check_data_free_space(inode, num_bytes);
3765         if (ret)
3766                 return ret;
3767
3768         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
3769         if (ret) {
3770                 btrfs_free_reserved_data_space(inode, num_bytes);
3771                 return ret;
3772         }
3773
3774         return 0;
3775 }
3776
3777 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
3778 {
3779         btrfs_delalloc_release_metadata(inode, num_bytes);
3780         btrfs_free_reserved_data_space(inode, num_bytes);
3781 }
3782
3783 static int update_block_group(struct btrfs_trans_handle *trans,
3784                               struct btrfs_root *root,
3785                               u64 bytenr, u64 num_bytes, int alloc)
3786 {
3787         struct btrfs_block_group_cache *cache;
3788         struct btrfs_fs_info *info = root->fs_info;
3789         int factor;
3790         u64 total = num_bytes;
3791         u64 old_val;
3792         u64 byte_in_group;
3793
3794         /* block accounting for super block */
3795         spin_lock(&info->delalloc_lock);
3796         old_val = btrfs_super_bytes_used(&info->super_copy);
3797         if (alloc)
3798                 old_val += num_bytes;
3799         else
3800                 old_val -= num_bytes;
3801         btrfs_set_super_bytes_used(&info->super_copy, old_val);
3802         spin_unlock(&info->delalloc_lock);
3803
3804         while (total) {
3805                 cache = btrfs_lookup_block_group(info, bytenr);
3806                 if (!cache)
3807                         return -1;
3808                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
3809                                     BTRFS_BLOCK_GROUP_RAID1 |
3810                                     BTRFS_BLOCK_GROUP_RAID10))
3811                         factor = 2;
3812                 else
3813                         factor = 1;
3814                 byte_in_group = bytenr - cache->key.objectid;
3815                 WARN_ON(byte_in_group > cache->key.offset);
3816
3817                 spin_lock(&cache->space_info->lock);
3818                 spin_lock(&cache->lock);
3819                 cache->dirty = 1;
3820                 old_val = btrfs_block_group_used(&cache->item);
3821                 num_bytes = min(total, cache->key.offset - byte_in_group);
3822                 if (alloc) {
3823                         old_val += num_bytes;
3824                         btrfs_set_block_group_used(&cache->item, old_val);
3825                         cache->reserved -= num_bytes;
3826                         cache->space_info->bytes_reserved -= num_bytes;
3827                         cache->space_info->bytes_used += num_bytes;
3828                         cache->space_info->disk_used += num_bytes * factor;
3829                         spin_unlock(&cache->lock);
3830                         spin_unlock(&cache->space_info->lock);
3831                 } else {
3832                         old_val -= num_bytes;
3833                         btrfs_set_block_group_used(&cache->item, old_val);
3834                         cache->pinned += num_bytes;
3835                         cache->space_info->bytes_pinned += num_bytes;
3836                         cache->space_info->bytes_used -= num_bytes;
3837                         cache->space_info->disk_used -= num_bytes * factor;
3838                         spin_unlock(&cache->lock);
3839                         spin_unlock(&cache->space_info->lock);
3840
3841                         set_extent_dirty(info->pinned_extents,
3842                                          bytenr, bytenr + num_bytes - 1,
3843                                          GFP_NOFS | __GFP_NOFAIL);
3844                 }
3845                 btrfs_put_block_group(cache);
3846                 total -= num_bytes;
3847                 bytenr += num_bytes;
3848         }
3849         return 0;
3850 }
3851
3852 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3853 {
3854         struct btrfs_block_group_cache *cache;
3855         u64 bytenr;
3856
3857         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3858         if (!cache)
3859                 return 0;
3860
3861         bytenr = cache->key.objectid;
3862         btrfs_put_block_group(cache);
3863
3864         return bytenr;
3865 }
3866
3867 static int pin_down_extent(struct btrfs_root *root,
3868                            struct btrfs_block_group_cache *cache,
3869                            u64 bytenr, u64 num_bytes, int reserved)
3870 {
3871         spin_lock(&cache->space_info->lock);
3872         spin_lock(&cache->lock);
3873         cache->pinned += num_bytes;
3874         cache->space_info->bytes_pinned += num_bytes;
3875         if (reserved) {
3876                 cache->reserved -= num_bytes;
3877                 cache->space_info->bytes_reserved -= num_bytes;
3878         }
3879         spin_unlock(&cache->lock);
3880         spin_unlock(&cache->space_info->lock);
3881
3882         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
3883                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
3884         return 0;
3885 }
3886
3887 /*
3888  * this function must be called within transaction
3889  */
3890 int btrfs_pin_extent(struct btrfs_root *root,
3891                      u64 bytenr, u64 num_bytes, int reserved)
3892 {
3893         struct btrfs_block_group_cache *cache;
3894
3895         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
3896         BUG_ON(!cache);
3897
3898         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
3899
3900         btrfs_put_block_group(cache);
3901         return 0;
3902 }
3903
3904 /*
3905  * update size of reserved extents. this function may return -EAGAIN
3906  * if 'reserve' is true or 'sinfo' is false.
3907  */
3908 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
3909                                  u64 num_bytes, int reserve, int sinfo)
3910 {
3911         int ret = 0;
3912         if (sinfo) {
3913                 struct btrfs_space_info *space_info = cache->space_info;
3914                 spin_lock(&space_info->lock);
3915                 spin_lock(&cache->lock);
3916                 if (reserve) {
3917                         if (cache->ro) {
3918                                 ret = -EAGAIN;
3919                         } else {
3920                                 cache->reserved += num_bytes;
3921                                 space_info->bytes_reserved += num_bytes;
3922                         }
3923                 } else {
3924                         if (cache->ro)
3925                                 space_info->bytes_readonly += num_bytes;
3926                         cache->reserved -= num_bytes;
3927                         space_info->bytes_reserved -= num_bytes;
3928                 }
3929                 spin_unlock(&cache->lock);
3930                 spin_unlock(&space_info->lock);
3931         } else {
3932                 spin_lock(&cache->lock);
3933                 if (cache->ro) {
3934                         ret = -EAGAIN;
3935                 } else {
3936                         if (reserve)
3937                                 cache->reserved += num_bytes;
3938                         else
3939                                 cache->reserved -= num_bytes;
3940                 }
3941                 spin_unlock(&cache->lock);
3942         }
3943         return ret;
3944 }
3945
3946 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3947                                 struct btrfs_root *root)
3948 {
3949         struct btrfs_fs_info *fs_info = root->fs_info;
3950         struct btrfs_caching_control *next;
3951         struct btrfs_caching_control *caching_ctl;
3952         struct btrfs_block_group_cache *cache;
3953
3954         down_write(&fs_info->extent_commit_sem);
3955
3956         list_for_each_entry_safe(caching_ctl, next,
3957                                  &fs_info->caching_block_groups, list) {
3958                 cache = caching_ctl->block_group;
3959                 if (block_group_cache_done(cache)) {
3960                         cache->last_byte_to_unpin = (u64)-1;
3961                         list_del_init(&caching_ctl->list);
3962                         put_caching_control(caching_ctl);
3963                 } else {
3964                         cache->last_byte_to_unpin = caching_ctl->progress;
3965                 }
3966         }
3967
3968         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3969                 fs_info->pinned_extents = &fs_info->freed_extents[1];
3970         else
3971                 fs_info->pinned_extents = &fs_info->freed_extents[0];
3972
3973         up_write(&fs_info->extent_commit_sem);
3974
3975         update_global_block_rsv(fs_info);
3976         return 0;
3977 }
3978
3979 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3980 {
3981         struct btrfs_fs_info *fs_info = root->fs_info;
3982         struct btrfs_block_group_cache *cache = NULL;
3983         u64 len;
3984
3985         while (start <= end) {
3986                 if (!cache ||
3987                     start >= cache->key.objectid + cache->key.offset) {
3988                         if (cache)
3989                                 btrfs_put_block_group(cache);
3990                         cache = btrfs_lookup_block_group(fs_info, start);
3991                         BUG_ON(!cache);
3992                 }
3993
3994                 len = cache->key.objectid + cache->key.offset - start;
3995                 len = min(len, end + 1 - start);
3996
3997                 if (start < cache->last_byte_to_unpin) {
3998                         len = min(len, cache->last_byte_to_unpin - start);
3999                         btrfs_add_free_space(cache, start, len);
4000                 }
4001
4002                 start += len;
4003
4004                 spin_lock(&cache->space_info->lock);
4005                 spin_lock(&cache->lock);
4006                 cache->pinned -= len;
4007                 cache->space_info->bytes_pinned -= len;
4008                 if (cache->ro) {
4009                         cache->space_info->bytes_readonly += len;
4010                 } else if (cache->reserved_pinned > 0) {
4011                         len = min(len, cache->reserved_pinned);
4012                         cache->reserved_pinned -= len;
4013                         cache->space_info->bytes_reserved += len;
4014                 }
4015                 spin_unlock(&cache->lock);
4016                 spin_unlock(&cache->space_info->lock);
4017         }
4018
4019         if (cache)
4020                 btrfs_put_block_group(cache);
4021         return 0;
4022 }
4023
4024 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4025                                struct btrfs_root *root)
4026 {
4027         struct btrfs_fs_info *fs_info = root->fs_info;
4028         struct extent_io_tree *unpin;
4029         struct btrfs_block_rsv *block_rsv;
4030         struct btrfs_block_rsv *next_rsv;
4031         u64 start;
4032         u64 end;
4033         int idx;
4034         int ret;
4035
4036         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4037                 unpin = &fs_info->freed_extents[1];
4038         else
4039                 unpin = &fs_info->freed_extents[0];
4040
4041         while (1) {
4042                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4043                                             EXTENT_DIRTY);
4044                 if (ret)
4045                         break;
4046
4047                 ret = btrfs_discard_extent(root, start, end + 1 - start);
4048
4049                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4050                 unpin_extent_range(root, start, end);
4051                 cond_resched();
4052         }
4053
4054         mutex_lock(&fs_info->durable_block_rsv_mutex);
4055         list_for_each_entry_safe(block_rsv, next_rsv,
4056                                  &fs_info->durable_block_rsv_list, list) {
4057
4058                 idx = trans->transid & 0x1;
4059                 if (block_rsv->freed[idx] > 0) {
4060                         block_rsv_add_bytes(block_rsv,
4061                                             block_rsv->freed[idx], 0);
4062                         block_rsv->freed[idx] = 0;
4063                 }
4064                 if (atomic_read(&block_rsv->usage) == 0) {
4065                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4066
4067                         if (block_rsv->freed[0] == 0 &&
4068                             block_rsv->freed[1] == 0) {
4069                                 list_del_init(&block_rsv->list);
4070                                 kfree(block_rsv);
4071                         }
4072                 } else {
4073                         btrfs_block_rsv_release(root, block_rsv, 0);
4074                 }
4075         }
4076         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4077
4078         return 0;
4079 }
4080
4081 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4082                                 struct btrfs_root *root,
4083                                 u64 bytenr, u64 num_bytes, u64 parent,
4084                                 u64 root_objectid, u64 owner_objectid,
4085                                 u64 owner_offset, int refs_to_drop,
4086                                 struct btrfs_delayed_extent_op *extent_op)
4087 {
4088         struct btrfs_key key;
4089         struct btrfs_path *path;
4090         struct btrfs_fs_info *info = root->fs_info;
4091         struct btrfs_root *extent_root = info->extent_root;
4092         struct extent_buffer *leaf;
4093         struct btrfs_extent_item *ei;
4094         struct btrfs_extent_inline_ref *iref;
4095         int ret;
4096         int is_data;
4097         int extent_slot = 0;
4098         int found_extent = 0;
4099         int num_to_del = 1;
4100         u32 item_size;
4101         u64 refs;
4102
4103         path = btrfs_alloc_path();
4104         if (!path)
4105                 return -ENOMEM;
4106
4107         path->reada = 1;
4108         path->leave_spinning = 1;
4109
4110         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4111         BUG_ON(!is_data && refs_to_drop != 1);
4112
4113         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4114                                     bytenr, num_bytes, parent,
4115                                     root_objectid, owner_objectid,
4116                                     owner_offset);
4117         if (ret == 0) {
4118                 extent_slot = path->slots[0];
4119                 while (extent_slot >= 0) {
4120                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4121                                               extent_slot);
4122                         if (key.objectid != bytenr)
4123                                 break;
4124                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4125                             key.offset == num_bytes) {
4126                                 found_extent = 1;
4127                                 break;
4128                         }
4129                         if (path->slots[0] - extent_slot > 5)
4130                                 break;
4131                         extent_slot--;
4132                 }
4133 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4134                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4135                 if (found_extent && item_size < sizeof(*ei))
4136                         found_extent = 0;
4137 #endif
4138                 if (!found_extent) {
4139                         BUG_ON(iref);
4140                         ret = remove_extent_backref(trans, extent_root, path,
4141                                                     NULL, refs_to_drop,
4142                                                     is_data);
4143                         BUG_ON(ret);
4144                         btrfs_release_path(extent_root, path);
4145                         path->leave_spinning = 1;
4146
4147                         key.objectid = bytenr;
4148                         key.type = BTRFS_EXTENT_ITEM_KEY;
4149                         key.offset = num_bytes;
4150
4151                         ret = btrfs_search_slot(trans, extent_root,
4152                                                 &key, path, -1, 1);
4153                         if (ret) {
4154                                 printk(KERN_ERR "umm, got %d back from search"
4155                                        ", was looking for %llu\n", ret,
4156                                        (unsigned long long)bytenr);
4157                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4158                         }
4159                         BUG_ON(ret);
4160                         extent_slot = path->slots[0];
4161                 }
4162         } else {
4163                 btrfs_print_leaf(extent_root, path->nodes[0]);
4164                 WARN_ON(1);
4165                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4166                        "parent %llu root %llu  owner %llu offset %llu\n",
4167                        (unsigned long long)bytenr,
4168                        (unsigned long long)parent,
4169                        (unsigned long long)root_objectid,
4170                        (unsigned long long)owner_objectid,
4171                        (unsigned long long)owner_offset);
4172         }
4173
4174         leaf = path->nodes[0];
4175         item_size = btrfs_item_size_nr(leaf, extent_slot);
4176 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4177         if (item_size < sizeof(*ei)) {
4178                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4179                 ret = convert_extent_item_v0(trans, extent_root, path,
4180                                              owner_objectid, 0);
4181                 BUG_ON(ret < 0);
4182
4183                 btrfs_release_path(extent_root, path);
4184                 path->leave_spinning = 1;
4185
4186                 key.objectid = bytenr;
4187                 key.type = BTRFS_EXTENT_ITEM_KEY;
4188                 key.offset = num_bytes;
4189
4190                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4191                                         -1, 1);
4192                 if (ret) {
4193                         printk(KERN_ERR "umm, got %d back from search"
4194                                ", was looking for %llu\n", ret,
4195                                (unsigned long long)bytenr);
4196                         btrfs_print_leaf(extent_root, path->nodes[0]);
4197                 }
4198                 BUG_ON(ret);
4199                 extent_slot = path->slots[0];
4200                 leaf = path->nodes[0];
4201                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4202         }
4203 #endif
4204         BUG_ON(item_size < sizeof(*ei));
4205         ei = btrfs_item_ptr(leaf, extent_slot,
4206                             struct btrfs_extent_item);
4207         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4208                 struct btrfs_tree_block_info *bi;
4209                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4210                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4211                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4212         }
4213
4214         refs = btrfs_extent_refs(leaf, ei);
4215         BUG_ON(refs < refs_to_drop);
4216         refs -= refs_to_drop;
4217
4218         if (refs > 0) {
4219                 if (extent_op)
4220                         __run_delayed_extent_op(extent_op, leaf, ei);
4221                 /*
4222                  * In the case of inline back ref, reference count will
4223                  * be updated by remove_extent_backref
4224                  */
4225                 if (iref) {
4226                         BUG_ON(!found_extent);
4227                 } else {
4228                         btrfs_set_extent_refs(leaf, ei, refs);
4229                         btrfs_mark_buffer_dirty(leaf);
4230                 }
4231                 if (found_extent) {
4232                         ret = remove_extent_backref(trans, extent_root, path,
4233                                                     iref, refs_to_drop,
4234                                                     is_data);
4235                         BUG_ON(ret);
4236                 }
4237         } else {
4238                 if (found_extent) {
4239                         BUG_ON(is_data && refs_to_drop !=
4240                                extent_data_ref_count(root, path, iref));
4241                         if (iref) {
4242                                 BUG_ON(path->slots[0] != extent_slot);
4243                         } else {
4244                                 BUG_ON(path->slots[0] != extent_slot + 1);
4245                                 path->slots[0] = extent_slot;
4246                                 num_to_del = 2;
4247                         }
4248                 }
4249
4250                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4251                                       num_to_del);
4252                 BUG_ON(ret);
4253                 btrfs_release_path(extent_root, path);
4254
4255                 if (is_data) {
4256                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4257                         BUG_ON(ret);
4258                 } else {
4259                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4260                              bytenr >> PAGE_CACHE_SHIFT,
4261                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4262                 }
4263
4264                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4265                 BUG_ON(ret);
4266         }
4267         btrfs_free_path(path);
4268         return ret;
4269 }
4270
4271 /*
4272  * when we free an block, it is possible (and likely) that we free the last
4273  * delayed ref for that extent as well.  This searches the delayed ref tree for
4274  * a given extent, and if there are no other delayed refs to be processed, it
4275  * removes it from the tree.
