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, u64 alloc_bytes)
3004 {
3005         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3006
3007         if (sinfo->bytes_used + sinfo->bytes_reserved +
3008             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3009                 return 0;
3010
3011         if (sinfo->bytes_used + sinfo->bytes_reserved +
3012             alloc_bytes < div_factor(num_bytes, 8))
3013                 return 0;
3014
3015         if (num_bytes > 256 * 1024 * 1024 &&
3016             sinfo->bytes_used < div_factor(num_bytes, 3))
3017                 return 0;
3018
3019         return 1;
3020 }
3021
3022 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3023                           struct btrfs_root *extent_root, u64 alloc_bytes,
3024                           u64 flags, int force)
3025 {
3026         struct btrfs_space_info *space_info;
3027         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3028         int ret = 0;
3029
3030         mutex_lock(&fs_info->chunk_mutex);
3031
3032         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3033
3034         space_info = __find_space_info(extent_root->fs_info, flags);
3035         if (!space_info) {
3036                 ret = update_space_info(extent_root->fs_info, flags,
3037                                         0, 0, &space_info);
3038                 BUG_ON(ret);
3039         }
3040         BUG_ON(!space_info);
3041
3042         spin_lock(&space_info->lock);
3043         if (space_info->force_alloc)
3044                 force = 1;
3045         if (space_info->full) {
3046                 spin_unlock(&space_info->lock);
3047                 goto out;
3048         }
3049
3050         if (!force && !should_alloc_chunk(space_info, alloc_bytes)) {
3051                 spin_unlock(&space_info->lock);
3052                 goto out;
3053         }
3054         spin_unlock(&space_info->lock);
3055
3056         /*
3057          * if we're doing a data chunk, go ahead and make sure that
3058          * we keep a reasonable number of metadata chunks allocated in the
3059          * FS as well.
3060          */
3061         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3062                 fs_info->data_chunk_allocations++;
3063                 if (!(fs_info->data_chunk_allocations %
3064                       fs_info->metadata_ratio))
3065                         force_metadata_allocation(fs_info);
3066         }
3067
3068         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3069         spin_lock(&space_info->lock);
3070         if (ret)
3071                 space_info->full = 1;
3072         else
3073                 ret = 1;
3074         space_info->force_alloc = 0;
3075         spin_unlock(&space_info->lock);
3076 out:
3077         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3078         return ret;
3079 }
3080
3081 static int maybe_allocate_chunk(struct btrfs_trans_handle *trans,
3082                                 struct btrfs_root *root,
3083                                 struct btrfs_space_info *sinfo, u64 num_bytes)
3084 {
3085         int ret;
3086         int end_trans = 0;
3087
3088         if (sinfo->full)
3089                 return 0;
3090
3091         spin_lock(&sinfo->lock);
3092         ret = should_alloc_chunk(sinfo, num_bytes + 2 * 1024 * 1024);
3093         spin_unlock(&sinfo->lock);
3094         if (!ret)
3095                 return 0;
3096
3097         if (!trans) {
3098                 trans = btrfs_join_transaction(root, 1);
3099                 BUG_ON(IS_ERR(trans));
3100                 end_trans = 1;
3101         }
3102
3103         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3104                              num_bytes + 2 * 1024 * 1024,
3105                              get_alloc_profile(root, sinfo->flags), 0);
3106
3107         if (end_trans)
3108                 btrfs_end_transaction(trans, root);
3109
3110         return ret == 1 ? 1 : 0;
3111 }
3112
3113 /*
3114  * shrink metadata reservation for delalloc
3115  */
3116 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3117                            struct btrfs_root *root, u64 to_reclaim, int sync)
3118 {
3119         struct btrfs_block_rsv *block_rsv;
3120         struct btrfs_space_info *space_info;
3121         u64 reserved;
3122         u64 max_reclaim;
3123         u64 reclaimed = 0;
3124         int no_reclaim = 0;
3125         int pause = 1;
3126         int ret;
3127
3128         block_rsv = &root->fs_info->delalloc_block_rsv;
3129         space_info = block_rsv->space_info;
3130         spin_lock(&space_info->lock);
3131         reserved = space_info->bytes_reserved;
3132         spin_unlock(&space_info->lock);
3133
3134         if (reserved == 0)
3135                 return 0;
3136
3137         max_reclaim = min(reserved, to_reclaim);
3138
3139         while (1) {
3140                 ret = btrfs_start_one_delalloc_inode(root, trans ? 1 : 0, sync);
3141                 if (!ret) {
3142                         if (no_reclaim > 2)
3143                                 break;
3144                         no_reclaim++;
3145                         __set_current_state(TASK_INTERRUPTIBLE);
3146                         schedule_timeout(pause);
3147                         pause <<= 1;
3148                         if (pause > HZ / 10)
3149                                 pause = HZ / 10;
3150                 } else {
3151                         no_reclaim = 0;
3152                         pause = 1;
3153                 }
3154
3155                 spin_lock(&space_info->lock);
3156                 if (reserved > space_info->bytes_reserved)
3157                         reclaimed += reserved - space_info->bytes_reserved;
3158                 reserved = space_info->bytes_reserved;
3159                 spin_unlock(&space_info->lock);
3160
3161                 if (reserved == 0 || reclaimed >= max_reclaim)
3162                         break;
3163
3164                 if (trans && trans->transaction->blocked)
3165                         return -EAGAIN;
3166         }
3167         return reclaimed >= to_reclaim;
3168 }
3169
3170 static int should_retry_reserve(struct btrfs_trans_handle *trans,
3171                                 struct btrfs_root *root,
3172                                 struct btrfs_block_rsv *block_rsv,
3173                                 u64 num_bytes, int *retries)
3174 {
3175         struct btrfs_space_info *space_info = block_rsv->space_info;
3176         int ret;
3177
3178         if ((*retries) > 2)
3179                 return -ENOSPC;
3180
3181         ret = maybe_allocate_chunk(trans, root, space_info, num_bytes);
3182         if (ret)
3183                 return 1;
3184
3185         if (trans && trans->transaction->in_commit)
3186                 return -ENOSPC;
3187
3188         ret = shrink_delalloc(trans, root, num_bytes, 0);
3189         if (ret)
3190                 return ret;
3191
3192         spin_lock(&space_info->lock);
3193         if (space_info->bytes_pinned < num_bytes)
3194                 ret = 1;
3195         spin_unlock(&space_info->lock);
3196         if (ret)
3197                 return -ENOSPC;
3198
3199         (*retries)++;
3200
3201         if (trans)
3202                 return -EAGAIN;
3203
3204         trans = btrfs_join_transaction(root, 1);
3205         BUG_ON(IS_ERR(trans));
3206         ret = btrfs_commit_transaction(trans, root);
3207         BUG_ON(ret);
3208
3209         return 1;
3210 }
3211
3212 static int reserve_metadata_bytes(struct btrfs_block_rsv *block_rsv,
3213                                   u64 num_bytes)
3214 {
3215         struct btrfs_space_info *space_info = block_rsv->space_info;
3216         u64 unused;
3217         int ret = -ENOSPC;
3218
3219         spin_lock(&space_info->lock);
3220         unused = space_info->bytes_used + space_info->bytes_reserved +
3221                  space_info->bytes_pinned + space_info->bytes_readonly +
3222                  space_info->bytes_may_use;
3223
3224         if (unused < space_info->total_bytes)
3225                 unused = space_info->total_bytes - unused;
3226         else
3227                 unused = 0;
3228
3229         if (unused >= num_bytes) {
3230                 if (block_rsv->priority >= 10) {
3231                         space_info->bytes_reserved += num_bytes;
3232                         ret = 0;
3233                 } else {
3234                         if ((unused + block_rsv->reserved) *
3235                             block_rsv->priority >=
3236                             (num_bytes + block_rsv->reserved) * 10) {
3237                                 space_info->bytes_reserved += num_bytes;
3238                                 ret = 0;
3239                         }
3240                 }
3241         }
3242         spin_unlock(&space_info->lock);
3243
3244         return ret;
3245 }
3246
3247 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3248                                              struct btrfs_root *root)
3249 {
3250         struct btrfs_block_rsv *block_rsv;
3251         if (root->ref_cows)
3252                 block_rsv = trans->block_rsv;
3253         else
3254                 block_rsv = root->block_rsv;
3255
3256         if (!block_rsv)
3257                 block_rsv = &root->fs_info->empty_block_rsv;
3258
3259         return block_rsv;
3260 }
3261
3262 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3263                                u64 num_bytes)
3264 {
3265         int ret = -ENOSPC;
3266         spin_lock(&block_rsv->lock);
3267         if (block_rsv->reserved >= num_bytes) {
3268                 block_rsv->reserved -= num_bytes;
3269                 if (block_rsv->reserved < block_rsv->size)
3270                         block_rsv->full = 0;
3271                 ret = 0;
3272         }
3273         spin_unlock(&block_rsv->lock);
3274         return ret;
3275 }
3276
3277 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3278                                 u64 num_bytes, int update_size)
3279 {
3280         spin_lock(&block_rsv->lock);
3281         block_rsv->reserved += num_bytes;
3282         if (update_size)
3283                 block_rsv->size += num_bytes;
3284         else if (block_rsv->reserved >= block_rsv->size)
3285                 block_rsv->full = 1;
3286         spin_unlock(&block_rsv->lock);
3287 }
3288
3289 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3290                              struct btrfs_block_rsv *dest, u64 num_bytes)
3291 {
3292         struct btrfs_space_info *space_info = block_rsv->space_info;
3293
3294         spin_lock(&block_rsv->lock);
3295         if (num_bytes == (u64)-1)
3296                 num_bytes = block_rsv->size;
3297         block_rsv->size -= num_bytes;
3298         if (block_rsv->reserved >= block_rsv->size) {
3299                 num_bytes = block_rsv->reserved - block_rsv->size;
3300                 block_rsv->reserved = block_rsv->size;
3301                 block_rsv->full = 1;
3302         } else {
3303                 num_bytes = 0;
3304         }
3305         spin_unlock(&block_rsv->lock);
3306
3307         if (num_bytes > 0) {
3308                 if (dest) {
3309                         block_rsv_add_bytes(dest, num_bytes, 0);
3310                 } else {
3311                         spin_lock(&space_info->lock);
3312                         space_info->bytes_reserved -= num_bytes;
3313                         spin_unlock(&space_info->lock);
3314                 }
3315         }
3316 }
3317
3318 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3319                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3320 {
3321         int ret;
3322
3323         ret = block_rsv_use_bytes(src, num_bytes);
3324         if (ret)
3325                 return ret;
3326
3327         block_rsv_add_bytes(dst, num_bytes, 1);
3328         return 0;
3329 }
3330
3331 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3332 {
3333         memset(rsv, 0, sizeof(*rsv));
3334         spin_lock_init(&rsv->lock);
3335         atomic_set(&rsv->usage, 1);
3336         rsv->priority = 6;
3337         INIT_LIST_HEAD(&rsv->list);
3338 }
3339
3340 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3341 {
3342         struct btrfs_block_rsv *block_rsv;
3343         struct btrfs_fs_info *fs_info = root->fs_info;
3344         u64 alloc_target;
3345
3346         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3347         if (!block_rsv)
3348                 return NULL;
3349
3350         btrfs_init_block_rsv(block_rsv);
3351
3352         alloc_target = btrfs_get_alloc_profile(root, 0);
3353         block_rsv->space_info = __find_space_info(fs_info,
3354                                                   BTRFS_BLOCK_GROUP_METADATA);
3355
3356         return block_rsv;
3357 }
3358
3359 void btrfs_free_block_rsv(struct btrfs_root *root,
3360                           struct btrfs_block_rsv *rsv)
3361 {
3362         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3363                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3364                 if (!rsv->durable)
3365                         kfree(rsv);
3366         }
3367 }
3368
3369 /*
3370  * make the block_rsv struct be able to capture freed space.