4276  */
4277 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4278                                       struct btrfs_root *root, u64 bytenr)
4279 {
4280         struct btrfs_delayed_ref_head *head;
4281         struct btrfs_delayed_ref_root *delayed_refs;
4282         struct btrfs_delayed_ref_node *ref;
4283         struct rb_node *node;
4284         int ret = 0;
4285
4286         delayed_refs = &trans->transaction->delayed_refs;
4287         spin_lock(&delayed_refs->lock);
4288         head = btrfs_find_delayed_ref_head(trans, bytenr);
4289         if (!head)
4290                 goto out;
4291
4292         node = rb_prev(&head->node.rb_node);
4293         if (!node)
4294                 goto out;
4295
4296         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4297
4298         /* there are still entries for this ref, we can't drop it */
4299         if (ref->bytenr == bytenr)
4300                 goto out;
4301
4302         if (head->extent_op) {
4303                 if (!head->must_insert_reserved)
4304                         goto out;
4305                 kfree(head->extent_op);
4306                 head->extent_op = NULL;
4307         }
4308
4309         /*
4310          * waiting for the lock here would deadlock.  If someone else has it
4311          * locked they are already in the process of dropping it anyway
4312          */
4313         if (!mutex_trylock(&head->mutex))
4314                 goto out;
4315
4316         /*
4317          * at this point we have a head with no other entries.  Go
4318          * ahead and process it.
4319          */
4320         head->node.in_tree = 0;
4321         rb_erase(&head->node.rb_node, &delayed_refs->root);
4322
4323         delayed_refs->num_entries--;
4324
4325         /*
4326          * we don't take a ref on the node because we're removing it from the
4327          * tree, so we just steal the ref the tree was holding.
4328          */
4329         delayed_refs->num_heads--;
4330         if (list_empty(&head->cluster))
4331                 delayed_refs->num_heads_ready--;
4332
4333         list_del_init(&head->cluster);
4334         spin_unlock(&delayed_refs->lock);
4335
4336         BUG_ON(head->extent_op);
4337         if (head->must_insert_reserved)
4338                 ret = 1;
4339
4340         mutex_unlock(&head->mutex);
4341         btrfs_put_delayed_ref(&head->node);
4342         return ret;
4343 out:
4344         spin_unlock(&delayed_refs->lock);
4345         return 0;
4346 }
4347
4348 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4349                            struct btrfs_root *root,
4350                            struct extent_buffer *buf,
4351                            u64 parent, int last_ref)
4352 {
4353         struct btrfs_block_rsv *block_rsv;
4354         struct btrfs_block_group_cache *cache = NULL;
4355         int ret;
4356
4357         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4358                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4359                                                 parent, root->root_key.objectid,
4360                                                 btrfs_header_level(buf),
4361                                                 BTRFS_DROP_DELAYED_REF, NULL);
4362                 BUG_ON(ret);
4363         }
4364
4365         if (!last_ref)
4366                 return;
4367
4368         block_rsv = get_block_rsv(trans, root);
4369         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4370         if (block_rsv->space_info != cache->space_info)
4371                 goto out;
4372
4373         if (btrfs_header_generation(buf) == trans->transid) {
4374                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4375                         ret = check_ref_cleanup(trans, root, buf->start);
4376                         if (!ret)
4377                                 goto pin;
4378                 }
4379
4380                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4381                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4382                         goto pin;
4383                 }
4384
4385                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4386
4387                 btrfs_add_free_space(cache, buf->start, buf->len);
4388                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4389                 if (ret == -EAGAIN) {
4390                         /* block group became read-only */
4391                         update_reserved_bytes(cache, buf->len, 0, 1);
4392                         goto out;
4393                 }
4394
4395                 ret = 1;
4396                 spin_lock(&block_rsv->lock);
4397                 if (block_rsv->reserved < block_rsv->size) {
4398                         block_rsv->reserved += buf->len;
4399                         ret = 0;
4400                 }
4401                 spin_unlock(&block_rsv->lock);
4402
4403                 if (ret) {
4404                         spin_lock(&cache->space_info->lock);
4405                         cache->space_info->bytes_reserved -= buf->len;
4406                         spin_unlock(&cache->space_info->lock);
4407                 }
4408                 goto out;
4409         }
4410 pin:
4411         if (block_rsv->durable && !cache->ro) {
4412                 ret = 0;
4413                 spin_lock(&cache->lock);
4414                 if (!cache->ro) {
4415                         cache->reserved_pinned += buf->len;
4416                         ret = 1;
4417                 }
4418                 spin_unlock(&cache->lock);
4419
4420                 if (ret) {
4421                         spin_lock(&block_rsv->lock);
4422                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4423                         spin_unlock(&block_rsv->lock);
4424                 }
4425         }
4426 out:
4427         btrfs_put_block_group(cache);
4428 }
4429
4430 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4431                       struct btrfs_root *root,
4432                       u64 bytenr, u64 num_bytes, u64 parent,
4433                       u64 root_objectid, u64 owner, u64 offset)
4434 {
4435         int ret;
4436
4437         /*
4438          * tree log blocks never actually go into the extent allocation
4439          * tree, just update pinning info and exit early.
4440          */
4441         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4442                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4443                 /* unlocks the pinned mutex */
4444                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4445                 ret = 0;
4446         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4447                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4448                                         parent, root_objectid, (int)owner,
4449                                         BTRFS_DROP_DELAYED_REF, NULL);
4450                 BUG_ON(ret);
4451         } else {
4452                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4453                                         parent, root_objectid, owner,
4454                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4455                 BUG_ON(ret);
4456         }
4457         return ret;
4458 }
4459
4460 static u64 stripe_align(struct btrfs_root *root, u64 val)
4461 {
4462         u64 mask = ((u64)root->stripesize - 1);
4463         u64 ret = (val + mask) & ~mask;
4464         return ret;
4465 }
4466
4467 /*
4468  * when we wait for progress in the block group caching, its because
4469  * our allocation attempt failed at least once.  So, we must sleep
4470  * and let some progress happen before we try again.
4471  *
4472  * This function will sleep at least once waiting for new free space to
4473  * show up, and then it will check the block group free space numbers
4474  * for our min num_bytes.  Another option is to have it go ahead
4475  * and look in the rbtree for a free extent of a given size, but this
4476  * is a good start.
4477  */
4478 static noinline int
4479 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4480                                 u64 num_bytes)
4481 {
4482         struct btrfs_caching_control *caching_ctl;
4483         DEFINE_WAIT(wait);
4484
4485         caching_ctl = get_caching_control(cache);
4486         if (!caching_ctl)
4487                 return 0;
4488
4489         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4490                    (cache->free_space >= num_bytes));
4491
4492         put_caching_control(caching_ctl);
4493         return 0;
4494 }
4495
4496 static noinline int
4497 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4498 {
4499         struct btrfs_caching_control *caching_ctl;
4500         DEFINE_WAIT(wait);
4501
4502         caching_ctl = get_caching_control(cache);
4503         if (!caching_ctl)
4504                 return 0;
4505
4506         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4507
4508         put_caching_control(caching_ctl);
4509         return 0;
4510 }
4511
4512 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4513 {
4514         int index;
4515         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4516                 index = 0;
4517         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4518                 index = 1;
4519         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4520                 index = 2;
4521         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4522                 index = 3;
4523         else
4524                 index = 4;
4525         return index;
4526 }
4527
4528 enum btrfs_loop_type {
4529         LOOP_FIND_IDEAL = 0,
4530         LOOP_CACHING_NOWAIT = 1,
4531         LOOP_CACHING_WAIT = 2,
4532         LOOP_ALLOC_CHUNK = 3,
4533         LOOP_NO_EMPTY_SIZE = 4,
4534 };
4535
4536 /*
4537  * walks the btree of allocated extents and find a hole of a given size.
4538  * The key ins is changed to record the hole:
4539  * ins->objectid == block start
4540  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4541  * ins->offset == number of blocks
4542  * Any available blocks before search_start are skipped.
4543  */
4544 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4545                                      struct btrfs_root *orig_root,
4546                                      u64 num_bytes, u64 empty_size,
4547                                      u64 search_start, u64 search_end,
4548                                      u64 hint_byte, struct btrfs_key *ins,
4549                                      int data)
4550 {
4551         int ret = 0;
4552         struct btrfs_root *root = orig_root->fs_info->extent_root;
4553         struct btrfs_free_cluster *last_ptr = NULL;
4554         struct btrfs_block_group_cache *block_group = NULL;
4555         int empty_cluster = 2 * 1024 * 1024;
4556         int allowed_chunk_alloc = 0;
4557         int done_chunk_alloc = 0;
4558         struct btrfs_space_info *space_info;
4559         int last_ptr_loop = 0;
4560         int loop = 0;
4561         int index = 0;
4562         bool found_uncached_bg = false;
4563         bool failed_cluster_refill = false;
4564         bool failed_alloc = false;
4565         u64 ideal_cache_percent = 0;
4566         u64 ideal_cache_offset = 0;
4567
4568         WARN_ON(num_bytes < root->sectorsize);
4569         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4570         ins->objectid = 0;
4571         ins->offset = 0;
4572
4573         space_info = __find_space_info(root->fs_info, data);
4574         if (!space_info) {
4575                 printk(KERN_ERR "No space info for %d\n", data);
4576                 return -ENOSPC;
4577         }
4578
4579         if (orig_root->ref_cows || empty_size)
4580                 allowed_chunk_alloc = 1;
4581
4582         if (data & BTRFS_BLOCK_GROUP_METADATA) {
4583                 last_ptr = &root->fs_info->meta_alloc_cluster;
4584                 if (!btrfs_test_opt(root, SSD))
4585                         empty_cluster = 64 * 1024;
4586         }
4587
4588         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4589                 last_ptr = &root->fs_info->data_alloc_cluster;
4590         }
4591
4592         if (last_ptr) {
4593                 spin_lock(&last_ptr->lock);
4594                 if (last_ptr->block_group)
4595                         hint_byte = last_ptr->window_start;
4596                 spin_unlock(&last_ptr->lock);
4597         }
4598
4599         search_start = max(search_start, first_logical_byte(root, 0));
4600         search_start = max(search_start, hint_byte);
4601
4602         if (!last_ptr)
4603                 empty_cluster = 0;
4604
4605         if (search_start == hint_byte) {
4606 ideal_cache:
4607                 block_group = btrfs_lookup_block_group(root->fs_info,
4608                                                        search_start);
4609                 /*
4610                  * we don't want to use the block group if it doesn't match our
4611                  * allocation bits, or if its not cached.
4612                  *
4613                  * However if we are re-searching with an ideal block group
4614                  * picked out then we don't care that the block group is cached.
4615                  */
4616                 if (block_group && block_group_bits(block_group, data) &&
4617                     (block_group->cached != BTRFS_CACHE_NO ||
4618                      search_start == ideal_cache_offset)) {
4619                         down_read(&space_info->groups_sem);
4620                         if (list_empty(&block_group->list) ||
4621                             block_group->ro) {
4622                                 /*
4623                                  * someone is removing this block group,
4624                                  * we can't jump into the have_block_group
4625                                  * target because our list pointers are not
4626                                  * valid
4627                                  */
4628                                 btrfs_put_block_group(block_group);
4629                                 up_read(&space_info->groups_sem);
4630                         } else {
4631                                 index = get_block_group_index(block_group);
4632                                 goto have_block_group;
4633                         }
4634                 } else if (block_group) {
4635                         btrfs_put_block_group(block_group);
4636                 }
4637         }
4638 search:
4639         down_read(&space_info->groups_sem);
4640         list_for_each_entry(block_group, &space_info->block_groups[index],
4641                             list) {
4642                 u64 offset;
4643                 int cached;
4644
4645                 btrfs_get_block_group(block_group);
4646                 search_start = block_group->key.objectid;
4647
4648 have_block_group:
4649                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4650                         u64 free_percent;
4651
4652                         free_percent = btrfs_block_group_used(&block_group->item);
4653                         free_percent *= 100;
4654                         free_percent = div64_u64(free_percent,
4655                                                  block_group->key.offset);
4656                         free_percent = 100 - free_percent;
4657                         if (free_percent > ideal_cache_percent &&
4658                             likely(!block_group->ro)) {
4659                                 ideal_cache_offset = block_group->key.objectid;
4660                                 ideal_cache_percent = free_percent;
4661                         }
4662
4663                         /*
4664                          * We only want to start kthread caching if we are at
4665                          * the point where we will wait for caching to make
4666                          * progress, or if our ideal search is over and we've
4667                          * found somebody to start caching.
4668                          */
4669                         if (loop > LOOP_CACHING_NOWAIT ||
4670                             (loop > LOOP_FIND_IDEAL &&
4671                              atomic_read(&space_info->caching_threads) < 2)) {
4672                                 ret = cache_block_group(block_group);
4673                                 BUG_ON(ret);
4674                         }
4675                         found_uncached_bg = true;
4676
4677                         /*
4678                          * If loop is set for cached only, try the next block
4679                          * group.
4680                          */
4681                         if (loop == LOOP_FIND_IDEAL)
4682                                 goto loop;
4683                 }
4684
4685                 cached = block_group_cache_done(block_group);
4686                 if (unlikely(!cached))
4687                         found_uncached_bg = true;
4688
4689                 if (unlikely(block_group->ro))
4690                         goto loop;
4691
4692                 /*
4693                  * Ok we want to try and use the cluster allocator, so lets look
4694                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4695                  * have tried the cluster allocator plenty of times at this
4696                  * point and not have found anything, so we are likely way too
4697                  * fragmented for the clustering stuff to find anything, so lets
4698                  * just skip it and let the allocator find whatever block it can
4699                  * find
4700                  */
4701                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4702                         /*
4703                          * the refill lock keeps out other
4704                          * people trying to start a new cluster
4705                          */
4706                         spin_lock(&last_ptr->refill_lock);
4707                         if (last_ptr->block_group &&
4708                             (last_ptr->block_group->ro ||
4709                             !block_group_bits(last_ptr->block_group, data))) {
4710                                 offset = 0;
4711                                 goto refill_cluster;
4712                         }
4713
4714                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4715                                                  num_bytes, search_start);
4716                         if (offset) {
4717                                 /* we have a block, we're done */
4718                                 spin_unlock(&last_ptr->refill_lock);
4719                                 goto checks;
4720                         }
4721
4722                         spin_lock(&last_ptr->lock);
4723                         /*
4724                          * whoops, this cluster doesn't actually point to
4725                          * this block group.  Get a ref on the block
4726                          * group is does point to and try again
4727                          */
4728                         if (!last_ptr_loop && last_ptr->block_group &&
4729                             last_ptr->block_group != block_group) {
4730
4731                                 btrfs_put_block_group(block_group);
4732                                 block_group = last_ptr->block_group;
4733                                 btrfs_get_block_group(block_group);
4734                                 spin_unlock(&last_ptr->lock);
4735                                 spin_unlock(&last_ptr->refill_lock);
4736
4737                                 last_ptr_loop = 1;
4738                                 search_start = block_group->key.objectid;
4739                                 /*
4740                                  * we know this block group is properly
4741                                  * in the list because
4742                                  * btrfs_remove_block_group, drops the
4743                                  * cluster before it removes the block
4744                                  * group from the list
4745                                  */
4746                                 goto have_block_group;
4747                         }
4748                         spin_unlock(&last_ptr->lock);
4749 refill_cluster:
4750                         /*
4751                          * this cluster didn't work out, free it and
4752                          * start over
4753                          */
4754                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4755
4756                         last_ptr_loop = 0;
4757
4758                         /* allocate a cluster in this block group */
4759                         ret = btrfs_find_space_cluster(trans, root,
4760                                                block_group, last_ptr,
4761                                                offset, num_bytes,
4762                                                empty_cluster + empty_size);
4763                         if (ret == 0) {
4764                                 /*
4765                                  * now pull our allocation out of this
4766                                  * cluster
4767                                  */
4768                                 offset = btrfs_alloc_from_cluster(block_group,
4769                                                   last_ptr, num_bytes,
4770                                                   search_start);
4771                                 if (offset) {
4772                                         /* we found one, proceed */
4773                                         spin_unlock(&last_ptr->refill_lock);
4774                                         goto checks;
4775                                 }
4776                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
4777                                    && !failed_cluster_refill) {
4778                                 spin_unlock(&last_ptr->refill_lock);
4779
4780                                 failed_cluster_refill = true;
4781                                 wait_block_group_cache_progress(block_group,
4782                                        num_bytes + empty_cluster + empty_size);
4783                                 goto have_block_group;
4784                         }
4785
4786                         /*
4787                          * at this point we either didn't find a cluster
4788                          * or we weren't able to allocate a block from our
4789                          * cluster.  Free the cluster we've been trying
4790                          * to use, and go to the next block group
4791                          */
4792                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4793                         spin_unlock(&last_ptr->refill_lock);
4794                         goto loop;
4795                 }
4796
4797                 offset = btrfs_find_space_for_alloc(block_group, search_start,
4798                                                     num_bytes, empty_size);
4799                 /*
4800                  * If we didn't find a chunk, and we haven't failed on this
4801                  * block group before, and this block group is in the middle of
4802                  * caching and we are ok with waiting, then go ahead and wait
4803                  * for progress to be made, and set failed_alloc to true.
4804                  *
4805                  * If failed_alloc is true then we've already waited on this
4806                  * block group once and should move on to the next block group.