3371  * the captured space will re-add to the the block_rsv struct
3372  * after transaction commit
3373  */
3374 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3375                                  struct btrfs_block_rsv *block_rsv)
3376 {
3377         block_rsv->durable = 1;
3378         mutex_lock(&fs_info->durable_block_rsv_mutex);
3379         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3380         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3381 }
3382
3383 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3384                         struct btrfs_root *root,
3385                         struct btrfs_block_rsv *block_rsv,
3386                         u64 num_bytes, int *retries)
3387 {
3388         int ret;
3389
3390         if (num_bytes == 0)
3391                 return 0;
3392 again:
3393         ret = reserve_metadata_bytes(block_rsv, num_bytes);
3394         if (!ret) {
3395                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3396                 return 0;
3397         }
3398
3399         ret = should_retry_reserve(trans, root, block_rsv, num_bytes, retries);
3400         if (ret > 0)
3401                 goto again;
3402
3403         return ret;
3404 }
3405
3406 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3407                           struct btrfs_root *root,
3408                           struct btrfs_block_rsv *block_rsv,
3409                           u64 min_reserved, int min_factor)
3410 {
3411         u64 num_bytes = 0;
3412         int commit_trans = 0;
3413         int ret = -ENOSPC;
3414
3415         if (!block_rsv)
3416                 return 0;
3417
3418         spin_lock(&block_rsv->lock);
3419         if (min_factor > 0)
3420                 num_bytes = div_factor(block_rsv->size, min_factor);
3421         if (min_reserved > num_bytes)
3422                 num_bytes = min_reserved;
3423
3424         if (block_rsv->reserved >= num_bytes) {
3425                 ret = 0;
3426         } else {
3427                 num_bytes -= block_rsv->reserved;
3428                 if (block_rsv->durable &&
3429                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3430                         commit_trans = 1;
3431         }
3432         spin_unlock(&block_rsv->lock);
3433         if (!ret)
3434                 return 0;
3435
3436         if (block_rsv->refill_used) {
3437                 ret = reserve_metadata_bytes(block_rsv, num_bytes);
3438                 if (!ret) {
3439                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3440                         return 0;
3441                 }
3442         }
3443
3444         if (commit_trans) {
3445                 if (trans)
3446                         return -EAGAIN;
3447
3448                 trans = btrfs_join_transaction(root, 1);
3449                 BUG_ON(IS_ERR(trans));
3450                 ret = btrfs_commit_transaction(trans, root);
3451                 return 0;
3452         }
3453
3454         WARN_ON(1);
3455         printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
3456                 block_rsv->size, block_rsv->reserved,
3457                 block_rsv->freed[0], block_rsv->freed[1]);
3458
3459         return -ENOSPC;
3460 }
3461
3462 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3463                             struct btrfs_block_rsv *dst_rsv,
3464                             u64 num_bytes)
3465 {
3466         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3467 }
3468
3469 void btrfs_block_rsv_release(struct btrfs_root *root,
3470                              struct btrfs_block_rsv *block_rsv,
3471                              u64 num_bytes)
3472 {
3473         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3474         if (global_rsv->full || global_rsv == block_rsv ||
3475             block_rsv->space_info != global_rsv->space_info)
3476                 global_rsv = NULL;
3477         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3478 }
3479
3480 /*
3481  * helper to calculate size of global block reservation.
3482  * the desired value is sum of space used by extent tree,
3483  * checksum tree and root tree
3484  */
3485 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3486 {
3487         struct btrfs_space_info *sinfo;
3488         u64 num_bytes;
3489         u64 meta_used;
3490         u64 data_used;
3491         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3492 #if 0
3493         /*
3494          * per tree used space accounting can be inaccuracy, so we
3495          * can't rely on it.
3496          */
3497         spin_lock(&fs_info->extent_root->accounting_lock);
3498         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3499         spin_unlock(&fs_info->extent_root->accounting_lock);
3500
3501         spin_lock(&fs_info->csum_root->accounting_lock);
3502         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3503         spin_unlock(&fs_info->csum_root->accounting_lock);
3504
3505         spin_lock(&fs_info->tree_root->accounting_lock);
3506         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3507         spin_unlock(&fs_info->tree_root->accounting_lock);
3508 #endif
3509         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3510         spin_lock(&sinfo->lock);
3511         data_used = sinfo->bytes_used;
3512         spin_unlock(&sinfo->lock);
3513
3514         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3515         spin_lock(&sinfo->lock);
3516         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3517                 data_used = 0;
3518         meta_used = sinfo->bytes_used;
3519         spin_unlock(&sinfo->lock);
3520
3521         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3522                     csum_size * 2;
3523         num_bytes += div64_u64(data_used + meta_used, 50);
3524
3525         if (num_bytes * 3 > meta_used)
3526                 num_bytes = div64_u64(meta_used, 3);
3527
3528         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3529 }
3530
3531 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3532 {
3533         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3534         struct btrfs_space_info *sinfo = block_rsv->space_info;
3535         u64 num_bytes;
3536
3537         num_bytes = calc_global_metadata_size(fs_info);
3538
3539         spin_lock(&block_rsv->lock);
3540         spin_lock(&sinfo->lock);
3541
3542         block_rsv->size = num_bytes;
3543
3544         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3545                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3546                     sinfo->bytes_may_use;
3547
3548         if (sinfo->total_bytes > num_bytes) {
3549                 num_bytes = sinfo->total_bytes - num_bytes;
3550                 block_rsv->reserved += num_bytes;
3551                 sinfo->bytes_reserved += num_bytes;
3552         }
3553
3554         if (block_rsv->reserved >= block_rsv->size) {
3555                 num_bytes = block_rsv->reserved - block_rsv->size;
3556                 sinfo->bytes_reserved -= num_bytes;
3557                 block_rsv->reserved = block_rsv->size;
3558                 block_rsv->full = 1;
3559         }
3560 #if 0
3561         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3562                 block_rsv->size, block_rsv->reserved);
3563 #endif
3564         spin_unlock(&sinfo->lock);
3565         spin_unlock(&block_rsv->lock);
3566 }
3567
3568 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3569 {
3570         struct btrfs_space_info *space_info;
3571
3572         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3573         fs_info->chunk_block_rsv.space_info = space_info;
3574         fs_info->chunk_block_rsv.priority = 10;
3575
3576         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3577         fs_info->global_block_rsv.space_info = space_info;
3578         fs_info->global_block_rsv.priority = 10;
3579         fs_info->global_block_rsv.refill_used = 1;
3580         fs_info->delalloc_block_rsv.space_info = space_info;
3581         fs_info->trans_block_rsv.space_info = space_info;
3582         fs_info->empty_block_rsv.space_info = space_info;
3583         fs_info->empty_block_rsv.priority = 10;
3584
3585         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3586         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3587         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3588         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3589         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3590
3591         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3592
3593         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3594
3595         update_global_block_rsv(fs_info);
3596 }
3597
3598 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3599 {
3600         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3601         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3602         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3603         WARN_ON(fs_info->trans_block_rsv.size > 0);
3604         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3605         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3606         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3607 }
3608
3609 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3610 {
3611         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3612                 3 * num_items;
3613 }
3614
3615 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3616                                  struct btrfs_root *root,
3617                                  int num_items, int *retries)
3618 {
3619         u64 num_bytes;
3620         int ret;
3621
3622         if (num_items == 0 || root->fs_info->chunk_root == root)
3623                 return 0;
3624
3625         num_bytes = calc_trans_metadata_size(root, num_items);
3626         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3627                                   num_bytes, retries);
3628         if (!ret) {
3629                 trans->bytes_reserved += num_bytes;
3630                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3631         }
3632         return ret;
3633 }
3634
3635 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3636                                   struct btrfs_root *root)
3637 {
3638         if (!trans->bytes_reserved)
3639                 return;
3640
3641         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3642         btrfs_block_rsv_release(root, trans->block_rsv,
3643                                 trans->bytes_reserved);
3644         trans->bytes_reserved = 0;
3645 }
3646
3647 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3648                                   struct inode *inode)
3649 {
3650         struct btrfs_root *root = BTRFS_I(inode)->root;
3651         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3652         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3653
3654         /*
3655          * one for deleting orphan item, one for updating inode and
3656          * two for calling btrfs_truncate_inode_items.
3657          *
3658          * btrfs_truncate_inode_items is a delete operation, it frees
3659          * more space than it uses in most cases. So two units of
3660          * metadata space should be enough for calling it many times.
3661          * If all of the metadata space is used, we can commit
3662          * transaction and use space it freed.
3663          */
3664         u64 num_bytes = calc_trans_metadata_size(root, 4);
3665         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3666 }
3667
3668 void btrfs_orphan_release_metadata(struct inode *inode)
3669 {
3670         struct btrfs_root *root = BTRFS_I(inode)->root;
3671         u64 num_bytes = calc_trans_metadata_size(root, 4);
3672         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3673 }
3674
3675 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3676                                 struct btrfs_pending_snapshot *pending)
3677 {
3678         struct btrfs_root *root = pending->root;
3679         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3680         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3681         /*
3682          * two for root back/forward refs, two for directory entries
3683          * and one for root of the snapshot.
3684          */
3685         u64 num_bytes = calc_trans_metadata_size(root, 5);
3686         dst_rsv->space_info = src_rsv->space_info;
3687         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3688 }
3689
3690 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3691 {
3692         return num_bytes >>= 3;
3693 }
3694
3695 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3696 {
3697         struct btrfs_root *root = BTRFS_I(inode)->root;
3698         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3699         u64 to_reserve;
3700         int nr_extents;
3701         int retries = 0;
3702         int ret;
3703
3704         if (btrfs_transaction_in_commit(root->fs_info))
3705                 schedule_timeout(1);
3706
3707         num_bytes = ALIGN(num_bytes, root->sectorsize);
3708 again:
3709         spin_lock(&BTRFS_I(inode)->accounting_lock);
3710         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
3711         if (nr_extents > BTRFS_I(inode)->reserved_extents) {
3712                 nr_extents -= BTRFS_I(inode)->reserved_extents;
3713                 to_reserve = calc_trans_metadata_size(root, nr_extents);
3714         } else {
3715                 nr_extents = 0;
3716                 to_reserve = 0;
3717         }
3718
3719         to_reserve += calc_csum_metadata_size(inode, num_bytes);
3720         ret = reserve_metadata_bytes(block_rsv, to_reserve);
3721         if (ret) {
3722                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3723                 ret = should_retry_reserve(NULL, root, block_rsv, to_reserve,
3724                                            &retries);
3725                 if (ret > 0)
3726                         goto again;
3727                 return ret;
3728         }
3729
3730         BTRFS_I(inode)->reserved_extents += nr_extents;
3731         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
3732         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3733
3734         block_rsv_add_bytes(block_rsv, to_reserve, 1);
3735
3736         if (block_rsv->size > 512 * 1024 * 1024)
3737                 shrink_delalloc(NULL, root, to_reserve, 0);
3738
3739         return 0;
3740 }
3741
3742 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
3743 {
3744         struct btrfs_root *root = BTRFS_I(inode)->root;
3745         u64 to_free;
3746         int nr_extents;
3747
3748         num_bytes = ALIGN(num_bytes, root->sectorsize);
3749         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
3750
3751         spin_lock(&BTRFS_I(inode)->accounting_lock);
3752         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
3753         if (nr_extents < BTRFS_I(inode)->reserved_extents) {
3754                 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
3755                 BTRFS_I(inode)->reserved_extents -= nr_extents;
3756         } else {
3757                 nr_extents = 0;
3758         }
3759         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3760
3761         to_free = calc_csum_metadata_size(inode, num_bytes);
3762         if (nr_extents > 0)
3763                 to_free += calc_trans_metadata_size(root, nr_extents);
3764
3765         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
3766                                 to_free);
3767 }
3768
3769 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
3770 {
3771         int ret;
3772
3773         ret = btrfs_check_data_free_space(inode, num_bytes);
3774         if (ret)
3775                 return ret;
3776
3777         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
3778         if (ret) {
3779                 btrfs_free_reserved_data_space(inode, num_bytes);
3780                 return ret;
3781         }
3782
3783         return 0;
3784 }
3785
3786 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
3787 {
3788         btrfs_delalloc_release_metadata(inode, num_bytes);
3789         btrfs_free_reserved_data_space(inode, num_bytes);
3790 }
3791
3792 static int update_block_group(struct btrfs_trans_handle *trans,
3793                               struct btrfs_root *root,
3794                               u64 bytenr, u64 num_bytes, int alloc)
3795 {
3796         struct btrfs_block_group_cache *cache;
3797         struct btrfs_fs_info *info = root->fs_info;
3798         int factor;
3799         u64 total = num_bytes;
3800         u64 old_val;
3801         u64 byte_in_group;
3802
3803         /* block accounting for super block */
3804         spin_lock(&info->delalloc_lock);
3805         old_val = btrfs_super_bytes_used(&info->super_copy);
3806         if (alloc)
3807                 old_val += num_bytes;
3808         else
3809                 old_val -= num_bytes;
3810         btrfs_set_super_bytes_used(&info->super_copy, old_val);
3811         spin_unlock(&info->delalloc_lock);
3812
3813         while (total) {
3814                 cache = btrfs_lookup_block_group(info, bytenr);
3815                 if (!cache)
3816                         return -1;
3817                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
3818                                     BTRFS_BLOCK_GROUP_RAID1 |
3819                                     BTRFS_BLOCK_GROUP_RAID10))
3820                         factor = 2;
3821                 else
3822                         factor = 1;
3823                 byte_in_group = bytenr - cache->key.objectid;
3824                 WARN_ON(byte_in_group > cache->key.offset);
3825
3826                 spin_lock(&cache->space_info->lock);
3827                 spin_lock(&cache->lock);
3828                 cache->dirty = 1;
3829                 old_val = btrfs_block_group_used(&cache->item);
3830                 num_bytes = min(total, cache->key.offset - byte_in_group);
3831                 if (alloc) {
3832                         old_val += num_bytes;
3833                         btrfs_set_block_group_used(&cache->item, old_val);
3834                         cache->reserved -= num_bytes;
3835                         cache->space_info->bytes_reserved -= 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                 } else {
3841                         old_val -= num_bytes;
3842                         btrfs_set_block_group_used(&cache->item, old_val);
3843                         cache->pinned += num_bytes;
3844                         cache->space_info->bytes_pinned += num_bytes;
3845                         cache->space_info->bytes_used -= num_bytes;
3846                         cache->space_info->disk_used -= num_bytes * factor;
3847                         spin_unlock(&cache->lock);
3848                         spin_unlock(&cache->space_info->lock);
3849
3850                         set_extent_dirty(info->pinned_extents,
3851                                          bytenr, bytenr + num_bytes - 1,
3852                                          GFP_NOFS | __GFP_NOFAIL);
3853                 }
3854                 btrfs_put_block_group(cache);
3855                 total -= num_bytes;
3856                 bytenr += num_bytes;
3857         }
3858         return 0;
3859 }
3860
3861 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3862 {
3863         struct btrfs_block_group_cache *cache;
3864         u64 bytenr;
3865
3866         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3867         if (!cache)
3868                 return 0;
3869
3870         bytenr = cache->key.objectid;
3871         btrfs_put_block_group(cache);
3872
3873         return bytenr;
3874 }
3875
3876 static int pin_down_extent(struct btrfs_root *root,
3877                            struct btrfs_block_group_cache *cache,
3878                            u64 bytenr, u64 num_bytes, int reserved)
3879 {
3880         spin_lock(&cache->space_info->lock);
3881         spin_lock(&cache->lock);
3882         cache->pinned += num_bytes;
3883         cache->space_info->bytes_pinned += num_bytes;
3884         if (reserved) {
3885                 cache->reserved -= num_bytes;
3886                 cache->space_info->bytes_reserved -= num_bytes;
3887         }
3888         spin_unlock(&cache->lock);
3889         spin_unlock(&cache->space_info->lock);
3890
3891         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
3892                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
3893         return 0;
3894 }
3895
3896 /*
3897  * this function must be called within transaction
3898  */
3899 int btrfs_pin_extent(struct btrfs_root *root,
3900                      u64 bytenr, u64 num_bytes, int reserved)
3901 {
3902         struct btrfs_block_group_cache *cache;
3903
3904         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
3905         BUG_ON(!cache);
3906
3907         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
3908
3909         btrfs_put_block_group(cache);
3910         return 0;
3911 }
3912
3913 /*
3914  * update size of reserved extents. this function may return -EAGAIN
3915  * if 'reserve' is true or 'sinfo' is false.
3916  */
3917 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
3918                                  u64 num_bytes, int reserve, int sinfo)
3919 {
3920         int ret = 0;
3921         if (sinfo) {
3922                 struct btrfs_space_info *space_info = cache->space_info;
3923                 spin_lock(&space_info->lock);
3924                 spin_lock(&cache->lock);
3925                 if (reserve) {
3926                         if (cache->ro) {
3927                                 ret = -EAGAIN;
3928                         } else {
3929                                 cache->reserved += num_bytes;
3930                                 space_info->bytes_reserved += num_bytes;
3931                         }
3932                 } else {
3933                         if (cache->ro)
3934                                 space_info->bytes_readonly += num_bytes;
3935                         cache->reserved -= num_bytes;
3936                         space_info->bytes_reserved -= num_bytes;
3937                 }
3938                 spin_unlock(&cache->lock);
3939                 spin_unlock(&space_info->lock);
3940         } else {
3941                 spin_lock(&cache->lock);
3942                 if (cache->ro) {
3943                         ret = -EAGAIN;
3944                 } else {
3945                         if (reserve)
3946                                 cache->reserved += num_bytes;
3947                         else
3948                                 cache->reserved -= num_bytes;
3949                 }
3950                 spin_unlock(&cache->lock);
3951         }
3952         return ret;
3953 }
3954
3955 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3956                                 struct btrfs_root *root)
3957 {
3958         struct btrfs_fs_info *fs_info = root->fs_info;
3959         struct btrfs_caching_control *next;
3960         struct btrfs_caching_control *caching_ctl;
3961         struct btrfs_block_group_cache *cache;
3962
3963         down_write(&fs_info->extent_commit_sem);
3964
3965         list_for_each_entry_safe(caching_ctl, next,
3966                                  &fs_info->caching_block_groups, list) {
3967                 cache = caching_ctl->block_group;
3968                 if (block_group_cache_done(cache)) {
3969                         cache->last_byte_to_unpin = (u64)-1;
3970                         list_del_init(&caching_ctl->list);
3971                         put_caching_control(caching_ctl);
3972                 } else {
3973                         cache->last_byte_to_unpin = caching_ctl->progress;
3974                 }
3975         }
3976
3977         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3978                 fs_info->pinned_extents = &fs_info->freed_extents[1];
3979         else
3980                 fs_info->pinned_extents = &fs_info->freed_extents[0];
3981
3982         up_write(&fs_info->extent_commit_sem);
3983
3984         update_global_block_rsv(fs_info);
3985         return 0;
3986 }
3987
3988 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3989 {
3990         struct btrfs_fs_info *fs_info = root->fs_info;
3991         struct btrfs_block_group_cache *cache = NULL;
3992         u64 len;
3993
3994         while (start <= end) {
3995                 if (!cache ||
3996                     start >= cache->key.objectid + cache->key.offset) {
3997                         if (cache)
3998                                 btrfs_put_block_group(cache);
3999                         cache = btrfs_lookup_block_group(fs_info, start);
4000                         BUG_ON(!cache);
4001                 }
4002
4003                 len = cache->key.objectid + cache->key.offset - start;
4004                 len = min(len, end + 1 - start);
4005
4006                 if (start < cache->last_byte_to_unpin) {
4007                         len = min(len, cache->last_byte_to_unpin - start);
4008                         btrfs_add_free_space(cache, start, len);
4009                 }
4010
4011                 start += len;
4012
4013                 spin_lock(&cache->space_info->lock);
4014                 spin_lock(&cache->lock);
4015                 cache->pinned -= len;
4016                 cache->space_info->bytes_pinned -= len;
4017                 if (cache->ro) {
4018                         cache->space_info->bytes_readonly += len;
4019                 } else if (cache->reserved_pinned > 0) {
4020                         len = min(len, cache->reserved_pinned);
4021                         cache->reserved_pinned -= len;
4022                         cache->space_info->bytes_reserved += len;
4023                 }
4024                 spin_unlock(&cache->lock);
4025                 spin_unlock(&cache->space_info->lock);
4026         }
4027
4028         if (cache)
4029                 btrfs_put_block_group(cache);
4030         return 0;
4031 }
4032
4033 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4034                                struct btrfs_root *root)
4035 {
4036         struct btrfs_fs_info *fs_info = root->fs_info;
4037         struct extent_io_tree *unpin;
4038         struct btrfs_block_rsv *block_rsv;
4039         struct btrfs_block_rsv *next_rsv;
4040         u64 start;
4041         u64 end;
4042         int idx;
4043         int ret;
4044
4045         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4046                 unpin = &fs_info->freed_extents[1];
4047         else
4048                 unpin = &fs_info->freed_extents[0];
4049
4050         while (1) {
4051                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4052                                             EXTENT_DIRTY);
4053                 if (ret)
4054                         break;
4055
4056                 ret = btrfs_discard_extent(root, start, end + 1 - start);
4057
4058                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4059                 unpin_extent_range(root, start, end);
4060                 cond_resched();
4061         }
4062
4063         mutex_lock(&fs_info->durable_block_rsv_mutex);
4064         list_for_each_entry_safe(block_rsv, next_rsv,
4065                                  &fs_info->durable_block_rsv_list, list) {
4066
4067                 idx = trans->transid & 0x1;
4068                 if (block_rsv->freed[idx] > 0) {
4069                         block_rsv_add_bytes(block_rsv,
4070                                             block_rsv->freed[idx], 0);
4071                         block_rsv->freed[idx] = 0;
4072                 }
4073                 if (atomic_read(&block_rsv->usage) == 0) {
4074                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4075
4076                         if (block_rsv->freed[0] == 0 &&
4077                             block_rsv->freed[1] == 0) {
4078                                 list_del_init(&block_rsv->list);
4079                                 kfree(block_rsv);
4080                         }
4081                 } else {
4082                         btrfs_block_rsv_release(root, block_rsv, 0);
4083                 }
4084         }
4085         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4086
4087         return 0;
4088 }
4089
4090 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4091                                 struct btrfs_root *root,
4092                                 u64 bytenr, u64 num_bytes, u64 parent,
4093                                 u64 root_objectid, u64 owner_objectid,
4094                                 u64 owner_offset, int refs_to_drop,
4095                                 struct btrfs_delayed_extent_op *extent_op)
4096 {
4097         struct btrfs_key key;
4098         struct btrfs_path *path;
4099         struct btrfs_fs_info *info = root->fs_info;
4100         struct btrfs_root *extent_root = info->extent_root;
4101         struct extent_buffer *leaf;
4102         struct btrfs_extent_item *ei;
4103         struct btrfs_extent_inline_ref *iref;
4104         int ret;
4105         int is_data;
4106         int extent_slot = 0;
4107         int found_extent = 0;
4108         int num_to_del = 1;
4109         u32 item_size;
4110         u64 refs;
4111
4112         path = btrfs_alloc_path();
4113         if (!path)
4114                 return -ENOMEM;
4115
4116         path->reada = 1;
4117         path->leave_spinning = 1;
4118
4119         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4120         BUG_ON(!is_data && refs_to_drop != 1);
4121
4122         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4123                                     bytenr, num_bytes, parent,
4124                                     root_objectid, owner_objectid,
4125                                     owner_offset);
4126         if (ret == 0) {
4127                 extent_slot = path->slots[0];
4128                 while (extent_slot >= 0) {
4129                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4130                                               extent_slot);
4131                         if (key.objectid != bytenr)
4132                                 break;
4133                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4134                             key.offset == num_bytes) {
4135                                 found_extent = 1;
4136                                 break;
4137                         }
4138                         if (path->slots[0] - extent_slot > 5)
4139                                 break;
4140                         extent_slot--;
4141                 }
4142 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4143                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4144                 if (found_extent && item_size < sizeof(*ei))
4145                         found_extent = 0;
4146 #endif
4147                 if (!found_extent) {
4148                         BUG_ON(iref);
4149                         ret = remove_extent_backref(trans, extent_root, path,
4150                                                     NULL, refs_to_drop,
4151                                                     is_data);
4152                         BUG_ON(ret);
4153                         btrfs_release_path(extent_root, path);
4154                         path->leave_spinning = 1;
4155
4156                         key.objectid = bytenr;
4157                         key.type = BTRFS_EXTENT_ITEM_KEY;
4158                         key.offset = num_bytes;
4159
4160                         ret = btrfs_search_slot(trans, extent_root,
4161                                                 &key, path, -1, 1);
4162                         if (ret) {
4163                                 printk(KERN_ERR "umm, got %d back from search"
4164                                        ", was looking for %llu\n", ret,
4165                                        (unsigned long long)bytenr);
4166                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4167                         }
4168                         BUG_ON(ret);
4169                         extent_slot = path->slots[0];
4170                 }
4171         } else {
4172                 btrfs_print_leaf(extent_root, path->nodes[0]);
4173                 WARN_ON(1);
4174                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4175                        "parent %llu root %llu  owner %llu offset %llu\n",
4176                        (unsigned long long)bytenr,
4177                        (unsigned long long)parent,
4178                        (unsigned long long)root_objectid,
4179                        (unsigned long long)owner_objectid,
4180                        (unsigned long long)owner_offset);
4181         }
4182
4183         leaf = path->nodes[0];
4184         item_size = btrfs_item_size_nr(leaf, extent_slot);
4185 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4186         if (item_size < sizeof(*ei)) {
4187                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4188                 ret = convert_extent_item_v0(trans, extent_root, path,
4189                                              owner_objectid, 0);
4190                 BUG_ON(ret < 0);
4191
4192                 btrfs_release_path(extent_root, path);
4193                 path->leave_spinning = 1;
4194
4195                 key.objectid = bytenr;
4196                 key.type = BTRFS_EXTENT_ITEM_KEY;
4197                 key.offset = num_bytes;
4198
4199                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4200                                         -1, 1);
4201                 if (ret) {
4202                         printk(KERN_ERR "umm, got %d back from search"
4203                                ", was looking for %llu\n", ret,
4204                                (unsigned long long)bytenr);
4205                         btrfs_print_leaf(extent_root, path->nodes[0]);
4206                 }
4207                 BUG_ON(ret);
4208                 extent_slot = path->slots[0];
4209                 leaf = path->nodes[0];
4210                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4211         }
4212 #endif
4213         BUG_ON(item_size < sizeof(*ei));
4214         ei = btrfs_item_ptr(leaf, extent_slot,
4215                             struct btrfs_extent_item);
4216         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4217                 struct btrfs_tree_block_info *bi;
4218                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4219                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4220                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4221         }
4222
4223         refs = btrfs_extent_refs(leaf, ei);
4224         BUG_ON(refs < refs_to_drop);
4225         refs -= refs_to_drop;
4226
4227         if (refs > 0) {
4228                 if (extent_op)
4229                         __run_delayed_extent_op(extent_op, leaf, ei);
4230                 /*
4231                  * In the case of inline back ref, reference count will
4232                  * be updated by remove_extent_backref
4233                  */
4234                 if (iref) {
4235                         BUG_ON(!found_extent);
4236                 } else {
4237                         btrfs_set_extent_refs(leaf, ei, refs);
4238                         btrfs_mark_buffer_dirty(leaf);
4239                 }
4240                 if (found_extent) {
4241                         ret = remove_extent_backref(trans, extent_root, path,
4242                                                     iref, refs_to_drop,
4243                                                     is_data);
4244                         BUG_ON(ret);
4245                 }
4246         } else {
4247                 if (found_extent) {
4248                         BUG_ON(is_data && refs_to_drop !=
4249                                extent_data_ref_count(root, path, iref));
4250                         if (iref) {
4251                                 BUG_ON(path->slots[0] != extent_slot);
4252                         } else {
4253                                 BUG_ON(path->slots[0] != extent_slot + 1);
4254                                 path->slots[0] = extent_slot;
4255                                 num_to_del = 2;
4256                         }
4257                 }
4258
4259                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4260                                       num_to_del);
4261                 BUG_ON(ret);
4262                 btrfs_release_path(extent_root, path);
4263
4264                 if (is_data) {
4265                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4266                         BUG_ON(ret);
4267                 } else {
4268                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4269                              bytenr >> PAGE_CACHE_SHIFT,
4270                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4271                 }
4272
4273                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4274                 BUG_ON(ret);
4275         }
4276         btrfs_free_path(path);
4277         return ret;
4278 }
4279
4280 /*
4281  * when we free an block, it is possible (and likely) that we free the last
4282  * delayed ref for that extent as well.  This searches the delayed ref tree for
4283  * a given extent, and if there are no other delayed refs to be processed, it
4284  * removes it from the tree.