4807                  */
4808                 if (!offset && !failed_alloc && !cached &&
4809                     loop > LOOP_CACHING_NOWAIT) {
4810                         wait_block_group_cache_progress(block_group,
4811                                                 num_bytes + empty_size);
4812                         failed_alloc = true;
4813                         goto have_block_group;
4814                 } else if (!offset) {
4815                         goto loop;
4816                 }
4817 checks:
4818                 search_start = stripe_align(root, offset);
4819                 /* move on to the next group */
4820                 if (search_start + num_bytes >= search_end) {
4821                         btrfs_add_free_space(block_group, offset, num_bytes);
4822                         goto loop;
4823                 }
4824
4825                 /* move on to the next group */
4826                 if (search_start + num_bytes >
4827                     block_group->key.objectid + block_group->key.offset) {
4828                         btrfs_add_free_space(block_group, offset, num_bytes);
4829                         goto loop;
4830                 }
4831
4832                 ins->objectid = search_start;
4833                 ins->offset = num_bytes;
4834
4835                 if (offset < search_start)
4836                         btrfs_add_free_space(block_group, offset,
4837                                              search_start - offset);
4838                 BUG_ON(offset > search_start);
4839
4840                 ret = update_reserved_bytes(block_group, num_bytes, 1,
4841                                             (data & BTRFS_BLOCK_GROUP_DATA));
4842                 if (ret == -EAGAIN) {
4843                         btrfs_add_free_space(block_group, offset, num_bytes);
4844                         goto loop;
4845                 }
4846
4847                 /* we are all good, lets return */
4848                 ins->objectid = search_start;
4849                 ins->offset = num_bytes;
4850
4851                 if (offset < search_start)
4852                         btrfs_add_free_space(block_group, offset,
4853                                              search_start - offset);
4854                 BUG_ON(offset > search_start);
4855                 break;
4856 loop:
4857                 failed_cluster_refill = false;
4858                 failed_alloc = false;
4859                 BUG_ON(index != get_block_group_index(block_group));
4860                 btrfs_put_block_group(block_group);
4861         }
4862         up_read(&space_info->groups_sem);
4863
4864         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
4865                 goto search;
4866
4867         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
4868          *                      for them to make caching progress.  Also
4869          *                      determine the best possible bg to cache
4870          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4871          *                      caching kthreads as we move along
4872          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4873          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4874          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4875          *                      again
4876          */
4877         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4878             (found_uncached_bg || empty_size || empty_cluster ||
4879              allowed_chunk_alloc)) {
4880                 index = 0;
4881                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
4882                         found_uncached_bg = false;
4883                         loop++;
4884                         if (!ideal_cache_percent &&
4885                             atomic_read(&space_info->caching_threads))
4886                                 goto search;
4887
4888                         /*
4889                          * 1 of the following 2 things have happened so far
4890                          *
4891                          * 1) We found an ideal block group for caching that
4892                          * is mostly full and will cache quickly, so we might
4893                          * as well wait for it.
4894                          *
4895                          * 2) We searched for cached only and we didn't find
4896                          * anything, and we didn't start any caching kthreads
4897                          * either, so chances are we will loop through and
4898                          * start a couple caching kthreads, and then come back
4899                          * around and just wait for them.  This will be slower
4900                          * because we will have 2 caching kthreads reading at
4901                          * the same time when we could have just started one
4902                          * and waited for it to get far enough to give us an
4903                          * allocation, so go ahead and go to the wait caching
4904                          * loop.
4905                          */
4906                         loop = LOOP_CACHING_WAIT;
4907                         search_start = ideal_cache_offset;
4908                         ideal_cache_percent = 0;
4909                         goto ideal_cache;
4910                 } else if (loop == LOOP_FIND_IDEAL) {
4911                         /*
4912                          * Didn't find a uncached bg, wait on anything we find
4913                          * next.
4914                          */
4915                         loop = LOOP_CACHING_WAIT;
4916                         goto search;
4917                 }
4918
4919                 if (loop < LOOP_CACHING_WAIT) {
4920                         loop++;
4921                         goto search;
4922                 }
4923
4924                 if (loop == LOOP_ALLOC_CHUNK) {
4925                         empty_size = 0;
4926                         empty_cluster = 0;
4927                 }
4928
4929                 if (allowed_chunk_alloc) {
4930                         ret = do_chunk_alloc(trans, root, num_bytes +
4931                                              2 * 1024 * 1024, data, 1);
4932                         allowed_chunk_alloc = 0;
4933                         done_chunk_alloc = 1;
4934                 } else if (!done_chunk_alloc) {
4935                         space_info->force_alloc = 1;
4936                 }
4937
4938                 if (loop < LOOP_NO_EMPTY_SIZE) {
4939                         loop++;
4940                         goto search;
4941                 }
4942                 ret = -ENOSPC;
4943         } else if (!ins->objectid) {
4944                 ret = -ENOSPC;
4945         }
4946
4947         /* we found what we needed */
4948         if (ins->objectid) {
4949                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4950                         trans->block_group = block_group->key.objectid;
4951
4952                 btrfs_put_block_group(block_group);
4953                 ret = 0;
4954         }
4955
4956         return ret;
4957 }
4958
4959 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4960                             int dump_block_groups)
4961 {
4962         struct btrfs_block_group_cache *cache;
4963         int index = 0;
4964
4965         spin_lock(&info->lock);
4966         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4967                (unsigned long long)(info->total_bytes - info->bytes_used -
4968                                     info->bytes_pinned - info->bytes_reserved -
4969                                     info->bytes_readonly),
4970                (info->full) ? "" : "not ");
4971         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
4972                "reserved=%llu, may_use=%llu, readonly=%llu\n",
4973                (unsigned long long)info->total_bytes,
4974                (unsigned long long)info->bytes_used,
4975                (unsigned long long)info->bytes_pinned,
4976                (unsigned long long)info->bytes_reserved,
4977                (unsigned long long)info->bytes_may_use,
4978                (unsigned long long)info->bytes_readonly);
4979         spin_unlock(&info->lock);
4980
4981         if (!dump_block_groups)
4982                 return;
4983
4984         down_read(&info->groups_sem);
4985 again:
4986         list_for_each_entry(cache, &info->block_groups[index], list) {
4987                 spin_lock(&cache->lock);
4988                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4989                        "%llu pinned %llu reserved\n",
4990                        (unsigned long long)cache->key.objectid,
4991                        (unsigned long long)cache->key.offset,
4992                        (unsigned long long)btrfs_block_group_used(&cache->item),
4993                        (unsigned long long)cache->pinned,
4994                        (unsigned long long)cache->reserved);
4995                 btrfs_dump_free_space(cache, bytes);
4996                 spin_unlock(&cache->lock);
4997         }
4998         if (++index < BTRFS_NR_RAID_TYPES)
4999                 goto again;
5000         up_read(&info->groups_sem);
5001 }
5002
5003 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5004                          struct btrfs_root *root,
5005                          u64 num_bytes, u64 min_alloc_size,
5006                          u64 empty_size, u64 hint_byte,
5007                          u64 search_end, struct btrfs_key *ins,
5008                          u64 data)
5009 {
5010         int ret;
5011         u64 search_start = 0;
5012
5013         data = btrfs_get_alloc_profile(root, data);
5014 again:
5015         /*
5016          * the only place that sets empty_size is btrfs_realloc_node, which
5017          * is not called recursively on allocations
5018          */
5019         if (empty_size || root->ref_cows)
5020                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5021                                      num_bytes + 2 * 1024 * 1024, data, 0);
5022
5023         WARN_ON(num_bytes < root->sectorsize);
5024         ret = find_free_extent(trans, root, num_bytes, empty_size,
5025                                search_start, search_end, hint_byte,
5026                                ins, data);
5027
5028         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5029                 num_bytes = num_bytes >> 1;
5030                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5031                 num_bytes = max(num_bytes, min_alloc_size);
5032                 do_chunk_alloc(trans, root->fs_info->extent_root,
5033                                num_bytes, data, 1);
5034                 goto again;
5035         }
5036         if (ret == -ENOSPC) {
5037                 struct btrfs_space_info *sinfo;
5038
5039                 sinfo = __find_space_info(root->fs_info, data);
5040                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5041                        "wanted %llu\n", (unsigned long long)data,
5042                        (unsigned long long)num_bytes);
5043                 dump_space_info(sinfo, num_bytes, 1);
5044         }
5045
5046         return ret;
5047 }
5048
5049 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5050 {
5051         struct btrfs_block_group_cache *cache;
5052         int ret = 0;
5053
5054         cache = btrfs_lookup_block_group(root->fs_info, start);
5055         if (!cache) {
5056                 printk(KERN_ERR "Unable to find block group for %llu\n",
5057                        (unsigned long long)start);
5058                 return -ENOSPC;
5059         }
5060
5061         ret = btrfs_discard_extent(root, start, len);
5062
5063         btrfs_add_free_space(cache, start, len);
5064         update_reserved_bytes(cache, len, 0, 1);
5065         btrfs_put_block_group(cache);
5066
5067         return ret;
5068 }
5069
5070 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5071                                       struct btrfs_root *root,
5072                                       u64 parent, u64 root_objectid,
5073                                       u64 flags, u64 owner, u64 offset,
5074                                       struct btrfs_key *ins, int ref_mod)
5075 {
5076         int ret;
5077         struct btrfs_fs_info *fs_info = root->fs_info;
5078         struct btrfs_extent_item *extent_item;
5079         struct btrfs_extent_inline_ref *iref;
5080         struct btrfs_path *path;
5081         struct extent_buffer *leaf;
5082         int type;
5083         u32 size;
5084
5085         if (parent > 0)
5086                 type = BTRFS_SHARED_DATA_REF_KEY;
5087         else
5088                 type = BTRFS_EXTENT_DATA_REF_KEY;
5089
5090         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5091
5092         path = btrfs_alloc_path();
5093         BUG_ON(!path);
5094
5095         path->leave_spinning = 1;
5096         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5097                                       ins, size);
5098         BUG_ON(ret);
5099
5100         leaf = path->nodes[0];
5101         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5102                                      struct btrfs_extent_item);
5103         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5104         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5105         btrfs_set_extent_flags(leaf, extent_item,
5106                                flags | BTRFS_EXTENT_FLAG_DATA);
5107
5108         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5109         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5110         if (parent > 0) {
5111                 struct btrfs_shared_data_ref *ref;
5112                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5113                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5114                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5115         } else {
5116                 struct btrfs_extent_data_ref *ref;
5117                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5118                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5119                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5120                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5121                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5122         }
5123
5124         btrfs_mark_buffer_dirty(path->nodes[0]);
5125         btrfs_free_path(path);
5126
5127         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5128         if (ret) {
5129                 printk(KERN_ERR "btrfs update block group failed for %llu "
5130                        "%llu\n", (unsigned long long)ins->objectid,
5131                        (unsigned long long)ins->offset);
5132                 BUG();
5133         }
5134         return ret;
5135 }
5136
5137 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5138                                      struct btrfs_root *root,
5139                                      u64 parent, u64 root_objectid,
5140                                      u64 flags, struct btrfs_disk_key *key,
5141                                      int level, struct btrfs_key *ins)
5142 {
5143         int ret;
5144         struct btrfs_fs_info *fs_info = root->fs_info;
5145         struct btrfs_extent_item *extent_item;
5146         struct btrfs_tree_block_info *block_info;
5147         struct btrfs_extent_inline_ref *iref;
5148         struct btrfs_path *path;
5149         struct extent_buffer *leaf;
5150         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5151
5152         path = btrfs_alloc_path();
5153         BUG_ON(!path);
5154
5155         path->leave_spinning = 1;
5156         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5157                                       ins, size);
5158         BUG_ON(ret);
5159
5160         leaf = path->nodes[0];
5161         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5162                                      struct btrfs_extent_item);
5163         btrfs_set_extent_refs(leaf, extent_item, 1);
5164         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5165         btrfs_set_extent_flags(leaf, extent_item,
5166                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5167         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5168
5169         btrfs_set_tree_block_key(leaf, block_info, key);
5170         btrfs_set_tree_block_level(leaf, block_info, level);
5171
5172         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5173         if (parent > 0) {
5174                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5175                 btrfs_set_extent_inline_ref_type(leaf, iref,
5176                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5177                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5178         } else {
5179                 btrfs_set_extent_inline_ref_type(leaf, iref,
5180                                                  BTRFS_TREE_BLOCK_REF_KEY);
5181                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5182         }
5183
5184         btrfs_mark_buffer_dirty(leaf);
5185         btrfs_free_path(path);
5186
5187         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5188         if (ret) {
5189                 printk(KERN_ERR "btrfs update block group failed for %llu "
5190                        "%llu\n", (unsigned long long)ins->objectid,
5191                        (unsigned long long)ins->offset);
5192                 BUG();
5193         }
5194         return ret;
5195 }
5196
5197 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5198                                      struct btrfs_root *root,
5199                                      u64 root_objectid, u64 owner,
5200                                      u64 offset, struct btrfs_key *ins)
5201 {
5202         int ret;
5203
5204         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5205
5206         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5207                                          0, root_objectid, owner, offset,
5208                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5209         return ret;
5210 }
5211
5212 /*
5213  * this is used by the tree logging recovery code.  It records that
5214  * an extent has been allocated and makes sure to clear the free
5215  * space cache bits as well
5216  */
5217 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5218                                    struct btrfs_root *root,
5219                                    u64 root_objectid, u64 owner, u64 offset,
5220                                    struct btrfs_key *ins)
5221 {
5222         int ret;
5223         struct btrfs_block_group_cache *block_group;
5224         struct btrfs_caching_control *caching_ctl;
5225         u64 start = ins->objectid;
5226         u64 num_bytes = ins->offset;
5227
5228         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5229         cache_block_group(block_group);
5230         caching_ctl = get_caching_control(block_group);
5231
5232         if (!caching_ctl) {
5233                 BUG_ON(!block_group_cache_done(block_group));
5234                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5235                 BUG_ON(ret);
5236         } else {
5237                 mutex_lock(&caching_ctl->mutex);
5238
5239                 if (start >= caching_ctl->progress) {
5240                         ret = add_excluded_extent(root, start, num_bytes);
5241                         BUG_ON(ret);
5242                 } else if (start + num_bytes <= caching_ctl->progress) {
5243                         ret = btrfs_remove_free_space(block_group,
5244                                                       start, num_bytes);
5245                         BUG_ON(ret);
5246                 } else {
5247                         num_bytes = caching_ctl->progress - start;
5248                         ret = btrfs_remove_free_space(block_group,
5249                                                       start, num_bytes);
5250                         BUG_ON(ret);
5251
5252                         start = caching_ctl->progress;
5253                         num_bytes = ins->objectid + ins->offset -
5254                                     caching_ctl->progress;
5255                         ret = add_excluded_extent(root, start, num_bytes);
5256                         BUG_ON(ret);
5257                 }
5258
5259                 mutex_unlock(&caching_ctl->mutex);
5260                 put_caching_control(caching_ctl);
5261         }
5262
5263         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5264         BUG_ON(ret);
5265         btrfs_put_block_group(block_group);
5266         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5267                                          0, owner, offset, ins, 1);
5268         return ret;
5269 }
5270
5271 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5272                                             struct btrfs_root *root,
5273                                             u64 bytenr, u32 blocksize,
5274                                             int level)
5275 {
5276         struct extent_buffer *buf;
5277
5278         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5279         if (!buf)
5280                 return ERR_PTR(-ENOMEM);
5281         btrfs_set_header_generation(buf, trans->transid);
5282         btrfs_set_buffer_lockdep_class(buf, level);
5283         btrfs_tree_lock(buf);
5284         clean_tree_block(trans, root, buf);
5285
5286         btrfs_set_lock_blocking(buf);
5287         btrfs_set_buffer_uptodate(buf);
5288
5289         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5290                 /*
5291                  * we allow two log transactions at a time, use different
5292                  * EXENT bit to differentiate dirty pages.
5293                  */
5294                 if (root->log_transid % 2 == 0)
5295                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5296                                         buf->start + buf->len - 1, GFP_NOFS);
5297                 else
5298                         set_extent_new(&root->dirty_log_pages, buf->start,
5299                                         buf->start + buf->len - 1, GFP_NOFS);
5300         } else {
5301                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5302                          buf->start + buf->len - 1, GFP_NOFS);
5303         }
5304         trans->blocks_used++;
5305         /* this returns a buffer locked for blocking */
5306         return buf;
5307 }
5308
5309 static struct btrfs_block_rsv *
5310 use_block_rsv(struct btrfs_trans_handle *trans,
5311               struct btrfs_root *root, u32 blocksize)
5312 {
5313         struct btrfs_block_rsv *block_rsv;
5314         int ret;
5315
5316         block_rsv = get_block_rsv(trans, root);
5317
5318         if (block_rsv->size == 0) {
5319                 ret = reserve_metadata_bytes(block_rsv, blocksize);
5320                 if (ret)
5321                         return ERR_PTR(ret);
5322                 return block_rsv;
5323         }
5324
5325         ret = block_rsv_use_bytes(block_rsv, blocksize);
5326         if (!ret)
5327                 return block_rsv;
5328
5329         WARN_ON(1);
5330         printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
5331                 block_rsv->size, block_rsv->reserved,
5332                 block_rsv->freed[0], block_rsv->freed[1]);
5333
5334         return ERR_PTR(-ENOSPC);
5335 }
5336
5337 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5338 {
5339         block_rsv_add_bytes(block_rsv, blocksize, 0);
5340         block_rsv_release_bytes(block_rsv, NULL, 0);
5341 }
5342
5343 /*
5344  * finds a free extent and does all the dirty work required for allocation
5345  * returns the key for the extent through ins, and a tree buffer for
5346  * the first block of the extent through buf.
5347  *
5348  * returns the tree buffer or NULL.