4285  */
4286 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4287                                       struct btrfs_root *root, u64 bytenr)
4288 {
4289         struct btrfs_delayed_ref_head *head;
4290         struct btrfs_delayed_ref_root *delayed_refs;
4291         struct btrfs_delayed_ref_node *ref;
4292         struct rb_node *node;
4293         int ret = 0;
4294
4295         delayed_refs = &trans->transaction->delayed_refs;
4296         spin_lock(&delayed_refs->lock);
4297         head = btrfs_find_delayed_ref_head(trans, bytenr);
4298         if (!head)
4299                 goto out;
4300
4301         node = rb_prev(&head->node.rb_node);
4302         if (!node)
4303                 goto out;
4304
4305         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4306
4307         /* there are still entries for this ref, we can't drop it */
4308         if (ref->bytenr == bytenr)
4309                 goto out;
4310
4311         if (head->extent_op) {
4312                 if (!head->must_insert_reserved)
4313                         goto out;
4314                 kfree(head->extent_op);
4315                 head->extent_op = NULL;
4316         }
4317
4318         /*
4319          * waiting for the lock here would deadlock.  If someone else has it
4320          * locked they are already in the process of dropping it anyway
4321          */
4322         if (!mutex_trylock(&head->mutex))
4323                 goto out;
4324
4325         /*
4326          * at this point we have a head with no other entries.  Go
4327          * ahead and process it.
4328          */
4329         head->node.in_tree = 0;
4330         rb_erase(&head->node.rb_node, &delayed_refs->root);
4331
4332         delayed_refs->num_entries--;
4333
4334         /*
4335          * we don't take a ref on the node because we're removing it from the
4336          * tree, so we just steal the ref the tree was holding.
4337          */
4338         delayed_refs->num_heads--;
4339         if (list_empty(&head->cluster))
4340                 delayed_refs->num_heads_ready--;
4341
4342         list_del_init(&head->cluster);
4343         spin_unlock(&delayed_refs->lock);
4344
4345         BUG_ON(head->extent_op);
4346         if (head->must_insert_reserved)
4347                 ret = 1;
4348
4349         mutex_unlock(&head->mutex);
4350         btrfs_put_delayed_ref(&head->node);
4351         return ret;
4352 out:
4353         spin_unlock(&delayed_refs->lock);
4354         return 0;
4355 }
4356
4357 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4358                            struct btrfs_root *root,
4359                            struct extent_buffer *buf,
4360                            u64 parent, int last_ref)
4361 {
4362         struct btrfs_block_rsv *block_rsv;
4363         struct btrfs_block_group_cache *cache = NULL;
4364         int ret;
4365
4366         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4367                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4368                                                 parent, root->root_key.objectid,
4369                                                 btrfs_header_level(buf),
4370                                                 BTRFS_DROP_DELAYED_REF, NULL);
4371                 BUG_ON(ret);
4372         }
4373
4374         if (!last_ref)
4375                 return;
4376
4377         block_rsv = get_block_rsv(trans, root);
4378         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4379         if (block_rsv->space_info != cache->space_info)
4380                 goto out;
4381
4382         if (btrfs_header_generation(buf) == trans->transid) {
4383                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4384                         ret = check_ref_cleanup(trans, root, buf->start);
4385                         if (!ret)
4386                                 goto pin;
4387                 }
4388
4389                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4390                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4391                         goto pin;
4392                 }
4393
4394                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4395
4396                 btrfs_add_free_space(cache, buf->start, buf->len);
4397                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4398                 if (ret == -EAGAIN) {
4399                         /* block group became read-only */
4400                         update_reserved_bytes(cache, buf->len, 0, 1);
4401                         goto out;
4402                 }
4403
4404                 ret = 1;
4405                 spin_lock(&block_rsv->lock);
4406                 if (block_rsv->reserved < block_rsv->size) {
4407                         block_rsv->reserved += buf->len;
4408                         ret = 0;
4409                 }
4410                 spin_unlock(&block_rsv->lock);
4411
4412                 if (ret) {
4413                         spin_lock(&cache->space_info->lock);
4414                         cache->space_info->bytes_reserved -= buf->len;
4415                         spin_unlock(&cache->space_info->lock);
4416                 }
4417                 goto out;
4418         }
4419 pin:
4420         if (block_rsv->durable && !cache->ro) {
4421                 ret = 0;
4422                 spin_lock(&cache->lock);
4423                 if (!cache->ro) {
4424                         cache->reserved_pinned += buf->len;
4425                         ret = 1;
4426                 }
4427                 spin_unlock(&cache->lock);
4428
4429                 if (ret) {
4430                         spin_lock(&block_rsv->lock);
4431                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4432                         spin_unlock(&block_rsv->lock);
4433                 }
4434         }
4435 out:
4436         btrfs_put_block_group(cache);
4437 }
4438
4439 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4440                       struct btrfs_root *root,
4441                       u64 bytenr, u64 num_bytes, u64 parent,
4442                       u64 root_objectid, u64 owner, u64 offset)
4443 {
4444         int ret;
4445
4446         /*
4447          * tree log blocks never actually go into the extent allocation
4448          * tree, just update pinning info and exit early.
4449          */
4450         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4451                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4452                 /* unlocks the pinned mutex */
4453                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4454                 ret = 0;
4455         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4456                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4457                                         parent, root_objectid, (int)owner,
4458                                         BTRFS_DROP_DELAYED_REF, NULL);
4459                 BUG_ON(ret);
4460         } else {
4461                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4462                                         parent, root_objectid, owner,
4463                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4464                 BUG_ON(ret);
4465         }
4466         return ret;
4467 }
4468
4469 static u64 stripe_align(struct btrfs_root *root, u64 val)
4470 {
4471         u64 mask = ((u64)root->stripesize - 1);
4472         u64 ret = (val + mask) & ~mask;
4473         return ret;
4474 }
4475
4476 /*
4477  * when we wait for progress in the block group caching, its because
4478  * our allocation attempt failed at least once.  So, we must sleep
4479  * and let some progress happen before we try again.
4480  *
4481  * This function will sleep at least once waiting for new free space to
4482  * show up, and then it will check the block group free space numbers
4483  * for our min num_bytes.  Another option is to have it go ahead
4484  * and look in the rbtree for a free extent of a given size, but this
4485  * is a good start.
4486  */
4487 static noinline int
4488 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4489                                 u64 num_bytes)
4490 {
4491         struct btrfs_caching_control *caching_ctl;
4492         DEFINE_WAIT(wait);
4493
4494         caching_ctl = get_caching_control(cache);
4495         if (!caching_ctl)
4496                 return 0;
4497
4498         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4499                    (cache->free_space >= num_bytes));
4500
4501         put_caching_control(caching_ctl);
4502         return 0;
4503 }
4504
4505 static noinline int
4506 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4507 {
4508         struct btrfs_caching_control *caching_ctl;
4509         DEFINE_WAIT(wait);
4510
4511         caching_ctl = get_caching_control(cache);
4512         if (!caching_ctl)
4513                 return 0;
4514
4515         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4516
4517         put_caching_control(caching_ctl);
4518         return 0;
4519 }
4520
4521 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4522 {
4523         int index;
4524         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4525                 index = 0;
4526         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4527                 index = 1;
4528         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4529                 index = 2;
4530         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4531                 index = 3;
4532         else
4533                 index = 4;
4534         return index;
4535 }
4536
4537 enum btrfs_loop_type {
4538         LOOP_FIND_IDEAL = 0,
4539         LOOP_CACHING_NOWAIT = 1,
4540         LOOP_CACHING_WAIT = 2,
4541         LOOP_ALLOC_CHUNK = 3,
4542         LOOP_NO_EMPTY_SIZE = 4,
4543 };
4544
4545 /*
4546  * walks the btree of allocated extents and find a hole of a given size.
4547  * The key ins is changed to record the hole:
4548  * ins->objectid == block start
4549  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4550  * ins->offset == number of blocks
4551  * Any available blocks before search_start are skipped.
4552  */
4553 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4554                                      struct btrfs_root *orig_root,
4555                                      u64 num_bytes, u64 empty_size,
4556                                      u64 search_start, u64 search_end,
4557                                      u64 hint_byte, struct btrfs_key *ins,
4558                                      int data)
4559 {
4560         int ret = 0;
4561         struct btrfs_root *root = orig_root->fs_info->extent_root;
4562         struct btrfs_free_cluster *last_ptr = NULL;
4563         struct btrfs_block_group_cache *block_group = NULL;
4564         int empty_cluster = 2 * 1024 * 1024;
4565         int allowed_chunk_alloc = 0;
4566         int done_chunk_alloc = 0;
4567         struct btrfs_space_info *space_info;
4568         int last_ptr_loop = 0;
4569         int loop = 0;
4570         int index = 0;
4571         bool found_uncached_bg = false;
4572         bool failed_cluster_refill = false;
4573         bool failed_alloc = false;
4574         u64 ideal_cache_percent = 0;
4575         u64 ideal_cache_offset = 0;
4576
4577         WARN_ON(num_bytes < root->sectorsize);
4578         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4579         ins->objectid = 0;
4580         ins->offset = 0;
4581
4582         space_info = __find_space_info(root->fs_info, data);
4583         if (!space_info) {
4584                 printk(KERN_ERR "No space info for %d\n", data);
4585                 return -ENOSPC;
4586         }
4587
4588         if (orig_root->ref_cows || empty_size)
4589                 allowed_chunk_alloc = 1;
4590
4591         if (data & BTRFS_BLOCK_GROUP_METADATA) {
4592                 last_ptr = &root->fs_info->meta_alloc_cluster;
4593                 if (!btrfs_test_opt(root, SSD))
4594                         empty_cluster = 64 * 1024;
4595         }
4596
4597         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4598                 last_ptr = &root->fs_info->data_alloc_cluster;
4599         }
4600
4601         if (last_ptr) {
4602                 spin_lock(&last_ptr->lock);
4603                 if (last_ptr->block_group)
4604                         hint_byte = last_ptr->window_start;
4605                 spin_unlock(&last_ptr->lock);
4606         }
4607
4608         search_start = max(search_start, first_logical_byte(root, 0));
4609         search_start = max(search_start, hint_byte);
4610
4611         if (!last_ptr)
4612                 empty_cluster = 0;
4613
4614         if (search_start == hint_byte) {
4615 ideal_cache:
4616                 block_group = btrfs_lookup_block_group(root->fs_info,
4617                                                        search_start);
4618                 /*
4619                  * we don't want to use the block group if it doesn't match our
4620                  * allocation bits, or if its not cached.
4621                  *
4622                  * However if we are re-searching with an ideal block group
4623                  * picked out then we don't care that the block group is cached.
4624                  */
4625                 if (block_group && block_group_bits(block_group, data) &&
4626                     (block_group->cached != BTRFS_CACHE_NO ||
4627                      search_start == ideal_cache_offset)) {
4628                         down_read(&space_info->groups_sem);
4629                         if (list_empty(&block_group->list) ||
4630                             block_group->ro) {
4631                                 /*
4632                                  * someone is removing this block group,
4633                                  * we can't jump into the have_block_group
4634                                  * target because our list pointers are not
4635                                  * valid
4636                                  */
4637                                 btrfs_put_block_group(block_group);
4638                                 up_read(&space_info->groups_sem);
4639                         } else {
4640                                 index = get_block_group_index(block_group);
4641                                 goto have_block_group;
4642                         }
4643                 } else if (block_group) {
4644                         btrfs_put_block_group(block_group);
4645                 }
4646         }
4647 search:
4648         down_read(&space_info->groups_sem);
4649         list_for_each_entry(block_group, &space_info->block_groups[index],
4650                             list) {
4651                 u64 offset;
4652                 int cached;
4653
4654                 btrfs_get_block_group(block_group);
4655                 search_start = block_group->key.objectid;
4656
4657 have_block_group:
4658                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4659                         u64 free_percent;
4660
4661                         free_percent = btrfs_block_group_used(&block_group->item);
4662                         free_percent *= 100;
4663                         free_percent = div64_u64(free_percent,
4664                                                  block_group->key.offset);
4665                         free_percent = 100 - free_percent;
4666                         if (free_percent > ideal_cache_percent &&
4667                             likely(!block_group->ro)) {
4668                                 ideal_cache_offset = block_group->key.objectid;
4669                                 ideal_cache_percent = free_percent;
4670                         }
4671
4672                         /*
4673                          * We only want to start kthread caching if we are at
4674                          * the point where we will wait for caching to make
4675                          * progress, or if our ideal search is over and we've
4676                          * found somebody to start caching.
4677                          */
4678                         if (loop > LOOP_CACHING_NOWAIT ||
4679                             (loop > LOOP_FIND_IDEAL &&
4680                              atomic_read(&space_info->caching_threads) < 2)) {
4681                                 ret = cache_block_group(block_group);
4682                                 BUG_ON(ret);
4683                         }
4684                         found_uncached_bg = true;
4685
4686                         /*
4687                          * If loop is set for cached only, try the next block
4688                          * group.
4689                          */
4690                         if (loop == LOOP_FIND_IDEAL)
4691                                 goto loop;
4692                 }
4693
4694                 cached = block_group_cache_done(block_group);
4695                 if (unlikely(!cached))
4696                         found_uncached_bg = true;
4697
4698                 if (unlikely(block_group->ro))
4699                         goto loop;
4700
4701                 /*
4702                  * Ok we want to try and use the cluster allocator, so lets look
4703                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4704                  * have tried the cluster allocator plenty of times at this
4705                  * point and not have found anything, so we are likely way too
4706                  * fragmented for the clustering stuff to find anything, so lets
4707                  * just skip it and let the allocator find whatever block it can
4708                  * find
4709                  */
4710                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4711                         /*
4712                          * the refill lock keeps out other
4713                          * people trying to start a new cluster
4714                          */
4715                         spin_lock(&last_ptr->refill_lock);
4716                         if (last_ptr->block_group &&
4717                             (last_ptr->block_group->ro ||
4718                             !block_group_bits(last_ptr->block_group, data))) {
4719                                 offset = 0;
4720                                 goto refill_cluster;
4721                         }
4722
4723                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4724                                                  num_bytes, search_start);
4725                         if (offset) {
4726                                 /* we have a block, we're done */
4727                                 spin_unlock(&last_ptr->refill_lock);
4728                                 goto checks;
4729                         }
4730
4731                         spin_lock(&last_ptr->lock);
4732                         /*
4733                          * whoops, this cluster doesn't actually point to
4734                          * this block group.  Get a ref on the block
4735                          * group is does point to and try again
4736                          */
4737                         if (!last_ptr_loop && last_ptr->block_group &&
4738                             last_ptr->block_group != block_group) {
4739
4740                                 btrfs_put_block_group(block_group);
4741                                 block_group = last_ptr->block_group;
4742                                 btrfs_get_block_group(block_group);
4743                                 spin_unlock(&last_ptr->lock);
4744                                 spin_unlock(&last_ptr->refill_lock);
4745
4746                                 last_ptr_loop = 1;
4747                                 search_start = block_group->key.objectid;
4748                                 /*
4749                                  * we know this block group is properly
4750                                  * in the list because
4751                                  * btrfs_remove_block_group, drops the
4752                                  * cluster before it removes the block
4753                                  * group from the list
4754                                  */
4755                                 goto have_block_group;
4756                         }
4757                         spin_unlock(&last_ptr->lock);
4758 refill_cluster:
4759                         /*
4760                          * this cluster didn't work out, free it and
4761                          * start over
4762                          */
4763                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4764
4765                         last_ptr_loop = 0;
4766
4767                         /* allocate a cluster in this block group */
4768                         ret = btrfs_find_space_cluster(trans, root,
4769                                                block_group, last_ptr,
4770                                                offset, num_bytes,
4771                                                empty_cluster + empty_size);
4772                         if (ret == 0) {
4773                                 /*
4774                                  * now pull our allocation out of this
4775                                  * cluster
4776                                  */
4777                                 offset = btrfs_alloc_from_cluster(block_group,
4778                                                   last_ptr, num_bytes,
4779                                                   search_start);
4780                                 if (offset) {
4781                                         /* we found one, proceed */
4782                                         spin_unlock(&last_ptr->refill_lock);
4783                                         goto checks;
4784                                 }
4785                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
4786                                    && !failed_cluster_refill) {
4787                                 spin_unlock(&last_ptr->refill_lock);
4788
4789                                 failed_cluster_refill = true;
4790                                 wait_block_group_cache_progress(block_group,
4791                                        num_bytes + empty_cluster + empty_size);
4792                                 goto have_block_group;
4793                         }
4794
4795                         /*
4796                          * at this point we either didn't find a cluster
4797                          * or we weren't able to allocate a block from our
4798                          * cluster.  Free the cluster we've been trying
4799                          * to use, and go to the next block group
4800                          */
4801                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4802                         spin_unlock(&last_ptr->refill_lock);
4803                         goto loop;
4804                 }
4805
4806                 offset = btrfs_find_space_for_alloc(block_group, search_start,
4807                                                     num_bytes, empty_size);
4808                 /*
4809                  * If we didn't find a chunk, and we haven't failed on this
4810                  * block group before, and this block group is in the middle of
4811                  * caching and we are ok with waiting, then go ahead and wait
4812                  * for progress to be made, and set failed_alloc to true.
4813                  *
4814                  * If failed_alloc is true then we've already waited on this
4815                  * block group once and should move on to the next block group.
4816                  */
4817                 if (!offset && !failed_alloc && !cached &&
4818                     loop > LOOP_CACHING_NOWAIT) {
4819                         wait_block_group_cache_progress(block_group,
4820                                                 num_bytes + empty_size);
4821                         failed_alloc = true;
4822                         goto have_block_group;
4823                 } else if (!offset) {
4824                         goto loop;
4825                 }
4826 checks:
4827                 search_start = stripe_align(root, offset);
4828                 /* move on to the next group */
4829                 if (search_start + num_bytes >= search_end) {
4830                         btrfs_add_free_space(block_group, offset, num_bytes);
4831                         goto loop;
4832                 }
4833
4834                 /* move on to the next group */
4835                 if (search_start + num_bytes >
4836                     block_group->key.objectid + block_group->key.offset) {
4837                         btrfs_add_free_space(block_group, offset, num_bytes);
4838                         goto loop;
4839                 }
4840
4841                 ins->objectid = search_start;
4842                 ins->offset = num_bytes;
4843
4844                 if (offset < search_start)
4845                         btrfs_add_free_space(block_group, offset,
4846                                              search_start - offset);
4847                 BUG_ON(offset > search_start);
4848
4849                 ret = update_reserved_bytes(block_group, num_bytes, 1,
4850                                             (data & BTRFS_BLOCK_GROUP_DATA));
4851                 if (ret == -EAGAIN) {
4852                         btrfs_add_free_space(block_group, offset, num_bytes);
4853                         goto loop;
4854                 }
4855
4856                 /* we are all good, lets return */
4857                 ins->objectid = search_start;
4858                 ins->offset = num_bytes;
4859
4860                 if (offset < search_start)
4861                         btrfs_add_free_space(block_group, offset,
4862                                              search_start - offset);
4863                 BUG_ON(offset > search_start);
4864                 break;
4865 loop:
4866                 failed_cluster_refill = false;
4867                 failed_alloc = false;
4868                 BUG_ON(index != get_block_group_index(block_group));
4869                 btrfs_put_block_group(block_group);
4870         }
4871         up_read(&space_info->groups_sem);
4872
4873         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
4874                 goto search;
4875
4876         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
4877          *                      for them to make caching progress.  Also
4878          *                      determine the best possible bg to cache
4879          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4880          *                      caching kthreads as we move along
4881          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4882          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4883          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4884          *                      again
4885          */
4886         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4887             (found_uncached_bg || empty_size || empty_cluster ||
4888              allowed_chunk_alloc)) {
4889                 index = 0;
4890                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
4891                         found_uncached_bg = false;
4892                         loop++;
4893                         if (!ideal_cache_percent &&
4894                             atomic_read(&space_info->caching_threads))
4895                                 goto search;
4896
4897                         /*
4898                          * 1 of the following 2 things have happened so far
4899                          *
4900                          * 1) We found an ideal block group for caching that
4901                          * is mostly full and will cache quickly, so we might
4902                          * as well wait for it.
4903                          *
4904                          * 2) We searched for cached only and we didn't find
4905                          * anything, and we didn't start any caching kthreads
4906                          * either, so chances are we will loop through and
4907                          * start a couple caching kthreads, and then come back
4908                          * around and just wait for them.  This will be slower
4909                          * because we will have 2 caching kthreads reading at
4910                          * the same time when we could have just started one
4911                          * and waited for it to get far enough to give us an
4912                          * allocation, so go ahead and go to the wait caching
4913                          * loop.
4914                          */
4915                         loop = LOOP_CACHING_WAIT;
4916                         search_start = ideal_cache_offset;
4917                         ideal_cache_percent = 0;
4918                         goto ideal_cache;
4919                 } else if (loop == LOOP_FIND_IDEAL) {
4920                         /*
4921                          * Didn't find a uncached bg, wait on anything we find
4922                          * next.