5349  */
5350 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5351                                         struct btrfs_root *root, u32 blocksize,
5352                                         u64 parent, u64 root_objectid,
5353                                         struct btrfs_disk_key *key, int level,
5354                                         u64 hint, u64 empty_size)
5355 {
5356         struct btrfs_key ins;
5357         struct btrfs_block_rsv *block_rsv;
5358         struct extent_buffer *buf;
5359         u64 flags = 0;
5360         int ret;
5361
5362
5363         block_rsv = use_block_rsv(trans, root, blocksize);
5364         if (IS_ERR(block_rsv))
5365                 return ERR_CAST(block_rsv);
5366
5367         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5368                                    empty_size, hint, (u64)-1, &ins, 0);
5369         if (ret) {
5370                 unuse_block_rsv(block_rsv, blocksize);
5371                 return ERR_PTR(ret);
5372         }
5373
5374         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5375                                     blocksize, level);
5376         BUG_ON(IS_ERR(buf));
5377
5378         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5379                 if (parent == 0)
5380                         parent = ins.objectid;
5381                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5382         } else
5383                 BUG_ON(parent > 0);
5384
5385         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5386                 struct btrfs_delayed_extent_op *extent_op;
5387                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5388                 BUG_ON(!extent_op);
5389                 if (key)
5390                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5391                 else
5392                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5393                 extent_op->flags_to_set = flags;
5394                 extent_op->update_key = 1;
5395                 extent_op->update_flags = 1;
5396                 extent_op->is_data = 0;
5397
5398                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5399                                         ins.offset, parent, root_objectid,
5400                                         level, BTRFS_ADD_DELAYED_EXTENT,
5401                                         extent_op);
5402                 BUG_ON(ret);
5403         }
5404         return buf;
5405 }
5406
5407 struct walk_control {
5408         u64 refs[BTRFS_MAX_LEVEL];
5409         u64 flags[BTRFS_MAX_LEVEL];
5410         struct btrfs_key update_progress;
5411         int stage;
5412         int level;
5413         int shared_level;
5414         int update_ref;
5415         int keep_locks;
5416         int reada_slot;
5417         int reada_count;
5418 };
5419
5420 #define DROP_REFERENCE  1
5421 #define UPDATE_BACKREF  2
5422
5423 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5424                                      struct btrfs_root *root,
5425                                      struct walk_control *wc,
5426                                      struct btrfs_path *path)
5427 {
5428         u64 bytenr;
5429         u64 generation;
5430         u64 refs;
5431         u64 flags;
5432         u64 last = 0;
5433         u32 nritems;
5434         u32 blocksize;
5435         struct btrfs_key key;
5436         struct extent_buffer *eb;
5437         int ret;
5438         int slot;
5439         int nread = 0;
5440
5441         if (path->slots[wc->level] < wc->reada_slot) {
5442                 wc->reada_count = wc->reada_count * 2 / 3;
5443                 wc->reada_count = max(wc->reada_count, 2);
5444         } else {
5445                 wc->reada_count = wc->reada_count * 3 / 2;
5446                 wc->reada_count = min_t(int, wc->reada_count,
5447                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5448         }
5449
5450         eb = path->nodes[wc->level];
5451         nritems = btrfs_header_nritems(eb);
5452         blocksize = btrfs_level_size(root, wc->level - 1);
5453
5454         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5455                 if (nread >= wc->reada_count)
5456                         break;
5457
5458                 cond_resched();
5459                 bytenr = btrfs_node_blockptr(eb, slot);
5460                 generation = btrfs_node_ptr_generation(eb, slot);
5461
5462                 if (slot == path->slots[wc->level])
5463                         goto reada;
5464
5465                 if (wc->stage == UPDATE_BACKREF &&
5466                     generation <= root->root_key.offset)
5467                         continue;
5468
5469                 /* We don't lock the tree block, it's OK to be racy here */
5470                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5471                                                &refs, &flags);
5472                 BUG_ON(ret);
5473                 BUG_ON(refs == 0);
5474
5475                 if (wc->stage == DROP_REFERENCE) {
5476                         if (refs == 1)
5477                                 goto reada;
5478
5479                         if (wc->level == 1 &&
5480                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5481                                 continue;
5482                         if (!wc->update_ref ||
5483                             generation <= root->root_key.offset)
5484                                 continue;
5485                         btrfs_node_key_to_cpu(eb, &key, slot);
5486                         ret = btrfs_comp_cpu_keys(&key,
5487                                                   &wc->update_progress);
5488                         if (ret < 0)
5489                                 continue;
5490                 } else {
5491                         if (wc->level == 1 &&
5492                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5493                                 continue;
5494                 }
5495 reada:
5496                 ret = readahead_tree_block(root, bytenr, blocksize,
5497                                            generation);
5498                 if (ret)
5499                         break;
5500                 last = bytenr + blocksize;
5501                 nread++;
5502         }
5503         wc->reada_slot = slot;
5504 }
5505
5506 /*
5507  * hepler to process tree block while walking down the tree.
5508  *
5509  * when wc->stage == UPDATE_BACKREF, this function updates
5510  * back refs for pointers in the block.
5511  *
5512  * NOTE: return value 1 means we should stop walking down.
5513  */
5514 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5515                                    struct btrfs_root *root,
5516                                    struct btrfs_path *path,
5517                                    struct walk_control *wc, int lookup_info)
5518 {
5519         int level = wc->level;
5520         struct extent_buffer *eb = path->nodes[level];
5521         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5522         int ret;
5523
5524         if (wc->stage == UPDATE_BACKREF &&
5525             btrfs_header_owner(eb) != root->root_key.objectid)
5526                 return 1;
5527
5528         /*
5529          * when reference count of tree block is 1, it won't increase
5530          * again. once full backref flag is set, we never clear it.
5531          */
5532         if (lookup_info &&
5533             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5534              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5535                 BUG_ON(!path->locks[level]);
5536                 ret = btrfs_lookup_extent_info(trans, root,
5537                                                eb->start, eb->len,
5538                                                &wc->refs[level],
5539                                                &wc->flags[level]);
5540                 BUG_ON(ret);
5541                 BUG_ON(wc->refs[level] == 0);
5542         }
5543
5544         if (wc->stage == DROP_REFERENCE) {
5545                 if (wc->refs[level] > 1)
5546                         return 1;
5547
5548                 if (path->locks[level] && !wc->keep_locks) {
5549                         btrfs_tree_unlock(eb);
5550                         path->locks[level] = 0;
5551                 }
5552                 return 0;
5553         }
5554
5555         /* wc->stage == UPDATE_BACKREF */
5556         if (!(wc->flags[level] & flag)) {
5557                 BUG_ON(!path->locks[level]);
5558                 ret = btrfs_inc_ref(trans, root, eb, 1);
5559                 BUG_ON(ret);
5560                 ret = btrfs_dec_ref(trans, root, eb, 0);
5561                 BUG_ON(ret);
5562                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5563                                                   eb->len, flag, 0);
5564                 BUG_ON(ret);
5565                 wc->flags[level] |= flag;
5566         }
5567
5568         /*
5569          * the block is shared by multiple trees, so it's not good to
5570          * keep the tree lock
5571          */
5572         if (path->locks[level] && level > 0) {
5573                 btrfs_tree_unlock(eb);
5574                 path->locks[level] = 0;
5575         }
5576         return 0;
5577 }
5578
5579 /*
5580  * hepler to process tree block pointer.
5581  *
5582  * when wc->stage == DROP_REFERENCE, this function checks
5583  * reference count of the block pointed to. if the block
5584  * is shared and we need update back refs for the subtree
5585  * rooted at the block, this function changes wc->stage to
5586  * UPDATE_BACKREF. if the block is shared and there is no
5587  * need to update back, this function drops the reference
5588  * to the block.
5589  *
5590  * NOTE: return value 1 means we should stop walking down.
5591  */
5592 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5593                                  struct btrfs_root *root,
5594                                  struct btrfs_path *path,
5595                                  struct walk_control *wc, int *lookup_info)
5596 {
5597         u64 bytenr;
5598         u64 generation;
5599         u64 parent;
5600         u32 blocksize;
5601         struct btrfs_key key;
5602         struct extent_buffer *next;
5603         int level = wc->level;
5604         int reada = 0;
5605         int ret = 0;
5606
5607         generation = btrfs_node_ptr_generation(path->nodes[level],
5608                                                path->slots[level]);
5609         /*
5610          * if the lower level block was created before the snapshot
5611          * was created, we know there is no need to update back refs
5612          * for the subtree
5613          */
5614         if (wc->stage == UPDATE_BACKREF &&
5615             generation <= root->root_key.offset) {
5616                 *lookup_info = 1;
5617                 return 1;
5618         }
5619
5620         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5621         blocksize = btrfs_level_size(root, level - 1);
5622
5623         next = btrfs_find_tree_block(root, bytenr, blocksize);
5624         if (!next) {
5625                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5626                 if (!next)
5627                         return -ENOMEM;
5628                 reada = 1;
5629         }
5630         btrfs_tree_lock(next);
5631         btrfs_set_lock_blocking(next);
5632
5633         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5634                                        &wc->refs[level - 1],
5635                                        &wc->flags[level - 1]);
5636         BUG_ON(ret);
5637         BUG_ON(wc->refs[level - 1] == 0);
5638         *lookup_info = 0;
5639
5640         if (wc->stage == DROP_REFERENCE) {
5641                 if (wc->refs[level - 1] > 1) {
5642                         if (level == 1 &&
5643                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5644                                 goto skip;
5645
5646                         if (!wc->update_ref ||
5647                             generation <= root->root_key.offset)
5648                                 goto skip;
5649
5650                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5651                                               path->slots[level]);
5652                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5653                         if (ret < 0)
5654                                 goto skip;
5655
5656                         wc->stage = UPDATE_BACKREF;
5657                         wc->shared_level = level - 1;
5658                 }
5659         } else {
5660                 if (level == 1 &&
5661                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5662                         goto skip;
5663         }
5664
5665         if (!btrfs_buffer_uptodate(next, generation)) {
5666                 btrfs_tree_unlock(next);
5667                 free_extent_buffer(next);
5668                 next = NULL;
5669                 *lookup_info = 1;
5670         }
5671
5672         if (!next) {
5673                 if (reada && level == 1)
5674                         reada_walk_down(trans, root, wc, path);
5675                 next = read_tree_block(root, bytenr, blocksize, generation);
5676                 btrfs_tree_lock(next);
5677                 btrfs_set_lock_blocking(next);
5678         }
5679
5680         level--;
5681         BUG_ON(level != btrfs_header_level(next));
5682         path->nodes[level] = next;
5683         path->slots[level] = 0;
5684         path->locks[level] = 1;
5685         wc->level = level;
5686         if (wc->level == 1)
5687                 wc->reada_slot = 0;
5688         return 0;
5689 skip:
5690         wc->refs[level - 1] = 0;
5691         wc->flags[level - 1] = 0;
5692         if (wc->stage == DROP_REFERENCE) {
5693                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5694                         parent = path->nodes[level]->start;
5695                 } else {
5696                         BUG_ON(root->root_key.objectid !=
5697                                btrfs_header_owner(path->nodes[level]));
5698                         parent = 0;
5699                 }
5700
5701                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5702                                         root->root_key.objectid, level - 1, 0);
5703                 BUG_ON(ret);
5704         }
5705         btrfs_tree_unlock(next);
5706         free_extent_buffer(next);
5707         *lookup_info = 1;
5708         return 1;
5709 }
5710
5711 /*
5712  * hepler to process tree block while walking up the tree.
5713  *
5714  * when wc->stage == DROP_REFERENCE, this function drops
5715  * reference count on the block.
5716  *
5717  * when wc->stage == UPDATE_BACKREF, this function changes
5718  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5719  * to UPDATE_BACKREF previously while processing the block.
5720  *
5721  * NOTE: return value 1 means we should stop walking up.
5722  */
5723 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5724                                  struct btrfs_root *root,
5725                                  struct btrfs_path *path,
5726                                  struct walk_control *wc)
5727 {
5728         int ret;
5729         int level = wc->level;
5730         struct extent_buffer *eb = path->nodes[level];
5731         u64 parent = 0;
5732
5733         if (wc->stage == UPDATE_BACKREF) {
5734                 BUG_ON(wc->shared_level < level);
5735                 if (level < wc->shared_level)
5736                         goto out;
5737
5738                 ret = find_next_key(path, level + 1, &wc->update_progress);
5739                 if (ret > 0)
5740                         wc->update_ref = 0;
5741
5742                 wc->stage = DROP_REFERENCE;
5743                 wc->shared_level = -1;
5744                 path->slots[level] = 0;
5745
5746                 /*
5747                  * check reference count again if the block isn't locked.
5748                  * we should start walking down the tree again if reference
5749                  * count is one.
5750                  */
5751                 if (!path->locks[level]) {
5752                         BUG_ON(level == 0);
5753                         btrfs_tree_lock(eb);
5754                         btrfs_set_lock_blocking(eb);
5755                         path->locks[level] = 1;
5756
5757                         ret = btrfs_lookup_extent_info(trans, root,
5758                                                        eb->start, eb->len,
5759                                                        &wc->refs[level],
5760                                                        &wc->flags[level]);
5761                         BUG_ON(ret);
5762                         BUG_ON(wc->refs[level] == 0);
5763                         if (wc->refs[level] == 1) {
5764                                 btrfs_tree_unlock(eb);
5765                                 path->locks[level] = 0;
5766                                 return 1;
5767                         }
5768                 }
5769         }
5770
5771         /* wc->stage == DROP_REFERENCE */
5772         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5773
5774         if (wc->refs[level] == 1) {
5775                 if (level == 0) {
5776                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5777                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5778                         else
5779                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5780                         BUG_ON(ret);
5781                 }
5782                 /* make block locked assertion in clean_tree_block happy */
5783                 if (!path->locks[level] &&
5784                     btrfs_header_generation(eb) == trans->transid) {
5785                         btrfs_tree_lock(eb);
5786                         btrfs_set_lock_blocking(eb);
5787                         path->locks[level] = 1;
5788                 }
5789                 clean_tree_block(trans, root, eb);
5790         }
5791
5792         if (eb == root->node) {
5793                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5794                         parent = eb->start;
5795                 else
5796                         BUG_ON(root->root_key.objectid !=
5797                                btrfs_header_owner(eb));
5798         } else {
5799                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5800                         parent = path->nodes[level + 1]->start;
5801                 else
5802                         BUG_ON(root->root_key.objectid !=
5803                                btrfs_header_owner(path->nodes[level + 1]));
5804         }
5805
5806         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
5807 out:
5808         wc->refs[level] = 0;
5809         wc->flags[level] = 0;
5810         return 0;
5811 }
5812
5813 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5814                                    struct btrfs_root *root,
5815                                    struct btrfs_path *path,
5816                                    struct walk_control *wc)
5817 {
5818         int level = wc->level;
5819         int lookup_info = 1;
5820         int ret;
5821
5822         while (level >= 0) {
5823                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5824                 if (ret > 0)
5825                         break;
5826
5827                 if (level == 0)
5828                         break;
5829
5830                 if (path->slots[level] >=
5831                     btrfs_header_nritems(path->nodes[level]))
5832                         break;
5833
5834                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5835                 if (ret > 0) {
5836                         path->slots[level]++;
5837                         continue;
5838                 } else if (ret < 0)
5839                         return ret;
5840                 level = wc->level;
5841         }
5842         return 0;
5843 }
5844
5845 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5846                                  struct btrfs_root *root,
5847                                  struct btrfs_path *path,
5848                                  struct walk_control *wc, int max_level)
5849 {
5850         int level = wc->level;
5851         int ret;
5852
5853         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5854         while (level < max_level && path->nodes[level]) {
5855                 wc->level = level;
5856                 if (path->slots[level] + 1 <
5857                     btrfs_header_nritems(path->nodes[level])) {
5858                         path->slots[level]++;
5859                         return 0;
5860                 } else {
5861                         ret = walk_up_proc(trans, root, path, wc);
5862                         if (ret > 0)
5863                                 return 0;
5864
5865                         if (path->locks[level]) {
5866                                 btrfs_tree_unlock(path->nodes[level]);
5867                                 path->locks[level] = 0;
5868                         }
5869                         free_extent_buffer(path->nodes[level]);
5870                         path->nodes[level] = NULL;
5871                         level++;
5872                 }
5873         }
5874         return 1;
5875 }
5876
5877 /*
5878  * drop a subvolume tree.
5879  *
5880  * this function traverses the tree freeing any blocks that only
5881  * referenced by the tree.
5882  *
5883  * when a shared tree block is found. this function decreases its
5884  * reference count by one. if update_ref is true, this function
5885  * also make sure backrefs for the shared block and all lower level
5886  * blocks are properly updated.
5887  */
5888 int btrfs_drop_snapshot(struct btrfs_root *root,
5889                         struct btrfs_block_rsv *block_rsv, int update_ref)
5890 {
5891         struct btrfs_path *path;
5892         struct btrfs_trans_handle *trans;
5893         struct btrfs_root *tree_root = root->fs_info->tree_root;
5894         struct btrfs_root_item *root_item = &root->root_item;
5895         struct walk_control *wc;
5896         struct btrfs_key key;
5897         int err = 0;
5898         int ret;
5899         int level;
5900
5901         path = btrfs_alloc_path();
5902         BUG_ON(!path);
5903
5904         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5905         BUG_ON(!wc);
5906
5907         trans = btrfs_start_transaction(tree_root, 0);
5908         if (block_rsv)
5909                 trans->block_rsv = block_rsv;
5910
5911         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5912                 level = btrfs_header_level(root->node);
5913                 path->nodes[level] = btrfs_lock_root_node(root);
5914                 btrfs_set_lock_blocking(path->nodes[level]);
5915                 path->slots[level] = 0;
5916                 path->locks[level] = 1;
5917                 memset(&wc->update_progress, 0,
5918                        sizeof(wc->update_progress));
5919         } else {
5920                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5921                 memcpy(&wc->update_progress, &key,
5922                        sizeof(wc->update_progress));
5923
5924                 level = root_item->drop_level;
5925                 BUG_ON(level == 0);
5926                 path->lowest_level = level;
5927                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5928                 path->lowest_level = 0;
5929                 if (ret < 0) {
5930                         err = ret;
5931                         goto out;
5932                 }
5933                 WARN_ON(ret > 0);
5934
5935                 /*
5936                  * unlock our path, this is safe because only this
5937                  * function is allowed to delete this snapshot
5938                  */
5939                 btrfs_unlock_up_safe(path, 0);
5940
5941                 level = btrfs_header_level(root->node);
5942                 while (1) {
5943                         btrfs_tree_lock(path->nodes[level]);
5944                         btrfs_set_lock_blocking(path->nodes[level]);
5945
5946                         ret = btrfs_lookup_extent_info(trans, root,
5947                                                 path->nodes[level]->start,
5948                                                 path->nodes[level]->len,
5949                                                 &wc->refs[level],
5950                                                 &wc->flags[level]);
5951                         BUG_ON(ret);
5952                         BUG_ON(wc->refs[level] == 0);
5953
5954                         if (level == root_item->drop_level)
5955                                 break;
5956
5957                         btrfs_tree_unlock(path->nodes[level]);
5958                         WARN_ON(wc->refs[level] != 1);
5959                         level--;
5960                 }
5961         }
5962
5963         wc->level = level;
5964         wc->shared_level = -1;
5965         wc->stage = DROP_REFERENCE;
5966         wc->update_ref = update_ref;
5967         wc->keep_locks = 0;
5968         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5969
5970         while (1) {
5971                 ret = walk_down_tree(trans, root, path, wc);
5972                 if (ret < 0) {
5973                         err = ret;
5974                         break;
5975                 }
5976
5977                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5978                 if (ret < 0) {
5979                         err = ret;
5980                         break;
5981                 }
5982
5983                 if (ret > 0) {
5984                         BUG_ON(wc->stage != DROP_REFERENCE);
5985                         break;
5986                 }
5987
5988                 if (wc->stage == DROP_REFERENCE) {
5989                         level = wc->level;
5990                         btrfs_node_key(path->nodes[level],
5991                                        &root_item->drop_progress,
5992                                        path->slots[level]);
5993                         root_item->drop_level = level;
5994                 }
5995
5996                 BUG_ON(wc->level == 0);
5997                 if (btrfs_should_end_transaction(trans, tree_root)) {
5998                         ret = btrfs_update_root(trans, tree_root,
5999                                                 &root->root_key,
6000                                                 root_item);
6001                         BUG_ON(ret);
6002
6003                         btrfs_end_transaction_throttle(trans, tree_root);
6004                         trans = btrfs_start_transaction(tree_root, 0);
6005                         if (block_rsv)
6006                                 trans->block_rsv = block_rsv;
6007                 }
6008         }
6009         btrfs_release_path(root, path);
6010         BUG_ON(err);
6011
6012         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6013         BUG_ON(ret);
6014
6015         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6016                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6017                                            NULL, NULL);
6018                 BUG_ON(ret < 0);
6019                 if (ret > 0) {
6020                         ret = btrfs_del_orphan_item(trans, tree_root,
6021                                                     root->root_key.objectid);
6022                         BUG_ON(ret);
6023                 }
6024         }
6025
6026         if (root->in_radix) {
6027                 btrfs_free_fs_root(tree_root->fs_info, root);
6028         } else {
6029                 free_extent_buffer(root->node);
6030                 free_extent_buffer(root->commit_root);
6031                 kfree(root);
6032         }
6033 out:
6034         btrfs_end_transaction_throttle(trans, tree_root);
6035         kfree(wc);
6036         btrfs_free_path(path);
6037         return err;
6038 }
6039
6040 /*
6041  * drop subtree rooted at tree block 'node'.
6042  *
6043  * NOTE: this function will unlock and release tree block 'node'
6044  */
6045 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6046                         struct btrfs_root *root,
6047                         struct extent_buffer *node,
6048                         struct extent_buffer *parent)
6049 {
6050         struct btrfs_path *path;
6051         struct walk_control *wc;
6052         int level;
6053         int parent_level;
6054         int ret = 0;
6055         int wret;
6056
6057         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6058
6059         path = btrfs_alloc_path();
6060         BUG_ON(!path);
6061
6062         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6063         BUG_ON(!wc);
6064
6065         btrfs_assert_tree_locked(parent);
6066         parent_level = btrfs_header_level(parent);
6067         extent_buffer_get(parent);
6068         path->nodes[parent_level] = parent;
6069         path->slots[parent_level] = btrfs_header_nritems(parent);
6070
6071         btrfs_assert_tree_locked(node);
6072         level = btrfs_header_level(node);
6073         path->nodes[level] = node;
6074         path->slots[level] = 0;
6075         path->locks[level] = 1;
6076
6077         wc->refs[parent_level] = 1;
6078         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6079         wc->level = level;
6080         wc->shared_level = -1;
6081         wc->stage = DROP_REFERENCE;
6082         wc->update_ref = 0;
6083         wc->keep_locks = 1;
6084         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6085
6086         while (1) {
6087                 wret = walk_down_tree(trans, root, path, wc);
6088                 if (wret < 0) {
6089                         ret = wret;
6090                         break;
6091                 }
6092
6093                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6094                 if (wret < 0)
6095                         ret = wret;
6096                 if (wret != 0)
6097                         break;
6098         }
6099
6100         kfree(wc);
6101         btrfs_free_path(path);
6102         return ret;
6103 }
6104
6105 #if 0
6106 static unsigned long calc_ra(unsigned long start, unsigned long last,
6107                              unsigned long nr)
6108 {
6109         return min(last, start + nr - 1);
6110 }
6111
6112 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6113                                          u64 len)
6114 {
6115         u64 page_start;
6116         u64 page_end;
6117         unsigned long first_index;
6118         unsigned long last_index;
6119         unsigned long i;
6120         struct page *page;
6121         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6122         struct file_ra_state *ra;
6123         struct btrfs_ordered_extent *ordered;
6124         unsigned int total_read = 0;
6125         unsigned int total_dirty = 0;
6126         int ret = 0;
6127
6128         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6129
6130         mutex_lock(&inode->i_mutex);
6131         first_index = start >> PAGE_CACHE_SHIFT;
6132         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6133
6134         /* make sure the dirty trick played by the caller work */
6135         ret = invalidate_inode_pages2_range(inode->i_mapping,
6136                                             first_index, last_index);
6137         if (ret)
6138                 goto out_unlock;
6139
6140         file_ra_state_init(ra, inode->i_mapping);
6141
6142         for (i = first_index ; i <= last_index; i++) {
6143                 if (total_read % ra->ra_pages == 0) {
6144                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6145                                        calc_ra(i, last_index, ra->ra_pages));
6146                 }
6147                 total_read++;
6148 again:
6149                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6150                         BUG_ON(1);
6151                 page = grab_cache_page(inode->i_mapping, i);
6152                 if (!page) {
6153                         ret = -ENOMEM;
6154                         goto out_unlock;
6155                 }
6156                 if (!PageUptodate(page)) {
6157                         btrfs_readpage(NULL, page);
6158                         lock_page(page);
6159                         if (!PageUptodate(page)) {
6160                                 unlock_page(page);
6161                                 page_cache_release(page);
6162                                 ret = -EIO;
6163                                 goto out_unlock;
6164                         }
6165                 }
6166                 wait_on_page_writeback(page);
6167
6168                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6169                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6170                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6171
6172                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6173                 if (ordered) {
6174                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6175                         unlock_page(page);
6176                         page_cache_release(page);
6177                         btrfs_start_ordered_extent(inode, ordered, 1);
6178                         btrfs_put_ordered_extent(ordered);
6179                         goto again;
6180                 }
6181                 set_page_extent_mapped(page);
6182
6183                 if (i == first_index)
6184                         set_extent_bits(io_tree, page_start, page_end,
6185                                         EXTENT_BOUNDARY, GFP_NOFS);
6186                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6187
6188                 set_page_dirty(page);
6189                 total_dirty++;
6190
6191                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6192                 unlock_page(page);
6193                 page_cache_release(page);
6194         }
6195
6196 out_unlock:
6197         kfree(ra);
6198         mutex_unlock(&inode->i_mutex);
6199         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6200         return ret;
6201 }
6202
6203 static noinline int relocate_data_extent(struct inode *reloc_inode,
6204                                          struct btrfs_key *extent_key,
6205                                          u64 offset)
6206 {
6207         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6208         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6209         struct extent_map *em;
6210         u64 start = extent_key->objectid - offset;
6211         u64 end = start + extent_key->offset - 1;
6212
6213         em = alloc_extent_map(GFP_NOFS);
6214         BUG_ON(!em || IS_ERR(em));
6215
6216         em->start = start;
6217         em->len = extent_key->offset;
6218         em->block_len = extent_key->offset;
6219         em->block_start = extent_key->objectid;
6220         em->bdev = root->fs_info->fs_devices->latest_bdev;
6221         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6222
6223         /* setup extent map to cheat btrfs_readpage */
6224         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6225         while (1) {
6226                 int ret;
6227                 write_lock(&em_tree->lock);
6228                 ret = add_extent_mapping(em_tree, em);
6229                 write_unlock(&em_tree->lock);
6230                 if (ret != -EEXIST) {
6231                         free_extent_map(em);
6232                         break;
6233                 }
6234                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6235         }
6236         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6237
6238         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6239 }
6240
6241 struct btrfs_ref_path {
6242         u64 extent_start;
6243         u64 nodes[BTRFS_MAX_LEVEL];
6244         u64 root_objectid;
6245         u64 root_generation;
6246         u64 owner_objectid;
6247         u32 num_refs;
6248         int lowest_level;
6249         int current_level;
6250         int shared_level;
6251
6252         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6253         u64 new_nodes[BTRFS_MAX_LEVEL];
6254 };
6255
6256 struct disk_extent {
6257         u64 ram_bytes;
6258         u64 disk_bytenr;
6259         u64 disk_num_bytes;
6260         u64 offset;
6261         u64 num_bytes;
6262         u8 compression;
6263         u8 encryption;
6264         u16 other_encoding;
6265 };
6266
6267 static int is_cowonly_root(u64 root_objectid)
6268 {
6269         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6270             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6271             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6272             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6273             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6274             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6275                 return 1;
6276         return 0;
6277 }
6278
6279 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6280                                     struct btrfs_root *extent_root,
6281                                     struct btrfs_ref_path *ref_path,
6282                                     int first_time)
6283 {
6284         struct extent_buffer *leaf;
6285         struct btrfs_path *path;
6286         struct btrfs_extent_ref *ref;
6287         struct btrfs_key key;
6288         struct btrfs_key found_key;
6289         u64 bytenr;
6290         u32 nritems;
6291         int level;
6292         int ret = 1;
6293
6294         path = btrfs_alloc_path();
6295         if (!path)
6296                 return -ENOMEM;
6297
6298         if (first_time) {
6299                 ref_path->lowest_level = -1;
6300                 ref_path->current_level = -1;
6301                 ref_path->shared_level = -1;
6302                 goto walk_up;
6303         }
6304 walk_down:
6305         level = ref_path->current_level - 1;
6306         while (level >= -1) {
6307                 u64 parent;
6308                 if (level < ref_path->lowest_level)
6309                         break;
6310
6311                 if (level >= 0)
6312                         bytenr = ref_path->nodes[level];
6313                 else
6314                         bytenr = ref_path->extent_start;
6315                 BUG_ON(bytenr == 0);
6316
6317                 parent = ref_path->nodes[level + 1];
6318                 ref_path->nodes[level + 1] = 0;
6319                 ref_path->current_level = level;
6320                 BUG_ON(parent == 0);
6321
6322                 key.objectid = bytenr;
6323                 key.offset = parent + 1;
6324                 key.type = BTRFS_EXTENT_REF_KEY;
6325
6326                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6327                 if (ret < 0)
6328                         goto out;
6329                 BUG_ON(ret == 0);
6330
6331                 leaf = path->nodes[0];
6332                 nritems = btrfs_header_nritems(leaf);
6333                 if (path->slots[0] >= nritems) {
6334                         ret = btrfs_next_leaf(extent_root, path);
6335                         if (ret < 0)
6336                                 goto out;
6337                         if (ret > 0)
6338                                 goto next;
6339                         leaf = path->nodes[0];
6340                 }
6341
6342                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6343                 if (found_key.objectid == bytenr &&
6344                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6345                         if (level < ref_path->shared_level)
6346                                 ref_path->shared_level = level;
6347                         goto found;
6348                 }
6349 next:
6350                 level--;
6351                 btrfs_release_path(extent_root, path);
6352                 cond_resched();
6353         }
6354         /* reached lowest level */
6355         ret = 1;
6356         goto out;
6357 walk_up:
6358         level = ref_path->current_level;
6359         while (level < BTRFS_MAX_LEVEL - 1) {
6360                 u64 ref_objectid;
6361
6362                 if (level >= 0)
6363                         bytenr = ref_path->nodes[level];
6364                 else
6365                         bytenr = ref_path->extent_start;
6366
6367                 BUG_ON(bytenr == 0);
6368
6369                 key.objectid = bytenr;
6370                 key.offset = 0;
6371                 key.type = BTRFS_EXTENT_REF_KEY;
6372
6373                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6374                 if (ret < 0)
6375                         goto out;
6376
6377                 leaf = path->nodes[0];
6378                 nritems = btrfs_header_nritems(leaf);
6379                 if (path->slots[0] >= nritems) {
6380                         ret = btrfs_next_leaf(extent_root, path);
6381                         if (ret < 0)
6382                                 goto out;
6383                         if (ret > 0) {
6384                                 /* the extent was freed by someone */
6385                                 if (ref_path->lowest_level == level)
6386                                         goto out;
6387                                 btrfs_release_path(extent_root, path);
6388                                 goto walk_down;
6389                         }
6390                         leaf = path->nodes[0];
6391                 }
6392
6393                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6394                 if (found_key.objectid != bytenr ||
6395                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6396                         /* the extent was freed by someone */
6397                         if (ref_path->lowest_level == level) {
6398                                 ret = 1;
6399                                 goto out;
6400                         }
6401                         btrfs_release_path(extent_root, path);
6402                         goto walk_down;
6403                 }
6404 found:
6405                 ref = btrfs_item_ptr(leaf, path->slots[0],
6406                                 struct btrfs_extent_ref);
6407                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6408                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6409                         if (first_time) {
6410                                 level = (int)ref_objectid;
6411                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6412                                 ref_path->lowest_level = level;
6413                                 ref_path->current_level = level;
6414                                 ref_path->nodes[level] = bytenr;
6415                         } else {
6416                                 WARN_ON(ref_objectid != level);
6417                         }
6418                 } else {
6419                         WARN_ON(level != -1);
6420                 }
6421                 first_time = 0;
6422
6423                 if (ref_path->lowest_level == level) {
6424                         ref_path->owner_objectid = ref_objectid;
6425                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6426                 }
6427
6428                 /*
6429                  * the block is tree root or the block isn't in reference
6430                  * counted tree.
6431                  */
6432                 if (found_key.objectid == found_key.offset ||
6433                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6434                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6435                         ref_path->root_generation =
6436                                 btrfs_ref_generation(leaf, ref);
6437                         if (level < 0) {
6438                                 /* special reference from the tree log */
6439                                 ref_path->nodes[0] = found_key.offset;
6440                                 ref_path->current_level = 0;
6441                         }
6442                         ret = 0;
6443                         goto out;
6444                 }
6445
6446                 level++;
6447                 BUG_ON(ref_path->nodes[level] != 0);
6448                 ref_path->nodes[level] = found_key.offset;
6449                 ref_path->current_level = level;
6450
6451                 /*
6452                  * the reference was created in the running transaction,
6453                  * no need to continue walking up.