4923                          */
4924                         loop = LOOP_CACHING_WAIT;
4925                         goto search;
4926                 }
4927
4928                 if (loop < LOOP_CACHING_WAIT) {
4929                         loop++;
4930                         goto search;
4931                 }
4932
4933                 if (loop == LOOP_ALLOC_CHUNK) {
4934                         empty_size = 0;
4935                         empty_cluster = 0;
4936                 }
4937
4938                 if (allowed_chunk_alloc) {
4939                         ret = do_chunk_alloc(trans, root, num_bytes +
4940                                              2 * 1024 * 1024, data, 1);
4941                         allowed_chunk_alloc = 0;
4942                         done_chunk_alloc = 1;
4943                 } else if (!done_chunk_alloc) {
4944                         space_info->force_alloc = 1;
4945                 }
4946
4947                 if (loop < LOOP_NO_EMPTY_SIZE) {
4948                         loop++;
4949                         goto search;
4950                 }
4951                 ret = -ENOSPC;
4952         } else if (!ins->objectid) {
4953                 ret = -ENOSPC;
4954         }
4955
4956         /* we found what we needed */
4957         if (ins->objectid) {
4958                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4959                         trans->block_group = block_group->key.objectid;
4960
4961                 btrfs_put_block_group(block_group);
4962                 ret = 0;
4963         }
4964
4965         return ret;
4966 }
4967
4968 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4969                             int dump_block_groups)
4970 {
4971         struct btrfs_block_group_cache *cache;
4972         int index = 0;
4973
4974         spin_lock(&info->lock);
4975         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4976                (unsigned long long)(info->total_bytes - info->bytes_used -
4977                                     info->bytes_pinned - info->bytes_reserved -
4978                                     info->bytes_readonly),
4979                (info->full) ? "" : "not ");
4980         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
4981                "reserved=%llu, may_use=%llu, readonly=%llu\n",
4982                (unsigned long long)info->total_bytes,
4983                (unsigned long long)info->bytes_used,
4984                (unsigned long long)info->bytes_pinned,
4985                (unsigned long long)info->bytes_reserved,
4986                (unsigned long long)info->bytes_may_use,
4987                (unsigned long long)info->bytes_readonly);
4988         spin_unlock(&info->lock);
4989
4990         if (!dump_block_groups)
4991                 return;
4992
4993         down_read(&info->groups_sem);
4994 again:
4995         list_for_each_entry(cache, &info->block_groups[index], list) {
4996                 spin_lock(&cache->lock);
4997                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4998                        "%llu pinned %llu reserved\n",
4999                        (unsigned long long)cache->key.objectid,
5000                        (unsigned long long)cache->key.offset,
5001                        (unsigned long long)btrfs_block_group_used(&cache->item),
5002                        (unsigned long long)cache->pinned,
5003                        (unsigned long long)cache->reserved);
5004                 btrfs_dump_free_space(cache, bytes);
5005                 spin_unlock(&cache->lock);
5006         }
5007         if (++index < BTRFS_NR_RAID_TYPES)
5008                 goto again;
5009         up_read(&info->groups_sem);
5010 }
5011
5012 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5013                          struct btrfs_root *root,
5014                          u64 num_bytes, u64 min_alloc_size,
5015                          u64 empty_size, u64 hint_byte,
5016                          u64 search_end, struct btrfs_key *ins,
5017                          u64 data)
5018 {
5019         int ret;
5020         u64 search_start = 0;
5021
5022         data = btrfs_get_alloc_profile(root, data);
5023 again:
5024         /*
5025          * the only place that sets empty_size is btrfs_realloc_node, which
5026          * is not called recursively on allocations
5027          */
5028         if (empty_size || root->ref_cows)
5029                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5030                                      num_bytes + 2 * 1024 * 1024, data, 0);
5031
5032         WARN_ON(num_bytes < root->sectorsize);
5033         ret = find_free_extent(trans, root, num_bytes, empty_size,
5034                                search_start, search_end, hint_byte,
5035                                ins, data);
5036
5037         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5038                 num_bytes = num_bytes >> 1;
5039                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5040                 num_bytes = max(num_bytes, min_alloc_size);
5041                 do_chunk_alloc(trans, root->fs_info->extent_root,
5042                                num_bytes, data, 1);
5043                 goto again;
5044         }
5045         if (ret == -ENOSPC) {
5046                 struct btrfs_space_info *sinfo;
5047
5048                 sinfo = __find_space_info(root->fs_info, data);
5049                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5050                        "wanted %llu\n", (unsigned long long)data,
5051                        (unsigned long long)num_bytes);
5052                 dump_space_info(sinfo, num_bytes, 1);
5053         }
5054
5055         return ret;
5056 }
5057
5058 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5059 {
5060         struct btrfs_block_group_cache *cache;
5061         int ret = 0;
5062
5063         cache = btrfs_lookup_block_group(root->fs_info, start);
5064         if (!cache) {
5065                 printk(KERN_ERR "Unable to find block group for %llu\n",
5066                        (unsigned long long)start);
5067                 return -ENOSPC;
5068         }
5069
5070         ret = btrfs_discard_extent(root, start, len);
5071
5072         btrfs_add_free_space(cache, start, len);
5073         update_reserved_bytes(cache, len, 0, 1);
5074         btrfs_put_block_group(cache);
5075
5076         return ret;
5077 }
5078
5079 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5080                                       struct btrfs_root *root,
5081                                       u64 parent, u64 root_objectid,
5082                                       u64 flags, u64 owner, u64 offset,
5083                                       struct btrfs_key *ins, int ref_mod)
5084 {
5085         int ret;
5086         struct btrfs_fs_info *fs_info = root->fs_info;
5087         struct btrfs_extent_item *extent_item;
5088         struct btrfs_extent_inline_ref *iref;
5089         struct btrfs_path *path;
5090         struct extent_buffer *leaf;
5091         int type;
5092         u32 size;
5093
5094         if (parent > 0)
5095                 type = BTRFS_SHARED_DATA_REF_KEY;
5096         else
5097                 type = BTRFS_EXTENT_DATA_REF_KEY;
5098
5099         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5100
5101         path = btrfs_alloc_path();
5102         BUG_ON(!path);
5103
5104         path->leave_spinning = 1;
5105         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5106                                       ins, size);
5107         BUG_ON(ret);
5108
5109         leaf = path->nodes[0];
5110         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5111                                      struct btrfs_extent_item);
5112         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5113         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5114         btrfs_set_extent_flags(leaf, extent_item,
5115                                flags | BTRFS_EXTENT_FLAG_DATA);
5116
5117         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5118         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5119         if (parent > 0) {
5120                 struct btrfs_shared_data_ref *ref;
5121                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5122                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5123                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5124         } else {
5125                 struct btrfs_extent_data_ref *ref;
5126                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5127                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5128                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5129                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5130                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5131         }
5132
5133         btrfs_mark_buffer_dirty(path->nodes[0]);
5134         btrfs_free_path(path);
5135
5136         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5137         if (ret) {
5138                 printk(KERN_ERR "btrfs update block group failed for %llu "
5139                        "%llu\n", (unsigned long long)ins->objectid,
5140                        (unsigned long long)ins->offset);
5141                 BUG();
5142         }
5143         return ret;
5144 }
5145
5146 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5147                                      struct btrfs_root *root,
5148                                      u64 parent, u64 root_objectid,
5149                                      u64 flags, struct btrfs_disk_key *key,
5150                                      int level, struct btrfs_key *ins)
5151 {
5152         int ret;
5153         struct btrfs_fs_info *fs_info = root->fs_info;
5154         struct btrfs_extent_item *extent_item;
5155         struct btrfs_tree_block_info *block_info;
5156         struct btrfs_extent_inline_ref *iref;
5157         struct btrfs_path *path;
5158         struct extent_buffer *leaf;
5159         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5160
5161         path = btrfs_alloc_path();
5162         BUG_ON(!path);
5163
5164         path->leave_spinning = 1;
5165         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5166                                       ins, size);
5167         BUG_ON(ret);
5168
5169         leaf = path->nodes[0];
5170         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5171                                      struct btrfs_extent_item);
5172         btrfs_set_extent_refs(leaf, extent_item, 1);
5173         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5174         btrfs_set_extent_flags(leaf, extent_item,
5175                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5176         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5177
5178         btrfs_set_tree_block_key(leaf, block_info, key);
5179         btrfs_set_tree_block_level(leaf, block_info, level);
5180
5181         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5182         if (parent > 0) {
5183                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5184                 btrfs_set_extent_inline_ref_type(leaf, iref,
5185                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5186                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5187         } else {
5188                 btrfs_set_extent_inline_ref_type(leaf, iref,
5189                                                  BTRFS_TREE_BLOCK_REF_KEY);
5190                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5191         }
5192
5193         btrfs_mark_buffer_dirty(leaf);
5194         btrfs_free_path(path);
5195
5196         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5197         if (ret) {
5198                 printk(KERN_ERR "btrfs update block group failed for %llu "
5199                        "%llu\n", (unsigned long long)ins->objectid,
5200                        (unsigned long long)ins->offset);
5201                 BUG();
5202         }
5203         return ret;
5204 }
5205
5206 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5207                                      struct btrfs_root *root,
5208                                      u64 root_objectid, u64 owner,
5209                                      u64 offset, struct btrfs_key *ins)
5210 {
5211         int ret;
5212
5213         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5214
5215         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5216                                          0, root_objectid, owner, offset,
5217                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5218         return ret;
5219 }
5220
5221 /*
5222  * this is used by the tree logging recovery code.  It records that
5223  * an extent has been allocated and makes sure to clear the free
5224  * space cache bits as well
5225  */
5226 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5227                                    struct btrfs_root *root,
5228                                    u64 root_objectid, u64 owner, u64 offset,
5229                                    struct btrfs_key *ins)
5230 {
5231         int ret;
5232         struct btrfs_block_group_cache *block_group;
5233         struct btrfs_caching_control *caching_ctl;
5234         u64 start = ins->objectid;
5235         u64 num_bytes = ins->offset;
5236
5237         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5238         cache_block_group(block_group);
5239         caching_ctl = get_caching_control(block_group);
5240
5241         if (!caching_ctl) {
5242                 BUG_ON(!block_group_cache_done(block_group));
5243                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5244                 BUG_ON(ret);
5245         } else {
5246                 mutex_lock(&caching_ctl->mutex);
5247
5248                 if (start >= caching_ctl->progress) {
5249                         ret = add_excluded_extent(root, start, num_bytes);
5250                         BUG_ON(ret);
5251                 } else if (start + num_bytes <= caching_ctl->progress) {
5252                         ret = btrfs_remove_free_space(block_group,
5253                                                       start, num_bytes);
5254                         BUG_ON(ret);
5255                 } else {
5256                         num_bytes = caching_ctl->progress - start;
5257                         ret = btrfs_remove_free_space(block_group,
5258                                                       start, num_bytes);
5259                         BUG_ON(ret);
5260
5261                         start = caching_ctl->progress;
5262                         num_bytes = ins->objectid + ins->offset -
5263                                     caching_ctl->progress;
5264                         ret = add_excluded_extent(root, start, num_bytes);
5265                         BUG_ON(ret);
5266                 }
5267
5268                 mutex_unlock(&caching_ctl->mutex);
5269                 put_caching_control(caching_ctl);
5270         }
5271
5272         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5273         BUG_ON(ret);
5274         btrfs_put_block_group(block_group);
5275         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5276                                          0, owner, offset, ins, 1);
5277         return ret;
5278 }
5279
5280 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5281                                             struct btrfs_root *root,
5282                                             u64 bytenr, u32 blocksize,
5283                                             int level)
5284 {
5285         struct extent_buffer *buf;
5286
5287         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5288         if (!buf)
5289                 return ERR_PTR(-ENOMEM);
5290         btrfs_set_header_generation(buf, trans->transid);
5291         btrfs_set_buffer_lockdep_class(buf, level);
5292         btrfs_tree_lock(buf);
5293         clean_tree_block(trans, root, buf);
5294
5295         btrfs_set_lock_blocking(buf);
5296         btrfs_set_buffer_uptodate(buf);
5297
5298         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5299                 /*
5300                  * we allow two log transactions at a time, use different
5301                  * EXENT bit to differentiate dirty pages.
5302                  */
5303                 if (root->log_transid % 2 == 0)
5304                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5305                                         buf->start + buf->len - 1, GFP_NOFS);
5306                 else
5307                         set_extent_new(&root->dirty_log_pages, buf->start,
5308                                         buf->start + buf->len - 1, GFP_NOFS);
5309         } else {
5310                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5311                          buf->start + buf->len - 1, GFP_NOFS);
5312         }
5313         trans->blocks_used++;
5314         /* this returns a buffer locked for blocking */
5315         return buf;
5316 }
5317
5318 static struct btrfs_block_rsv *
5319 use_block_rsv(struct btrfs_trans_handle *trans,
5320               struct btrfs_root *root, u32 blocksize)
5321 {
5322         struct btrfs_block_rsv *block_rsv;
5323         int ret;
5324
5325         block_rsv = get_block_rsv(trans, root);
5326
5327         if (block_rsv->size == 0) {
5328                 ret = reserve_metadata_bytes(block_rsv, blocksize);
5329                 if (ret)
5330                         return ERR_PTR(ret);
5331                 return block_rsv;
5332         }
5333
5334         ret = block_rsv_use_bytes(block_rsv, blocksize);
5335         if (!ret)
5336                 return block_rsv;
5337
5338         WARN_ON(1);
5339         printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
5340                 block_rsv->size, block_rsv->reserved,
5341                 block_rsv->freed[0], block_rsv->freed[1]);
5342
5343         return ERR_PTR(-ENOSPC);
5344 }
5345
5346 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5347 {
5348         block_rsv_add_bytes(block_rsv, blocksize, 0);
5349         block_rsv_release_bytes(block_rsv, NULL, 0);
5350 }
5351
5352 /*
5353  * finds a free extent and does all the dirty work required for allocation
5354  * returns the key for the extent through ins, and a tree buffer for
5355  * the first block of the extent through buf.
5356  *
5357  * returns the tree buffer or NULL.
5358  */
5359 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5360                                         struct btrfs_root *root, u32 blocksize,
5361                                         u64 parent, u64 root_objectid,
5362                                         struct btrfs_disk_key *key, int level,
5363                                         u64 hint, u64 empty_size)
5364 {
5365         struct btrfs_key ins;
5366         struct btrfs_block_rsv *block_rsv;
5367         struct extent_buffer *buf;
5368         u64 flags = 0;
5369         int ret;
5370
5371
5372         block_rsv = use_block_rsv(trans, root, blocksize);
5373         if (IS_ERR(block_rsv))
5374                 return ERR_CAST(block_rsv);
5375
5376         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5377                                    empty_size, hint, (u64)-1, &ins, 0);
5378         if (ret) {
5379                 unuse_block_rsv(block_rsv, blocksize);
5380                 return ERR_PTR(ret);
5381         }
5382
5383         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5384                                     blocksize, level);
5385         BUG_ON(IS_ERR(buf));
5386
5387         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5388                 if (parent == 0)
5389                         parent = ins.objectid;
5390                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5391         } else
5392                 BUG_ON(parent > 0);
5393
5394         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5395                 struct btrfs_delayed_extent_op *extent_op;
5396                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5397                 BUG_ON(!extent_op);
5398                 if (key)
5399                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5400                 else
5401                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5402                 extent_op->flags_to_set = flags;
5403                 extent_op->update_key = 1;
5404                 extent_op->update_flags = 1;
5405                 extent_op->is_data = 0;
5406
5407                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5408                                         ins.offset, parent, root_objectid,
5409                                         level, BTRFS_ADD_DELAYED_EXTENT,
5410                                         extent_op);
5411                 BUG_ON(ret);
5412         }
5413         return buf;
5414 }
5415
5416 struct walk_control {
5417         u64 refs[BTRFS_MAX_LEVEL];
5418         u64 flags[BTRFS_MAX_LEVEL];
5419         struct btrfs_key update_progress;
5420         int stage;
5421         int level;
5422         int shared_level;
5423         int update_ref;
5424         int keep_locks;
5425         int reada_slot;
5426         int reada_count;
5427 };
5428
5429 #define DROP_REFERENCE  1
5430 #define UPDATE_BACKREF  2
5431
5432 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5433                                      struct btrfs_root *root,
5434                                      struct walk_control *wc,
5435                                      struct btrfs_path *path)
5436 {
5437         u64 bytenr;
5438         u64 generation;
5439         u64 refs;
5440         u64 flags;
5441         u64 last = 0;
5442         u32 nritems;
5443         u32 blocksize;
5444         struct btrfs_key key;
5445         struct extent_buffer *eb;
5446         int ret;
5447         int slot;
5448         int nread = 0;
5449
5450         if (path->slots[wc->level] < wc->reada_slot) {
5451                 wc->reada_count = wc->reada_count * 2 / 3;
5452                 wc->reada_count = max(wc->reada_count, 2);
5453         } else {
5454                 wc->reada_count = wc->reada_count * 3 / 2;
5455                 wc->reada_count = min_t(int, wc->reada_count,
5456                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5457         }
5458
5459         eb = path->nodes[wc->level];
5460         nritems = btrfs_header_nritems(eb);
5461         blocksize = btrfs_level_size(root, wc->level - 1);
5462
5463         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5464                 if (nread >= wc->reada_count)
5465                         break;
5466
5467                 cond_resched();
5468                 bytenr = btrfs_node_blockptr(eb, slot);
5469                 generation = btrfs_node_ptr_generation(eb, slot);
5470
5471                 if (slot == path->slots[wc->level])
5472                         goto reada;
5473
5474                 if (wc->stage == UPDATE_BACKREF &&
5475                     generation <= root->root_key.offset)
5476                         continue;
5477
5478                 /* We don't lock the tree block, it's OK to be racy here */
5479                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5480                                                &refs, &flags);
5481                 BUG_ON(ret);
5482                 BUG_ON(refs == 0);
5483
5484                 if (wc->stage == DROP_REFERENCE) {
5485                         if (refs == 1)
5486                                 goto reada;
5487
5488                         if (wc->level == 1 &&
5489                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5490                                 continue;
5491                         if (!wc->update_ref ||
5492                             generation <= root->root_key.offset)
5493                                 continue;
5494                         btrfs_node_key_to_cpu(eb, &key, slot);
5495                         ret = btrfs_comp_cpu_keys(&key,
5496                                                   &wc->update_progress);
5497                         if (ret < 0)
5498                                 continue;
5499                 } else {
5500                         if (wc->level == 1 &&
5501                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5502                                 continue;
5503                 }
5504 reada:
5505                 ret = readahead_tree_block(root, bytenr, blocksize,
5506                                            generation);
5507                 if (ret)
5508                         break;
5509                 last = bytenr + blocksize;
5510                 nread++;
5511         }
5512         wc->reada_slot = slot;
5513 }
5514
5515 /*
5516  * hepler to process tree block while walking down the tree.
5517  *
5518  * when wc->stage == UPDATE_BACKREF, this function updates
5519  * back refs for pointers in the block.
5520  *
5521  * NOTE: return value 1 means we should stop walking down.
5522  */
5523 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5524                                    struct btrfs_root *root,
5525                                    struct btrfs_path *path,
5526                                    struct walk_control *wc, int lookup_info)
5527 {
5528         int level = wc->level;
5529         struct extent_buffer *eb = path->nodes[level];
5530         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5531         int ret;
5532
5533         if (wc->stage == UPDATE_BACKREF &&
5534             btrfs_header_owner(eb) != root->root_key.objectid)
5535                 return 1;
5536
5537         /*
5538          * when reference count of tree block is 1, it won't increase
5539          * again. once full backref flag is set, we never clear it.
5540          */
5541         if (lookup_info &&
5542             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5543              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5544                 BUG_ON(!path->locks[level]);
5545                 ret = btrfs_lookup_extent_info(trans, root,
5546                                                eb->start, eb->len,
5547                                                &wc->refs[level],
5548                                                &wc->flags[level]);
5549                 BUG_ON(ret);
5550                 BUG_ON(wc->refs[level] == 0);
5551         }
5552
5553         if (wc->stage == DROP_REFERENCE) {
5554                 if (wc->refs[level] > 1)
5555                         return 1;
5556
5557                 if (path->locks[level] && !wc->keep_locks) {
5558                         btrfs_tree_unlock(eb);
5559                         path->locks[level] = 0;
5560                 }
5561                 return 0;
5562         }
5563
5564         /* wc->stage == UPDATE_BACKREF */
5565         if (!(wc->flags[level] & flag)) {
5566                 BUG_ON(!path->locks[level]);
5567                 ret = btrfs_inc_ref(trans, root, eb, 1);
5568                 BUG_ON(ret);
5569                 ret = btrfs_dec_ref(trans, root, eb, 0);
5570                 BUG_ON(ret);
5571                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5572                                                   eb->len, flag, 0);
5573                 BUG_ON(ret);
5574                 wc->flags[level] |= flag;
5575         }
5576
5577         /*
5578          * the block is shared by multiple trees, so it's not good to
5579          * keep the tree lock
5580          */
5581         if (path->locks[level] && level > 0) {
5582                 btrfs_tree_unlock(eb);
5583                 path->locks[level] = 0;
5584         }
5585         return 0;
5586 }
5587
5588 /*
5589  * hepler to process tree block pointer.
5590  *
5591  * when wc->stage == DROP_REFERENCE, this function checks
5592  * reference count of the block pointed to. if the block
5593  * is shared and we need update back refs for the subtree
5594  * rooted at the block, this function changes wc->stage to
5595  * UPDATE_BACKREF. if the block is shared and there is no
5596  * need to update back, this function drops the reference
5597  * to the block.
5598  *
5599  * NOTE: return value 1 means we should stop walking down.
5600  */
5601 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5602                                  struct btrfs_root *root,
5603                                  struct btrfs_path *path,
5604                                  struct walk_control *wc, int *lookup_info)
5605 {
5606         u64 bytenr;
5607         u64 generation;
5608         u64 parent;
5609         u32 blocksize;
5610         struct btrfs_key key;
5611         struct extent_buffer *next;
5612         int level = wc->level;
5613         int reada = 0;
5614         int ret = 0;
5615
5616         generation = btrfs_node_ptr_generation(path->nodes[level],
5617                                                path->slots[level]);
5618         /*
5619          * if the lower level block was created before the snapshot
5620          * was created, we know there is no need to update back refs
5621          * for the subtree
5622          */
5623         if (wc->stage == UPDATE_BACKREF &&
5624             generation <= root->root_key.offset) {
5625                 *lookup_info = 1;
5626                 return 1;
5627         }
5628
5629         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5630         blocksize = btrfs_level_size(root, level - 1);
5631
5632         next = btrfs_find_tree_block(root, bytenr, blocksize);
5633         if (!next) {
5634                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5635                 if (!next)
5636                         return -ENOMEM;
5637                 reada = 1;
5638         }
5639         btrfs_tree_lock(next);
5640         btrfs_set_lock_blocking(next);
5641
5642         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5643                                        &wc->refs[level - 1],
5644                                        &wc->flags[level - 1]);
5645         BUG_ON(ret);
5646         BUG_ON(wc->refs[level - 1] == 0);
5647         *lookup_info = 0;
5648
5649         if (wc->stage == DROP_REFERENCE) {
5650                 if (wc->refs[level - 1] > 1) {
5651                         if (level == 1 &&
5652                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5653                                 goto skip;
5654
5655                         if (!wc->update_ref ||
5656                             generation <= root->root_key.offset)
5657                                 goto skip;
5658
5659                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5660                                               path->slots[level]);
5661                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5662                         if (ret < 0)
5663                                 goto skip;
5664
5665                         wc->stage = UPDATE_BACKREF;
5666                         wc->shared_level = level - 1;
5667                 }
5668         } else {
5669                 if (level == 1 &&
5670                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5671                         goto skip;
5672         }
5673
5674         if (!btrfs_buffer_uptodate(next, generation)) {
5675                 btrfs_tree_unlock(next);
5676                 free_extent_buffer(next);
5677                 next = NULL;
5678                 *lookup_info = 1;
5679         }
5680
5681         if (!next) {
5682                 if (reada && level == 1)
5683                         reada_walk_down(trans, root, wc, path);
5684                 next = read_tree_block(root, bytenr, blocksize, generation);
5685                 btrfs_tree_lock(next);
5686                 btrfs_set_lock_blocking(next);
5687         }
5688
5689         level--;
5690         BUG_ON(level != btrfs_header_level(next));
5691         path->nodes[level] = next;
5692         path->slots[level] = 0;
5693         path->locks[level] = 1;
5694         wc->level = level;
5695         if (wc->level == 1)
5696                 wc->reada_slot = 0;
5697         return 0;
5698 skip:
5699         wc->refs[level - 1] = 0;
5700         wc->flags[level - 1] = 0;
5701         if (wc->stage == DROP_REFERENCE) {
5702                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5703                         parent = path->nodes[level]->start;
5704                 } else {
5705                         BUG_ON(root->root_key.objectid !=
5706                                btrfs_header_owner(path->nodes[level]));
5707                         parent = 0;
5708                 }
5709
5710                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5711                                         root->root_key.objectid, level - 1, 0);
5712                 BUG_ON(ret);
5713         }
5714         btrfs_tree_unlock(next);
5715         free_extent_buffer(next);
5716         *lookup_info = 1;
5717         return 1;
5718 }
5719
5720 /*
5721  * hepler to process tree block while walking up the tree.