6454                  */
6455                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6456                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6457                         ref_path->root_generation =
6458                                 btrfs_ref_generation(leaf, ref);
6459                         ret = 0;
6460                         goto out;
6461                 }
6462
6463                 btrfs_release_path(extent_root, path);
6464                 cond_resched();
6465         }
6466         /* reached max tree level, but no tree root found. */
6467         BUG();
6468 out:
6469         btrfs_free_path(path);
6470         return ret;
6471 }
6472
6473 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6474                                 struct btrfs_root *extent_root,
6475                                 struct btrfs_ref_path *ref_path,
6476                                 u64 extent_start)
6477 {
6478         memset(ref_path, 0, sizeof(*ref_path));
6479         ref_path->extent_start = extent_start;
6480
6481         return __next_ref_path(trans, extent_root, ref_path, 1);
6482 }
6483
6484 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6485                                struct btrfs_root *extent_root,
6486                                struct btrfs_ref_path *ref_path)
6487 {
6488         return __next_ref_path(trans, extent_root, ref_path, 0);
6489 }
6490
6491 static noinline int get_new_locations(struct inode *reloc_inode,
6492                                       struct btrfs_key *extent_key,
6493                                       u64 offset, int no_fragment,
6494                                       struct disk_extent **extents,
6495                                       int *nr_extents)
6496 {
6497         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6498         struct btrfs_path *path;
6499         struct btrfs_file_extent_item *fi;
6500         struct extent_buffer *leaf;
6501         struct disk_extent *exts = *extents;
6502         struct btrfs_key found_key;
6503         u64 cur_pos;
6504         u64 last_byte;
6505         u32 nritems;
6506         int nr = 0;
6507         int max = *nr_extents;
6508         int ret;
6509
6510         WARN_ON(!no_fragment && *extents);
6511         if (!exts) {
6512                 max = 1;
6513                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6514                 if (!exts)
6515                         return -ENOMEM;
6516         }
6517
6518         path = btrfs_alloc_path();
6519         BUG_ON(!path);
6520
6521         cur_pos = extent_key->objectid - offset;
6522         last_byte = extent_key->objectid + extent_key->offset;
6523         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6524                                        cur_pos, 0);
6525         if (ret < 0)
6526                 goto out;
6527         if (ret > 0) {
6528                 ret = -ENOENT;
6529                 goto out;
6530         }
6531
6532         while (1) {
6533                 leaf = path->nodes[0];
6534                 nritems = btrfs_header_nritems(leaf);
6535                 if (path->slots[0] >= nritems) {
6536                         ret = btrfs_next_leaf(root, path);
6537                         if (ret < 0)
6538                                 goto out;
6539                         if (ret > 0)
6540                                 break;
6541                         leaf = path->nodes[0];
6542                 }
6543
6544                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6545                 if (found_key.offset != cur_pos ||
6546                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6547                     found_key.objectid != reloc_inode->i_ino)
6548                         break;
6549
6550                 fi = btrfs_item_ptr(leaf, path->slots[0],
6551                                     struct btrfs_file_extent_item);
6552                 if (btrfs_file_extent_type(leaf, fi) !=
6553                     BTRFS_FILE_EXTENT_REG ||
6554                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6555                         break;
6556
6557                 if (nr == max) {
6558                         struct disk_extent *old = exts;
6559                         max *= 2;
6560                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6561                         memcpy(exts, old, sizeof(*exts) * nr);
6562                         if (old != *extents)
6563                                 kfree(old);
6564                 }
6565
6566                 exts[nr].disk_bytenr =
6567                         btrfs_file_extent_disk_bytenr(leaf, fi);
6568                 exts[nr].disk_num_bytes =
6569                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6570                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6571                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6572                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6573                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6574                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6575                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6576                                                                            fi);
6577                 BUG_ON(exts[nr].offset > 0);
6578                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6579                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6580
6581                 cur_pos += exts[nr].num_bytes;
6582                 nr++;
6583
6584                 if (cur_pos + offset >= last_byte)
6585                         break;
6586
6587                 if (no_fragment) {
6588                         ret = 1;
6589                         goto out;
6590                 }
6591                 path->slots[0]++;
6592         }
6593
6594         BUG_ON(cur_pos + offset > last_byte);
6595         if (cur_pos + offset < last_byte) {
6596                 ret = -ENOENT;
6597                 goto out;
6598         }
6599         ret = 0;
6600 out:
6601         btrfs_free_path(path);
6602         if (ret) {
6603                 if (exts != *extents)
6604                         kfree(exts);
6605         } else {
6606                 *extents = exts;
6607                 *nr_extents = nr;
6608         }
6609         return ret;
6610 }
6611
6612 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6613                                         struct btrfs_root *root,
6614                                         struct btrfs_path *path,
6615                                         struct btrfs_key *extent_key,
6616                                         struct btrfs_key *leaf_key,
6617                                         struct btrfs_ref_path *ref_path,
6618                                         struct disk_extent *new_extents,
6619                                         int nr_extents)
6620 {
6621         struct extent_buffer *leaf;
6622         struct btrfs_file_extent_item *fi;
6623         struct inode *inode = NULL;
6624         struct btrfs_key key;
6625         u64 lock_start = 0;
6626         u64 lock_end = 0;
6627         u64 num_bytes;
6628         u64 ext_offset;
6629         u64 search_end = (u64)-1;
6630         u32 nritems;
6631         int nr_scaned = 0;
6632         int extent_locked = 0;
6633         int extent_type;
6634         int ret;
6635
6636         memcpy(&key, leaf_key, sizeof(key));
6637         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6638                 if (key.objectid < ref_path->owner_objectid ||
6639                     (key.objectid == ref_path->owner_objectid &&
6640                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6641                         key.objectid = ref_path->owner_objectid;
6642                         key.type = BTRFS_EXTENT_DATA_KEY;
6643                         key.offset = 0;
6644                 }
6645         }
6646
6647         while (1) {
6648                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6649                 if (ret < 0)
6650                         goto out;
6651
6652                 leaf = path->nodes[0];
6653                 nritems = btrfs_header_nritems(leaf);
6654 next:
6655                 if (extent_locked && ret > 0) {
6656                         /*
6657                          * the file extent item was modified by someone
6658                          * before the extent got locked.
6659                          */
6660                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6661                                       lock_end, GFP_NOFS);
6662                         extent_locked = 0;
6663                 }
6664
6665                 if (path->slots[0] >= nritems) {
6666                         if (++nr_scaned > 2)
6667                                 break;
6668
6669                         BUG_ON(extent_locked);
6670                         ret = btrfs_next_leaf(root, path);
6671                         if (ret < 0)
6672                                 goto out;
6673                         if (ret > 0)
6674                                 break;
6675                         leaf = path->nodes[0];
6676                         nritems = btrfs_header_nritems(leaf);
6677                 }
6678
6679                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6680
6681                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6682                         if ((key.objectid > ref_path->owner_objectid) ||
6683                             (key.objectid == ref_path->owner_objectid &&
6684                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6685                             key.offset >= search_end)
6686                                 break;
6687                 }
6688
6689                 if (inode && key.objectid != inode->i_ino) {
6690                         BUG_ON(extent_locked);
6691                         btrfs_release_path(root, path);
6692                         mutex_unlock(&inode->i_mutex);
6693                         iput(inode);
6694                         inode = NULL;
6695                         continue;
6696                 }
6697
6698                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6699                         path->slots[0]++;
6700                         ret = 1;
6701                         goto next;
6702                 }
6703                 fi = btrfs_item_ptr(leaf, path->slots[0],
6704                                     struct btrfs_file_extent_item);
6705                 extent_type = btrfs_file_extent_type(leaf, fi);
6706                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6707                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6708                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6709                      extent_key->objectid)) {
6710                         path->slots[0]++;
6711                         ret = 1;
6712                         goto next;
6713                 }
6714
6715                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6716                 ext_offset = btrfs_file_extent_offset(leaf, fi);
6717
6718                 if (search_end == (u64)-1) {
6719                         search_end = key.offset - ext_offset +
6720                                 btrfs_file_extent_ram_bytes(leaf, fi);
6721                 }
6722
6723                 if (!extent_locked) {
6724                         lock_start = key.offset;
6725                         lock_end = lock_start + num_bytes - 1;
6726                 } else {
6727                         if (lock_start > key.offset ||
6728                             lock_end + 1 < key.offset + num_bytes) {
6729                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6730                                               lock_start, lock_end, GFP_NOFS);
6731                                 extent_locked = 0;
6732                         }
6733                 }
6734
6735                 if (!inode) {
6736                         btrfs_release_path(root, path);
6737
6738                         inode = btrfs_iget_locked(root->fs_info->sb,
6739                                                   key.objectid, root);
6740                         if (inode->i_state & I_NEW) {
6741                                 BTRFS_I(inode)->root = root;
6742                                 BTRFS_I(inode)->location.objectid =
6743                                         key.objectid;
6744                                 BTRFS_I(inode)->location.type =
6745                                         BTRFS_INODE_ITEM_KEY;
6746                                 BTRFS_I(inode)->location.offset = 0;
6747                                 btrfs_read_locked_inode(inode);
6748                                 unlock_new_inode(inode);
6749                         }
6750                         /*
6751                          * some code call btrfs_commit_transaction while
6752                          * holding the i_mutex, so we can't use mutex_lock
6753                          * here.
6754                          */
6755                         if (is_bad_inode(inode) ||
6756                             !mutex_trylock(&inode->i_mutex)) {
6757                                 iput(inode);
6758                                 inode = NULL;
6759                                 key.offset = (u64)-1;
6760                                 goto skip;
6761                         }
6762                 }
6763
6764                 if (!extent_locked) {
6765                         struct btrfs_ordered_extent *ordered;
6766
6767                         btrfs_release_path(root, path);
6768
6769                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6770                                     lock_end, GFP_NOFS);
6771                         ordered = btrfs_lookup_first_ordered_extent(inode,
6772                                                                     lock_end);
6773                         if (ordered &&
6774                             ordered->file_offset <= lock_end &&
6775                             ordered->file_offset + ordered->len > lock_start) {
6776                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6777                                               lock_start, lock_end, GFP_NOFS);
6778                                 btrfs_start_ordered_extent(inode, ordered, 1);
6779                                 btrfs_put_ordered_extent(ordered);
6780                                 key.offset += num_bytes;
6781                                 goto skip;
6782                         }
6783                         if (ordered)
6784                                 btrfs_put_ordered_extent(ordered);
6785
6786                         extent_locked = 1;
6787                         continue;
6788                 }
6789
6790                 if (nr_extents == 1) {
6791                         /* update extent pointer in place */
6792                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
6793                                                 new_extents[0].disk_bytenr);
6794                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6795                                                 new_extents[0].disk_num_bytes);
6796                         btrfs_mark_buffer_dirty(leaf);
6797
6798                         btrfs_drop_extent_cache(inode, key.offset,
6799                                                 key.offset + num_bytes - 1, 0);
6800
6801                         ret = btrfs_inc_extent_ref(trans, root,
6802                                                 new_extents[0].disk_bytenr,
6803                                                 new_extents[0].disk_num_bytes,
6804                                                 leaf->start,
6805                                                 root->root_key.objectid,
6806                                                 trans->transid,
6807                                                 key.objectid);
6808                         BUG_ON(ret);
6809
6810                         ret = btrfs_free_extent(trans, root,
6811                                                 extent_key->objectid,
6812                                                 extent_key->offset,
6813                                                 leaf->start,
6814                                                 btrfs_header_owner(leaf),
6815                                                 btrfs_header_generation(leaf),
6816                                                 key.objectid, 0);
6817                         BUG_ON(ret);
6818
6819                         btrfs_release_path(root, path);
6820                         key.offset += num_bytes;
6821                 } else {
6822                         BUG_ON(1);
6823 #if 0
6824                         u64 alloc_hint;
6825                         u64 extent_len;
6826                         int i;
6827                         /*
6828                          * drop old extent pointer at first, then insert the
6829                          * new pointers one bye one
6830                          */
6831                         btrfs_release_path(root, path);
6832                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
6833                                                  key.offset + num_bytes,
6834                                                  key.offset, &alloc_hint);
6835                         BUG_ON(ret);
6836
6837                         for (i = 0; i < nr_extents; i++) {
6838                                 if (ext_offset >= new_extents[i].num_bytes) {
6839                                         ext_offset -= new_extents[i].num_bytes;
6840                                         continue;
6841                                 }
6842                                 extent_len = min(new_extents[i].num_bytes -
6843                                                  ext_offset, num_bytes);
6844
6845                                 ret = btrfs_insert_empty_item(trans, root,
6846                                                               path, &key,
6847                                                               sizeof(*fi));
6848                                 BUG_ON(ret);
6849
6850                                 leaf = path->nodes[0];
6851                                 fi = btrfs_item_ptr(leaf, path->slots[0],
6852                                                 struct btrfs_file_extent_item);
6853                                 btrfs_set_file_extent_generation(leaf, fi,
6854                                                         trans->transid);
6855                                 btrfs_set_file_extent_type(leaf, fi,
6856                                                         BTRFS_FILE_EXTENT_REG);
6857                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6858                                                 new_extents[i].disk_bytenr);
6859                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6860                                                 new_extents[i].disk_num_bytes);
6861                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
6862                                                 new_extents[i].ram_bytes);
6863
6864                                 btrfs_set_file_extent_compression(leaf, fi,
6865                                                 new_extents[i].compression);
6866                                 btrfs_set_file_extent_encryption(leaf, fi,
6867                                                 new_extents[i].encryption);
6868                                 btrfs_set_file_extent_other_encoding(leaf, fi,
6869                                                 new_extents[i].other_encoding);
6870
6871                                 btrfs_set_file_extent_num_bytes(leaf, fi,
6872                                                         extent_len);
6873                                 ext_offset += new_extents[i].offset;
6874                                 btrfs_set_file_extent_offset(leaf, fi,
6875                                                         ext_offset);
6876                                 btrfs_mark_buffer_dirty(leaf);
6877
6878                                 btrfs_drop_extent_cache(inode, key.offset,
6879                                                 key.offset + extent_len - 1, 0);
6880
6881                                 ret = btrfs_inc_extent_ref(trans, root,
6882                                                 new_extents[i].disk_bytenr,
6883                                                 new_extents[i].disk_num_bytes,
6884                                                 leaf->start,
6885                                                 root->root_key.objectid,
6886                                                 trans->transid, key.objectid);
6887                                 BUG_ON(ret);
6888                                 btrfs_release_path(root, path);
6889
6890                                 inode_add_bytes(inode, extent_len);
6891
6892                                 ext_offset = 0;
6893                                 num_bytes -= extent_len;
6894                                 key.offset += extent_len;
6895
6896                                 if (num_bytes == 0)
6897                                         break;
6898                         }
6899                         BUG_ON(i >= nr_extents);
6900 #endif
6901                 }
6902
6903                 if (extent_locked) {
6904                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6905                                       lock_end, GFP_NOFS);
6906                         extent_locked = 0;
6907                 }
6908 skip:
6909                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6910                     key.offset >= search_end)
6911                         break;
6912
6913                 cond_resched();
6914         }
6915         ret = 0;
6916 out:
6917         btrfs_release_path(root, path);
6918         if (inode) {
6919                 mutex_unlock(&inode->i_mutex);
6920                 if (extent_locked) {
6921                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6922                                       lock_end, GFP_NOFS);
6923                 }
6924                 iput(inode);
6925         }
6926         return ret;
6927 }
6928
6929 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6930                                struct btrfs_root *root,
6931                                struct extent_buffer *buf, u64 orig_start)
6932 {
6933         int level;
6934         int ret;
6935
6936         BUG_ON(btrfs_header_generation(buf) != trans->transid);
6937         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6938
6939         level = btrfs_header_level(buf);
6940         if (level == 0) {
6941                 struct btrfs_leaf_ref *ref;
6942                 struct btrfs_leaf_ref *orig_ref;
6943
6944                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6945                 if (!orig_ref)
6946                         return -ENOENT;
6947
6948                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6949                 if (!ref) {
6950                         btrfs_free_leaf_ref(root, orig_ref);
6951                         return -ENOMEM;
6952                 }
6953
6954                 ref->nritems = orig_ref->nritems;
6955                 memcpy(ref->extents, orig_ref->extents,
6956                         sizeof(ref->extents[0]) * ref->nritems);
6957
6958                 btrfs_free_leaf_ref(root, orig_ref);
6959
6960                 ref->root_gen = trans->transid;
6961                 ref->bytenr = buf->start;
6962                 ref->owner = btrfs_header_owner(buf);
6963                 ref->generation = btrfs_header_generation(buf);
6964
6965                 ret = btrfs_add_leaf_ref(root, ref, 0);
6966                 WARN_ON(ret);
6967                 btrfs_free_leaf_ref(root, ref);
6968         }
6969         return 0;
6970 }
6971
6972 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6973                                         struct extent_buffer *leaf,
6974                                         struct btrfs_block_group_cache *group,
6975                                         struct btrfs_root *target_root)
6976 {
6977         struct btrfs_key key;
6978         struct inode *inode = NULL;
6979         struct btrfs_file_extent_item *fi;
6980         struct extent_state *cached_state = NULL;
6981         u64 num_bytes;
6982         u64 skip_objectid = 0;
6983         u32 nritems;
6984         u32 i;
6985
6986         nritems = btrfs_header_nritems(leaf);
6987         for (i = 0; i < nritems; i++) {
6988                 btrfs_item_key_to_cpu(leaf, &key, i);
6989                 if (key.objectid == skip_objectid ||
6990                     key.type != BTRFS_EXTENT_DATA_KEY)
6991                         continue;
6992                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6993                 if (btrfs_file_extent_type(leaf, fi) ==
6994                     BTRFS_FILE_EXTENT_INLINE)
6995                         continue;
6996                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6997                         continue;
6998                 if (!inode || inode->i_ino != key.objectid) {
6999                         iput(inode);
7000                         inode = btrfs_ilookup(target_root->fs_info->sb,
7001                                               key.objectid, target_root, 1);
7002                 }
7003                 if (!inode) {
7004                         skip_objectid = key.objectid;
7005                         continue;
7006                 }
7007                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7008
7009                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7010                                  key.offset + num_bytes - 1, 0, &cached_state,
7011                                  GFP_NOFS);
7012                 btrfs_drop_extent_cache(inode, key.offset,
7013                                         key.offset + num_bytes - 1, 1);
7014                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7015                                      key.offset + num_bytes - 1, &cached_state,
7016                                      GFP_NOFS);
7017                 cond_resched();
7018         }
7019         iput(inode);
7020         return 0;
7021 }
7022
7023 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7024                                         struct btrfs_root *root,
7025                                         struct extent_buffer *leaf,
7026                                         struct btrfs_block_group_cache *group,
7027                                         struct inode *reloc_inode)
7028 {
7029         struct btrfs_key key;
7030         struct btrfs_key extent_key;
7031         struct btrfs_file_extent_item *fi;
7032         struct btrfs_leaf_ref *ref;
7033         struct disk_extent *new_extent;
7034         u64 bytenr;
7035         u64 num_bytes;
7036         u32 nritems;
7037         u32 i;
7038         int ext_index;
7039         int nr_extent;
7040         int ret;
7041
7042         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7043         BUG_ON(!new_extent);
7044
7045         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7046         BUG_ON(!ref);
7047
7048         ext_index = -1;
7049         nritems = btrfs_header_nritems(leaf);
7050         for (i = 0; i < nritems; i++) {
7051                 btrfs_item_key_to_cpu(leaf, &key, i);
7052                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7053                         continue;
7054                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7055                 if (btrfs_file_extent_type(leaf, fi) ==
7056                     BTRFS_FILE_EXTENT_INLINE)
7057                         continue;
7058                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7059                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7060                 if (bytenr == 0)
7061                         continue;
7062
7063                 ext_index++;
7064                 if (bytenr >= group->key.objectid + group->key.offset ||
7065                     bytenr + num_bytes <= group->key.objectid)
7066                         continue;
7067
7068                 extent_key.objectid = bytenr;
7069                 extent_key.offset = num_bytes;
7070                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7071                 nr_extent = 1;
7072                 ret = get_new_locations(reloc_inode, &extent_key,
7073                                         group->key.objectid, 1,
7074                                         &new_extent, &nr_extent);
7075                 if (ret > 0)
7076                         continue;
7077                 BUG_ON(ret < 0);
7078
7079                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7080                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7081                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7082                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7083
7084                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7085                                                 new_extent->disk_bytenr);
7086                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7087                                                 new_extent->disk_num_bytes);
7088                 btrfs_mark_buffer_dirty(leaf);
7089
7090                 ret = btrfs_inc_extent_ref(trans, root,
7091                                         new_extent->disk_bytenr,
7092                                         new_extent->disk_num_bytes,
7093                                         leaf->start,
7094                                         root->root_key.objectid,
7095                                         trans->transid, key.objectid);
7096                 BUG_ON(ret);
7097
7098                 ret = btrfs_free_extent(trans, root,
7099                                         bytenr, num_bytes, leaf->start,
7100                                         btrfs_header_owner(leaf),
7101                                         btrfs_header_generation(leaf),
7102                                         key.objectid, 0);
7103                 BUG_ON(ret);
7104                 cond_resched();
7105         }
7106         kfree(new_extent);
7107         BUG_ON(ext_index + 1 != ref->nritems);
7108         btrfs_free_leaf_ref(root, ref);
7109         return 0;
7110 }
7111
7112 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7113                           struct btrfs_root *root)
7114 {
7115         struct btrfs_root *reloc_root;
7116         int ret;
7117
7118         if (root->reloc_root) {
7119                 reloc_root = root->reloc_root;
7120                 root->reloc_root = NULL;
7121                 list_add(&reloc_root->dead_list,
7122                          &root->fs_info->dead_reloc_roots);
7123
7124                 btrfs_set_root_bytenr(&reloc_root->root_item,
7125                                       reloc_root->node->start);
7126                 btrfs_set_root_level(&root->root_item,
7127                                      btrfs_header_level(reloc_root->node));
7128                 memset(&reloc_root->root_item.drop_progress, 0,
7129                         sizeof(struct btrfs_disk_key));
7130                 reloc_root->root_item.drop_level = 0;
7131
7132                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7133                                         &reloc_root->root_key,
7134                                         &reloc_root->root_item);
7135                 BUG_ON(ret);
7136         }
7137         return 0;
7138 }
7139
7140 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7141 {
7142         struct btrfs_trans_handle *trans;
7143         struct btrfs_root *reloc_root;
7144         struct btrfs_root *prev_root = NULL;
7145         struct list_head dead_roots;
7146         int ret;
7147         unsigned long nr;
7148
7149         INIT_LIST_HEAD(&dead_roots);
7150         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7151
7152         while (!list_empty(&dead_roots)) {
7153                 reloc_root = list_entry(dead_roots.prev,
7154                                         struct btrfs_root, dead_list);
7155                 list_del_init(&reloc_root->dead_list);
7156
7157                 BUG_ON(reloc_root->commit_root != NULL);
7158                 while (1) {
7159                         trans = btrfs_join_transaction(root, 1);
7160                         BUG_ON(!trans);
7161
7162                         mutex_lock(&root->fs_info->drop_mutex);
7163                         ret = btrfs_drop_snapshot(trans, reloc_root);
7164                         if (ret != -EAGAIN)
7165                                 break;
7166                         mutex_unlock(&root->fs_info->drop_mutex);
7167
7168                         nr = trans->blocks_used;
7169                         ret = btrfs_end_transaction(trans, root);
7170                         BUG_ON(ret);
7171                         btrfs_btree_balance_dirty(root, nr);
7172                 }
7173
7174                 free_extent_buffer(reloc_root->node);
7175
7176                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7177                                      &reloc_root->root_key);
7178                 BUG_ON(ret);
7179                 mutex_unlock(&root->fs_info->drop_mutex);
7180
7181                 nr = trans->blocks_used;
7182                 ret = btrfs_end_transaction(trans, root);
7183                 BUG_ON(ret);
7184                 btrfs_btree_balance_dirty(root, nr);
7185
7186                 kfree(prev_root);
7187                 prev_root = reloc_root;
7188         }
7189         if (prev_root) {
7190                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7191                 kfree(prev_root);
7192         }
7193         return 0;
7194 }
7195
7196 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7197 {
7198         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7199         return 0;
7200 }
7201
7202 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7203 {
7204         struct btrfs_root *reloc_root;
7205         struct btrfs_trans_handle *trans;
7206         struct btrfs_key location;
7207         int found;
7208         int ret;
7209
7210         mutex_lock(&root->fs_info->tree_reloc_mutex);
7211         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7212         BUG_ON(ret);
7213         found = !list_empty(&root->fs_info->dead_reloc_roots);
7214         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7215
7216         if (found) {
7217                 trans = btrfs_start_transaction(root, 1);
7218                 BUG_ON(!trans);
7219                 ret = btrfs_commit_transaction(trans, root);
7220                 BUG_ON(ret);
7221         }
7222
7223         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7224         location.offset = (u64)-1;
7225         location.type = BTRFS_ROOT_ITEM_KEY;
7226
7227         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7228         BUG_ON(!reloc_root);
7229         btrfs_orphan_cleanup(reloc_root);
7230         return 0;
7231 }
7232
7233 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7234                                     struct btrfs_root *root)
7235 {
7236         struct btrfs_root *reloc_root;
7237         struct extent_buffer *eb;
7238         struct btrfs_root_item *root_item;
7239         struct btrfs_key root_key;
7240         int ret;
7241
7242         BUG_ON(!root->ref_cows);
7243         if (root->reloc_root)
7244                 return 0;
7245
7246         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7247         BUG_ON(!root_item);
7248
7249         ret = btrfs_copy_root(trans, root, root->commit_root,
7250                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7251         BUG_ON(ret);
7252
7253         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7254         root_key.offset = root->root_key.objectid;
7255         root_key.type = BTRFS_ROOT_ITEM_KEY;
7256
7257         memcpy(root_item, &root->root_item, sizeof(root_item));
7258         btrfs_set_root_refs(root_item, 0);
7259         btrfs_set_root_bytenr(root_item, eb->start);
7260         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7261         btrfs_set_root_generation(root_item, trans->transid);
7262
7263         btrfs_tree_unlock(eb);
7264         free_extent_buffer(eb);
7265
7266         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7267                                 &root_key, root_item);
7268         BUG_ON(ret);
7269         kfree(root_item);
7270
7271         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7272                                                  &root_key);
7273         BUG_ON(!reloc_root);
7274         reloc_root->last_trans = trans->transid;
7275         reloc_root->commit_root = NULL;
7276         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7277
7278         root->reloc_root = reloc_root;
7279         return 0;
7280 }
7281
7282 /*
7283  * Core function of space balance.
7284  *
7285  * The idea is using reloc trees to relocate tree blocks in reference
7286  * counted roots. There is one reloc tree for each subvol, and all
7287  * reloc trees share same root key objectid. Reloc trees are snapshots
7288  * of the latest committed roots of subvols (root->commit_root).
7289  *
7290  * To relocate a tree block referenced by a subvol, there are two steps.
7291  * COW the block through subvol's reloc tree, then update block pointer
7292  * in the subvol to point to the new block. Since all reloc trees share
7293  * same root key objectid, doing special handing for tree blocks owned
7294  * by them is easy. Once a tree block has been COWed in one reloc tree,
7295  * we can use the resulting new block directly when the same block is
7296  * required to COW again through other reloc trees. By this way, relocated
7297  * tree blocks are shared between reloc trees, so they are also shared
7298  * between subvols.
7299  */
7300 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7301                                       struct btrfs_root *root,
7302                                       struct btrfs_path *path,
7303                                       struct btrfs_key *first_key,
7304                                       struct btrfs_ref_path *ref_path,
7305                                       struct btrfs_block_group_cache *group,
7306                                       struct inode *reloc_inode)
7307 {
7308         struct btrfs_root *reloc_root;
7309         struct extent_buffer *eb = NULL;
7310         struct btrfs_key *keys;
7311         u64 *nodes;
7312         int level;
7313         int shared_level;
7314         int lowest_level = 0;
7315         int ret;
7316
7317         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7318                 lowest_level = ref_path->owner_objectid;
7319
7320         if (!root->ref_cows) {
7321                 path->lowest_level = lowest_level;
7322                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7323                 BUG_ON(ret < 0);
7324                 path->lowest_level = 0;
7325                 btrfs_release_path(root, path);
7326                 return 0;
7327         }
7328
7329         mutex_lock(&root->fs_info->tree_reloc_mutex);
7330         ret = init_reloc_tree(trans, root);
7331         BUG_ON(ret);
7332         reloc_root = root->reloc_root;
7333
7334         shared_level = ref_path->shared_level;
7335         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7336
7337         keys = ref_path->node_keys;
7338         nodes = ref_path->new_nodes;
7339         memset(&keys[shared_level + 1], 0,
7340                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7341         memset(&nodes[shared_level + 1], 0,
7342                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7343
7344         if (nodes[lowest_level] == 0) {
7345                 path->lowest_level = lowest_level;
7346                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7347                                         0, 1);
7348                 BUG_ON(ret);
7349                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7350                         eb = path->nodes[level];
7351                         if (!eb || eb == reloc_root->node)
7352                                 break;
7353                         nodes[level] = eb->start;
7354                         if (level == 0)
7355                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7356                         else
7357                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7358                 }
7359                 if (nodes[0] &&
7360                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7361                         eb = path->nodes[0];
7362                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7363                                                       group, reloc_inode);
7364                         BUG_ON(ret);
7365                 }
7366                 btrfs_release_path(reloc_root, path);
7367         } else {
7368                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7369                                        lowest_level);
7370                 BUG_ON(ret);
7371         }
7372
7373         /*
7374          * replace tree blocks in the fs tree with tree blocks in
7375          * the reloc tree.
7376          */
7377         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7378         BUG_ON(ret < 0);
7379
7380         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7381                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7382                                         0, 0);
7383                 BUG_ON(ret);
7384                 extent_buffer_get(path->nodes[0]);
7385                 eb = path->nodes[0];
7386                 btrfs_release_path(reloc_root, path);
7387                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7388                 BUG_ON(ret);
7389                 free_extent_buffer(eb);
7390         }
7391
7392         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7393         path->lowest_level = 0;
7394         return 0;
7395 }
7396
7397 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7398                                         struct btrfs_root *root,
7399                                         struct btrfs_path *path,
7400                                         struct btrfs_key *first_key,
7401                                         struct btrfs_ref_path *ref_path)
7402 {
7403         int ret;
7404
7405         ret = relocate_one_path(trans, root, path, first_key,
7406                                 ref_path, NULL, NULL);
7407         BUG_ON(ret);
7408
7409         return 0;
7410 }
7411
7412 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7413                                     struct btrfs_root *extent_root,
7414                                     struct btrfs_path *path,
7415                                     struct btrfs_key *extent_key)
7416 {
7417         int ret;
7418
7419         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7420         if (ret)
7421                 goto out;
7422         ret = btrfs_del_item(trans, extent_root, path);
7423 out:
7424         btrfs_release_path(extent_root, path);
7425         return ret;
7426 }
7427
7428 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7429                                                 struct btrfs_ref_path *ref_path)
7430 {
7431         struct btrfs_key root_key;
7432
7433         root_key.objectid = ref_path->root_objectid;
7434         root_key.type = BTRFS_ROOT_ITEM_KEY;
7435         if (is_cowonly_root(ref_path->root_objectid))
7436                 root_key.offset = 0;
7437         else
7438                 root_key.offset = (u64)-1;
7439
7440         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7441 }
7442
7443 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7444                                         struct btrfs_path *path,
7445                                         struct btrfs_key *extent_key,
7446                                         struct btrfs_block_group_cache *group,
7447                                         struct inode *reloc_inode, int pass)
7448 {
7449         struct btrfs_trans_handle *trans;
7450         struct btrfs_root *found_root;
7451         struct btrfs_ref_path *ref_path = NULL;
7452         struct disk_extent *new_extents = NULL;
7453         int nr_extents = 0;
7454         int loops;
7455         int ret;
7456         int level;
7457         struct btrfs_key first_key;
7458         u64 prev_block = 0;
7459
7460
7461         trans = btrfs_start_transaction(extent_root, 1);
7462         BUG_ON(!trans);
7463
7464         if (extent_key->objectid == 0) {
7465                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7466                 goto out;
7467         }
7468
7469         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7470         if (!ref_path) {
7471                 ret = -ENOMEM;
7472                 goto out;
7473         }
7474
7475         for (loops = 0; ; loops++) {
7476                 if (loops == 0) {
7477                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7478                                                    extent_key->objectid);
7479                 } else {
7480                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7481                 }
7482                 if (ret < 0)
7483                         goto out;
7484                 if (ret > 0)
7485                         break;
7486
7487                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7488                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7489                         continue;
7490
7491                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7492                 BUG_ON(!found_root);
7493                 /*
7494                  * for reference counted tree, only process reference paths
7495                  * rooted at the latest committed root.
7496                  */
7497                 if (found_root->ref_cows &&
7498                     ref_path->root_generation != found_root->root_key.offset)
7499                         continue;
7500
7501                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7502                         if (pass == 0) {
7503                                 /*
7504                                  * copy data extents to new locations
7505                                  */
7506                                 u64 group_start = group->key.objectid;
7507                                 ret = relocate_data_extent(reloc_inode,
7508                                                            extent_key,
7509                                                            group_start);
7510                                 if (ret < 0)
7511                                         goto out;
7512                                 break;
7513                         }
7514                         level = 0;
7515                 } else {
7516                         level = ref_path->owner_objectid;
7517                 }
7518
7519                 if (prev_block != ref_path->nodes[level]) {
7520                         struct extent_buffer *eb;
7521                         u64 block_start = ref_path->nodes[level];
7522                         u64 block_size = btrfs_level_size(found_root, level);
7523
7524                         eb = read_tree_block(found_root, block_start,
7525                                              block_size, 0);
7526                         btrfs_tree_lock(eb);
7527                         BUG_ON(level != btrfs_header_level(eb));
7528
7529                         if (level == 0)
7530                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7531                         else
7532                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7533
7534                         btrfs_tree_unlock(eb);
7535                         free_extent_buffer(eb);
7536                         prev_block = block_start;
7537                 }
7538
7539                 mutex_lock(&extent_root->fs_info->trans_mutex);
7540                 btrfs_record_root_in_trans(found_root);
7541                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7542                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7543                         /*
7544                          * try to update data extent references while
7545                          * keeping metadata shared between snapshots.
7546                          */
7547                         if (pass == 1) {
7548                                 ret = relocate_one_path(trans, found_root,
7549                                                 path, &first_key, ref_path,
7550                                                 group, reloc_inode);
7551                                 if (ret < 0)
7552                                         goto out;
7553                                 continue;
7554                         }
7555                         /*
7556                          * use fallback method to process the remaining
7557                          * references.
7558                          */
7559                         if (!new_extents) {
7560                                 u64 group_start = group->key.objectid;
7561                                 new_extents = kmalloc(sizeof(*new_extents),
7562                                                       GFP_NOFS);
7563                                 nr_extents = 1;
7564                                 ret = get_new_locations(reloc_inode,
7565                                                         extent_key,
7566                                                         group_start, 1,
7567                                                         &new_extents,
7568                                                         &nr_extents);
7569                                 if (ret)
7570                                         goto out;
7571                         }
7572                         ret = replace_one_extent(trans, found_root,
7573                                                 path, extent_key,
7574                                                 &first_key, ref_path,
7575                                                 new_extents, nr_extents);
7576                 } else {
7577                         ret = relocate_tree_block(trans, found_root, path,
7578                                                   &first_key, ref_path);
7579                 }
7580                 if (ret < 0)
7581                         goto out;
7582         }
7583         ret = 0;
7584 out:
7585         btrfs_end_transaction(trans, extent_root);
7586         kfree(new_extents);
7587         kfree(ref_path);
7588         return ret;
7589 }
7590 #endif
7591
7592 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7593 {
7594         u64 num_devices;
7595         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7596                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7597
7598         num_devices = root->fs_info->fs_devices->rw_devices;
7599         if (num_devices == 1) {
7600                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7601                 stripped = flags & ~stripped;
7602
7603                 /* turn raid0 into single device chunks */
7604                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7605                         return stripped;
7606
7607                 /* turn mirroring into duplication */
7608                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7609                              BTRFS_BLOCK_GROUP_RAID10))
7610                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7611                 return flags;
7612         } else {
7613                 /* they already had raid on here, just return */
7614                 if (flags & stripped)
7615                         return flags;
7616
7617                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7618                 stripped = flags & ~stripped;
7619
7620                 /* switch duplicated blocks with raid1 */
7621                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7622                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7623
7624                 /* turn single device chunks into raid0 */
7625                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7626         }
7627         return flags;
7628 }
7629
7630 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
7631 {
7632         struct btrfs_space_info *sinfo = cache->space_info;
7633         u64 num_bytes;
7634         int ret = -ENOSPC;
7635
7636         if (cache->ro)
7637                 return 0;
7638
7639         spin_lock(&sinfo->lock);
7640         spin_lock(&cache->lock);
7641         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7642                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7643
7644         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7645             sinfo->bytes_may_use + sinfo->bytes_readonly +
7646             cache->reserved_pinned + num_bytes < sinfo->total_bytes) {
7647                 sinfo->bytes_readonly += num_bytes;
7648                 sinfo->bytes_reserved += cache->reserved_pinned;
7649                 cache->reserved_pinned = 0;
7650                 cache->ro = 1;
7651                 ret = 0;
7652         }
7653         spin_unlock(&cache->lock);
7654         spin_unlock(&sinfo->lock);
7655         return ret;
7656 }
7657
7658 int btrfs_set_block_group_ro(struct btrfs_root *root,
7659                              struct btrfs_block_group_cache *cache)
7660
7661 {
7662         struct btrfs_trans_handle *trans;
7663         u64 alloc_flags;
7664         int ret;
7665
7666         BUG_ON(cache->ro);
7667
7668         trans = btrfs_join_transaction(root, 1);
7669         BUG_ON(IS_ERR(trans));
7670
7671         alloc_flags = update_block_group_flags(root, cache->flags);
7672         if (alloc_flags != cache->flags)
7673                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7674
7675         ret = set_block_group_ro(cache);
7676         if (!ret)
7677                 goto out;
7678         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7679         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7680         if (ret < 0)
7681                 goto out;
7682         ret = set_block_group_ro(cache);
7683 out:
7684         btrfs_end_transaction(trans, root);
7685         return ret;
7686 }
7687
7688 int btrfs_set_block_group_rw(struct btrfs_root *root,
7689                               struct btrfs_block_group_cache *cache)
7690 {
7691         struct btrfs_space_info *sinfo = cache->space_info;
7692         u64 num_bytes;
7693
7694         BUG_ON(!cache->ro);
7695
7696         spin_lock(&sinfo->lock);
7697         spin_lock(&cache->lock);
7698         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7699                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7700         sinfo->bytes_readonly -= num_bytes;
7701         cache->ro = 0;
7702         spin_unlock(&cache->lock);
7703         spin_unlock(&sinfo->lock);
7704         return 0;
7705 }
7706
7707 /*
7708  * checks to see if its even possible to relocate this block group.
7709  *
7710  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7711  * ok to go ahead and try.
7712  */
7713 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7714 {
7715         struct btrfs_block_group_cache *block_group;
7716         struct btrfs_space_info *space_info;
7717         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7718         struct btrfs_device *device;
7719         int full = 0;
7720         int ret = 0;
7721
7722         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7723
7724         /* odd, couldn't find the block group, leave it alone */
7725         if (!block_group)
7726                 return -1;
7727
7728         /* no bytes used, we're good */
7729         if (!btrfs_block_group_used(&block_group->item))
7730                 goto out;
7731
7732         space_info = block_group->space_info;
7733         spin_lock(&space_info->lock);
7734
7735         full = space_info->full;
7736
7737         /*
7738          * if this is the last block group we have in this space, we can't
7739          * relocate it unless we're able to allocate a new chunk below.
7740          *
7741          * Otherwise, we need to make sure we have room in the space to handle
7742          * all of the extents from this block group.  If we can, we're good
7743          */
7744         if ((space_info->total_bytes != block_group->key.offset) &&
7745            (space_info->bytes_used + space_info->bytes_reserved +
7746             space_info->bytes_pinned + space_info->bytes_readonly +
7747             btrfs_block_group_used(&block_group->item) <
7748             space_info->total_bytes)) {
7749                 spin_unlock(&space_info->lock);
7750                 goto out;
7751         }
7752         spin_unlock(&space_info->lock);
7753
7754         /*
7755          * ok we don't have enough space, but maybe we have free space on our
7756          * devices to allocate new chunks for relocation, so loop through our
7757          * alloc devices and guess if we have enough space.  However, if we
7758          * were marked as full, then we know there aren't enough chunks, and we
7759          * can just return.
7760          */
7761         ret = -1;
7762         if (full)
7763                 goto out;
7764
7765         mutex_lock(&root->fs_info->chunk_mutex);
7766         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7767                 u64 min_free = btrfs_block_group_used(&block_group->item);
7768                 u64 dev_offset, max_avail;
7769
7770                 /*
7771                  * check to make sure we can actually find a chunk with enough
7772                  * space to fit our block group in.
7773                  */
7774                 if (device->total_bytes > device->bytes_used + min_free) {
7775                         ret = find_free_dev_extent(NULL, device, min_free,
7776                                                    &dev_offset, &max_avail);
7777                         if (!ret)
7778                                 break;
7779                         ret = -1;
7780                 }
7781         }
7782         mutex_unlock(&root->fs_info->chunk_mutex);
7783 out:
7784         btrfs_put_block_group(block_group);
7785         return ret;
7786 }
7787
7788 static int find_first_block_group(struct btrfs_root *root,
7789                 struct btrfs_path *path, struct btrfs_key *key)
7790 {
7791         int ret = 0;
7792         struct btrfs_key found_key;
7793         struct extent_buffer *leaf;
7794         int slot;
7795
7796         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7797         if (ret < 0)
7798                 goto out;
7799
7800         while (1) {
7801                 slot = path->slots[0];
7802                 leaf = path->nodes[0];
7803                 if (slot >= btrfs_header_nritems(leaf)) {
7804                         ret = btrfs_next_leaf(root, path);
7805                         if (ret == 0)
7806                                 continue;
7807                         if (ret < 0)
7808                                 goto out;
7809                         break;
7810                 }
7811                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7812
7813                 if (found_key.objectid >= key->objectid &&
7814                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7815                         ret = 0;
7816                         goto out;
7817                 }
7818                 path->slots[0]++;
7819         }
7820 out:
7821         return ret;
7822 }
7823
7824 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7825 {
7826         struct btrfs_block_group_cache *block_group;
7827         struct btrfs_space_info *space_info;
7828         struct btrfs_caching_control *caching_ctl;
7829         struct rb_node *n;
7830
7831         down_write(&info->extent_commit_sem);
7832         while (!list_empty(&info->caching_block_groups)) {
7833                 caching_ctl = list_entry(info->caching_block_groups.next,
7834                                          struct btrfs_caching_control, list);
7835                 list_del(&caching_ctl->list);
7836                 put_caching_control(caching_ctl);
7837         }
7838         up_write(&info->extent_commit_sem);
7839
7840         spin_lock(&info->block_group_cache_lock);
7841         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7842                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7843                                        cache_node);
7844                 rb_erase(&block_group->cache_node,
7845                          &info->block_group_cache_tree);
7846                 spin_unlock(&info->block_group_cache_lock);
7847
7848                 down_write(&block_group->space_info->groups_sem);
7849                 list_del(&block_group->list);
7850                 up_write(&block_group->space_info->groups_sem);
7851
7852                 if (block_group->cached == BTRFS_CACHE_STARTED)
7853                         wait_block_group_cache_done(block_group);
7854
7855                 btrfs_remove_free_space_cache(block_group);
7856                 btrfs_put_block_group(block_group);
7857
7858                 spin_lock(&info->block_group_cache_lock);
7859         }
7860         spin_unlock(&info->block_group_cache_lock);
7861
7862         /* now that all the block groups are freed, go through and
7863          * free all the space_info structs.  This is only called during
7864          * the final stages of unmount, and so we know nobody is
7865          * using them.  We call synchronize_rcu() once before we start,
7866          * just to be on the safe side.
7867          */
7868         synchronize_rcu();
7869
7870         release_global_block_rsv(info);
7871
7872         while(!list_empty(&info->space_info)) {
7873                 space_info = list_entry(info->space_info.next,
7874                                         struct btrfs_space_info,
7875                                         list);
7876                 if (space_info->bytes_pinned > 0 ||
7877                     space_info->bytes_reserved > 0) {
7878                         WARN_ON(1);
7879                         dump_space_info(space_info, 0, 0);
7880                 }
7881                 list_del(&space_info->list);
7882                 kfree(space_info);
7883         }
7884         return 0;
7885 }
7886
7887 static void __link_block_group(struct btrfs_space_info *space_info,
7888                                struct btrfs_block_group_cache *cache)
7889 {
7890         int index = get_block_group_index(cache);
7891
7892         down_write(&space_info->groups_sem);
7893         list_add_tail(&cache->list, &space_info->block_groups[index]);
7894         up_write(&space_info->groups_sem);
7895 }
7896
7897 int btrfs_read_block_groups(struct btrfs_root *root)
7898 {
7899         struct btrfs_path *path;
7900         int ret;
7901         struct btrfs_block_group_cache *cache;
7902         struct btrfs_fs_info *info = root->fs_info;
7903         struct btrfs_space_info *space_info;
7904         struct btrfs_key key;
7905         struct btrfs_key found_key;
7906         struct extent_buffer *leaf;
7907
7908         root = info->extent_root;
7909         key.objectid = 0;
7910         key.offset = 0;
7911         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7912         path = btrfs_alloc_path();
7913         if (!path)
7914                 return -ENOMEM;
7915
7916         while (1) {
7917                 ret = find_first_block_group(root, path, &key);
7918                 if (ret > 0)
7919                         break;
7920                 if (ret != 0)
7921                         goto error;
7922
7923                 leaf = path->nodes[0];
7924                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7925                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7926                 if (!cache) {
7927                         ret = -ENOMEM;
7928                         goto error;
7929                 }
7930
7931                 atomic_set(&cache->count, 1);
7932                 spin_lock_init(&cache->lock);
7933                 spin_lock_init(&cache->tree_lock);
7934                 cache->fs_info = info;
7935                 INIT_LIST_HEAD(&cache->list);
7936                 INIT_LIST_HEAD(&cache->cluster_list);
7937
7938                 /*
7939                  * we only want to have 32k of ram per block group for keeping
7940                  * track of free space, and if we pass 1/2 of that we want to
7941                  * start converting things over to using bitmaps
7942                  */
7943                 cache->extents_thresh = ((1024 * 32) / 2) /
7944                         sizeof(struct btrfs_free_space);
7945
7946                 read_extent_buffer(leaf, &cache->item,
7947                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7948                                    sizeof(cache->item));
7949                 memcpy(&cache->key, &found_key, sizeof(found_key));
7950
7951                 key.objectid = found_key.objectid + found_key.offset;
7952                 btrfs_release_path(root, path);
7953                 cache->flags = btrfs_block_group_flags(&cache->item);
7954                 cache->sectorsize = root->sectorsize;
7955
7956                 /*
7957                  * check for two cases, either we are full, and therefore
7958                  * don't need to bother with the caching work since we won't
7959                  * find any space, or we are empty, and we can just add all
7960                  * the space in and be done with it.  This saves us _alot_ of
7961                  * time, particularly in the full case.
7962                  */
7963                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7964                         exclude_super_stripes(root, cache);
7965                         cache->last_byte_to_unpin = (u64)-1;
7966                         cache->cached = BTRFS_CACHE_FINISHED;
7967                         free_excluded_extents(root, cache);
7968                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7969                         exclude_super_stripes(root, cache);
7970                         cache->last_byte_to_unpin = (u64)-1;
7971                         cache->cached = BTRFS_CACHE_FINISHED;
7972                         add_new_free_space(cache, root->fs_info,
7973                                            found_key.objectid,
7974                                            found_key.objectid +
7975                                            found_key.offset);
7976                         free_excluded_extents(root, cache);
7977                 }
7978
7979                 ret = update_space_info(info, cache->flags, found_key.offset,
7980                                         btrfs_block_group_used(&cache->item),
7981                                         &space_info);
7982                 BUG_ON(ret);
7983                 cache->space_info = space_info;
7984                 spin_lock(&cache->space_info->lock);
7985                 cache->space_info->bytes_readonly += cache->bytes_super;
7986                 spin_unlock(&cache->space_info->lock);
7987
7988                 __link_block_group(space_info, cache);
7989
7990                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7991                 BUG_ON(ret);
7992
7993                 set_avail_alloc_bits(root->fs_info, cache->flags);
7994                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7995                         set_block_group_ro(cache);
7996         }
7997
7998         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7999                 if (!(get_alloc_profile(root, space_info->flags) &
8000                       (BTRFS_BLOCK_GROUP_RAID10 |
8001                        BTRFS_BLOCK_GROUP_RAID1 |
8002                        BTRFS_BLOCK_GROUP_DUP)))
8003                         continue;
8004                 /*
8005                  * avoid allocating from un-mirrored block group if there are
8006                  * mirrored block groups.
8007                  */
8008                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8009                         set_block_group_ro(cache);
8010                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8011                         set_block_group_ro(cache);
8012         }
8013
8014         init_global_block_rsv(info);
8015         ret = 0;
8016 error:
8017         btrfs_free_path(path);
8018         return ret;
8019 }
8020
8021 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8022                            struct btrfs_root *root, u64 bytes_used,
8023                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8024                            u64 size)
8025 {
8026         int ret;
8027         struct btrfs_root *extent_root;
8028         struct btrfs_block_group_cache *cache;
8029
8030         extent_root = root->fs_info->extent_root;
8031
8032         root->fs_info->last_trans_log_full_commit = trans->transid;
8033
8034         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8035         if (!cache)
8036                 return -ENOMEM;
8037
8038         cache->key.objectid = chunk_offset;
8039         cache->key.offset = size;
8040         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8041         cache->sectorsize = root->sectorsize;
8042
8043         /*
8044          * we only want to have 32k of ram per block group for keeping track
8045          * of free space, and if we pass 1/2 of that we want to start
8046          * converting things over to using bitmaps
8047          */
8048         cache->extents_thresh = ((1024 * 32) / 2) /
8049                 sizeof(struct btrfs_free_space);
8050         atomic_set(&cache->count, 1);
8051         spin_lock_init(&cache->lock);
8052         spin_lock_init(&cache->tree_lock);
8053         INIT_LIST_HEAD(&cache->list);
8054         INIT_LIST_HEAD(&cache->cluster_list);
8055
8056         btrfs_set_block_group_used(&cache->item, bytes_used);
8057         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8058         cache->flags = type;
8059         btrfs_set_block_group_flags(&cache->item, type);
8060
8061         cache->last_byte_to_unpin = (u64)-1;
8062         cache->cached = BTRFS_CACHE_FINISHED;
8063         exclude_super_stripes(root, cache);
8064
8065         add_new_free_space(cache, root->fs_info, chunk_offset,
8066                            chunk_offset + size);
8067
8068         free_excluded_extents(root, cache);
8069
8070         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8071                                 &cache->space_info);
8072         BUG_ON(ret);
8073
8074         spin_lock(&cache->space_info->lock);
8075         cache->space_info->bytes_readonly += cache->bytes_super;
8076         spin_unlock(&cache->space_info->lock);
8077
8078         __link_block_group(cache->space_info, cache);
8079
8080         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8081         BUG_ON(ret);
8082
8083         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8084                                 sizeof(cache->item));
8085         BUG_ON(ret);
8086
8087         set_avail_alloc_bits(extent_root->fs_info, type);
8088
8089         return 0;
8090 }
8091
8092 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8093                              struct btrfs_root *root, u64 group_start)
8094 {
8095         struct btrfs_path *path;
8096         struct btrfs_block_group_cache *block_group;
8097         struct btrfs_free_cluster *cluster;
8098         struct btrfs_key key;
8099         int ret;
8100         int factor;
8101
8102         root = root->fs_info->extent_root;
8103
8104         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8105         BUG_ON(!block_group);
8106         BUG_ON(!block_group->ro);
8107
8108         memcpy(&key, &block_group->key, sizeof(key));
8109         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8110                                   BTRFS_BLOCK_GROUP_RAID1 |
8111                                   BTRFS_BLOCK_GROUP_RAID10))
8112                 factor = 2;
8113         else
8114                 factor = 1;
8115
8116         /* make sure this block group isn't part of an allocation cluster */
8117         cluster = &root->fs_info->data_alloc_cluster;
8118         spin_lock(&cluster->refill_lock);
8119         btrfs_return_cluster_to_free_space(block_group, cluster);
8120         spin_unlock(&cluster->refill_lock);
8121
8122         /*
8123          * make sure this block group isn't part of a metadata
8124          * allocation cluster
8125          */
8126         cluster = &root->fs_info->meta_alloc_cluster;
8127         spin_lock(&cluster->refill_lock);
8128         btrfs_return_cluster_to_free_space(block_group, cluster);
8129         spin_unlock(&cluster->refill_lock);
8130
8131         path = btrfs_alloc_path();
8132         BUG_ON(!path);
8133
8134         spin_lock(&root->fs_info->block_group_cache_lock);
8135         rb_erase(&block_group->cache_node,
8136                  &root->fs_info->block_group_cache_tree);
8137         spin_unlock(&root->fs_info->block_group_cache_lock);
8138
8139         down_write(&block_group->space_info->groups_sem);
8140         /*
8141          * we must use list_del_init so people can check to see if they
8142          * are still on the list after taking the semaphore
8143          */
8144         list_del_init(&block_group->list);
8145         up_write(&block_group->space_info->groups_sem);
8146
8147         if (block_group->cached == BTRFS_CACHE_STARTED)
8148                 wait_block_group_cache_done(block_group);
8149
8150         btrfs_remove_free_space_cache(block_group);
8151
8152         spin_lock(&block_group->space_info->lock);
8153         block_group->space_info->total_bytes -= block_group->key.offset;
8154         block_group->space_info->bytes_readonly -= block_group->key.offset;
8155         block_group->space_info->disk_total -= block_group->key.offset * factor;
8156         spin_unlock(&block_group->space_info->lock);
8157
8158         btrfs_clear_space_info_full(root->fs_info);
8159
8160         btrfs_put_block_group(block_group);
8161         btrfs_put_block_group(block_group);
8162
8163         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8164         if (ret > 0)
8165                 ret = -EIO;
8166         if (ret < 0)
8167                 goto out;
8168
8169         ret = btrfs_del_item(trans, root, path);
8170 out:
8171         btrfs_free_path(path);
8172         return ret;
8173 }