5722  *
5723  * when wc->stage == DROP_REFERENCE, this function drops
5724  * reference count on the block.
5725  *
5726  * when wc->stage == UPDATE_BACKREF, this function changes
5727  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5728  * to UPDATE_BACKREF previously while processing the block.
5729  *
5730  * NOTE: return value 1 means we should stop walking up.
5731  */
5732 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5733                                  struct btrfs_root *root,
5734                                  struct btrfs_path *path,
5735                                  struct walk_control *wc)
5736 {
5737         int ret;
5738         int level = wc->level;
5739         struct extent_buffer *eb = path->nodes[level];
5740         u64 parent = 0;
5741
5742         if (wc->stage == UPDATE_BACKREF) {
5743                 BUG_ON(wc->shared_level < level);
5744                 if (level < wc->shared_level)
5745                         goto out;
5746
5747                 ret = find_next_key(path, level + 1, &wc->update_progress);
5748                 if (ret > 0)
5749                         wc->update_ref = 0;
5750
5751                 wc->stage = DROP_REFERENCE;
5752                 wc->shared_level = -1;
5753                 path->slots[level] = 0;
5754
5755                 /*
5756                  * check reference count again if the block isn't locked.
5757                  * we should start walking down the tree again if reference
5758                  * count is one.
5759                  */
5760                 if (!path->locks[level]) {
5761                         BUG_ON(level == 0);
5762                         btrfs_tree_lock(eb);
5763                         btrfs_set_lock_blocking(eb);
5764                         path->locks[level] = 1;
5765
5766                         ret = btrfs_lookup_extent_info(trans, root,
5767                                                        eb->start, eb->len,
5768                                                        &wc->refs[level],
5769                                                        &wc->flags[level]);
5770                         BUG_ON(ret);
5771                         BUG_ON(wc->refs[level] == 0);
5772                         if (wc->refs[level] == 1) {
5773                                 btrfs_tree_unlock(eb);
5774                                 path->locks[level] = 0;
5775                                 return 1;
5776                         }
5777                 }
5778         }
5779
5780         /* wc->stage == DROP_REFERENCE */
5781         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5782
5783         if (wc->refs[level] == 1) {
5784                 if (level == 0) {
5785                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5786                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5787                         else
5788                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5789                         BUG_ON(ret);
5790                 }
5791                 /* make block locked assertion in clean_tree_block happy */
5792                 if (!path->locks[level] &&
5793                     btrfs_header_generation(eb) == trans->transid) {
5794                         btrfs_tree_lock(eb);
5795                         btrfs_set_lock_blocking(eb);
5796                         path->locks[level] = 1;
5797                 }
5798                 clean_tree_block(trans, root, eb);
5799         }
5800
5801         if (eb == root->node) {
5802                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5803                         parent = eb->start;
5804                 else
5805                         BUG_ON(root->root_key.objectid !=
5806                                btrfs_header_owner(eb));
5807         } else {
5808                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5809                         parent = path->nodes[level + 1]->start;
5810                 else
5811                         BUG_ON(root->root_key.objectid !=
5812                                btrfs_header_owner(path->nodes[level + 1]));
5813         }
5814
5815         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
5816 out:
5817         wc->refs[level] = 0;
5818         wc->flags[level] = 0;
5819         return 0;
5820 }
5821
5822 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5823                                    struct btrfs_root *root,
5824                                    struct btrfs_path *path,
5825                                    struct walk_control *wc)
5826 {
5827         int level = wc->level;
5828         int lookup_info = 1;
5829         int ret;
5830
5831         while (level >= 0) {
5832                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5833                 if (ret > 0)
5834                         break;
5835
5836                 if (level == 0)
5837                         break;
5838
5839                 if (path->slots[level] >=
5840                     btrfs_header_nritems(path->nodes[level]))
5841                         break;
5842
5843                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5844                 if (ret > 0) {
5845                         path->slots[level]++;
5846                         continue;
5847                 } else if (ret < 0)
5848                         return ret;
5849                 level = wc->level;
5850         }
5851         return 0;
5852 }
5853
5854 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5855                                  struct btrfs_root *root,
5856                                  struct btrfs_path *path,
5857                                  struct walk_control *wc, int max_level)
5858 {
5859         int level = wc->level;
5860         int ret;
5861
5862         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5863         while (level < max_level && path->nodes[level]) {
5864                 wc->level = level;
5865                 if (path->slots[level] + 1 <
5866                     btrfs_header_nritems(path->nodes[level])) {
5867                         path->slots[level]++;
5868                         return 0;
5869                 } else {
5870                         ret = walk_up_proc(trans, root, path, wc);
5871                         if (ret > 0)
5872                                 return 0;
5873
5874                         if (path->locks[level]) {
5875                                 btrfs_tree_unlock(path->nodes[level]);
5876                                 path->locks[level] = 0;
5877                         }
5878                         free_extent_buffer(path->nodes[level]);
5879                         path->nodes[level] = NULL;
5880                         level++;
5881                 }
5882         }
5883         return 1;
5884 }
5885
5886 /*
5887  * drop a subvolume tree.
5888  *
5889  * this function traverses the tree freeing any blocks that only
5890  * referenced by the tree.
5891  *
5892  * when a shared tree block is found. this function decreases its
5893  * reference count by one. if update_ref is true, this function
5894  * also make sure backrefs for the shared block and all lower level
5895  * blocks are properly updated.
5896  */
5897 int btrfs_drop_snapshot(struct btrfs_root *root,
5898                         struct btrfs_block_rsv *block_rsv, int update_ref)
5899 {
5900         struct btrfs_path *path;
5901         struct btrfs_trans_handle *trans;
5902         struct btrfs_root *tree_root = root->fs_info->tree_root;
5903         struct btrfs_root_item *root_item = &root->root_item;
5904         struct walk_control *wc;
5905         struct btrfs_key key;
5906         int err = 0;
5907         int ret;
5908         int level;
5909
5910         path = btrfs_alloc_path();
5911         BUG_ON(!path);
5912
5913         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5914         BUG_ON(!wc);
5915
5916         trans = btrfs_start_transaction(tree_root, 0);
5917         if (block_rsv)
5918                 trans->block_rsv = block_rsv;
5919
5920         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5921                 level = btrfs_header_level(root->node);
5922                 path->nodes[level] = btrfs_lock_root_node(root);
5923                 btrfs_set_lock_blocking(path->nodes[level]);
5924                 path->slots[level] = 0;
5925                 path->locks[level] = 1;
5926                 memset(&wc->update_progress, 0,
5927                        sizeof(wc->update_progress));
5928         } else {
5929                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5930                 memcpy(&wc->update_progress, &key,
5931                        sizeof(wc->update_progress));
5932
5933                 level = root_item->drop_level;
5934                 BUG_ON(level == 0);
5935                 path->lowest_level = level;
5936                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5937                 path->lowest_level = 0;
5938                 if (ret < 0) {
5939                         err = ret;
5940                         goto out;
5941                 }
5942                 WARN_ON(ret > 0);
5943
5944                 /*
5945                  * unlock our path, this is safe because only this
5946                  * function is allowed to delete this snapshot
5947                  */
5948                 btrfs_unlock_up_safe(path, 0);
5949
5950                 level = btrfs_header_level(root->node);
5951                 while (1) {
5952                         btrfs_tree_lock(path->nodes[level]);
5953                         btrfs_set_lock_blocking(path->nodes[level]);
5954
5955                         ret = btrfs_lookup_extent_info(trans, root,
5956                                                 path->nodes[level]->start,
5957                                                 path->nodes[level]->len,
5958                                                 &wc->refs[level],
5959                                                 &wc->flags[level]);
5960                         BUG_ON(ret);
5961                         BUG_ON(wc->refs[level] == 0);
5962
5963                         if (level == root_item->drop_level)
5964                                 break;
5965
5966                         btrfs_tree_unlock(path->nodes[level]);
5967                         WARN_ON(wc->refs[level] != 1);
5968                         level--;
5969                 }
5970         }
5971
5972         wc->level = level;
5973         wc->shared_level = -1;
5974         wc->stage = DROP_REFERENCE;
5975         wc->update_ref = update_ref;
5976         wc->keep_locks = 0;
5977         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5978
5979         while (1) {
5980                 ret = walk_down_tree(trans, root, path, wc);
5981                 if (ret < 0) {
5982                         err = ret;
5983                         break;
5984                 }
5985
5986                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5987                 if (ret < 0) {
5988                         err = ret;
5989                         break;
5990                 }
5991
5992                 if (ret > 0) {
5993                         BUG_ON(wc->stage != DROP_REFERENCE);
5994                         break;
5995                 }
5996
5997                 if (wc->stage == DROP_REFERENCE) {
5998                         level = wc->level;
5999                         btrfs_node_key(path->nodes[level],
6000                                        &root_item->drop_progress,
6001                                        path->slots[level]);
6002                         root_item->drop_level = level;
6003                 }
6004
6005                 BUG_ON(wc->level == 0);
6006                 if (btrfs_should_end_transaction(trans, tree_root)) {
6007                         ret = btrfs_update_root(trans, tree_root,
6008                                                 &root->root_key,
6009                                                 root_item);
6010                         BUG_ON(ret);
6011
6012                         btrfs_end_transaction_throttle(trans, tree_root);
6013                         trans = btrfs_start_transaction(tree_root, 0);
6014                         if (block_rsv)
6015                                 trans->block_rsv = block_rsv;
6016                 }
6017         }
6018         btrfs_release_path(root, path);
6019         BUG_ON(err);
6020
6021         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6022         BUG_ON(ret);
6023
6024         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6025                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6026                                            NULL, NULL);
6027                 BUG_ON(ret < 0);
6028                 if (ret > 0) {
6029                         ret = btrfs_del_orphan_item(trans, tree_root,
6030                                                     root->root_key.objectid);
6031                         BUG_ON(ret);
6032                 }
6033         }
6034
6035         if (root->in_radix) {
6036                 btrfs_free_fs_root(tree_root->fs_info, root);
6037         } else {
6038                 free_extent_buffer(root->node);
6039                 free_extent_buffer(root->commit_root);
6040                 kfree(root);
6041         }
6042 out:
6043         btrfs_end_transaction_throttle(trans, tree_root);
6044         kfree(wc);
6045         btrfs_free_path(path);
6046         return err;
6047 }
6048
6049 /*
6050  * drop subtree rooted at tree block 'node'.
6051  *
6052  * NOTE: this function will unlock and release tree block 'node'
6053  */
6054 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6055                         struct btrfs_root *root,
6056                         struct extent_buffer *node,
6057                         struct extent_buffer *parent)
6058 {
6059         struct btrfs_path *path;
6060         struct walk_control *wc;
6061         int level;
6062         int parent_level;
6063         int ret = 0;
6064         int wret;
6065
6066         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6067
6068         path = btrfs_alloc_path();
6069         BUG_ON(!path);
6070
6071         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6072         BUG_ON(!wc);
6073
6074         btrfs_assert_tree_locked(parent);
6075         parent_level = btrfs_header_level(parent);
6076         extent_buffer_get(parent);
6077         path->nodes[parent_level] = parent;
6078         path->slots[parent_level] = btrfs_header_nritems(parent);
6079
6080         btrfs_assert_tree_locked(node);
6081         level = btrfs_header_level(node);
6082         path->nodes[level] = node;
6083         path->slots[level] = 0;
6084         path->locks[level] = 1;
6085
6086         wc->refs[parent_level] = 1;
6087         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6088         wc->level = level;
6089         wc->shared_level = -1;
6090         wc->stage = DROP_REFERENCE;
6091         wc->update_ref = 0;
6092         wc->keep_locks = 1;
6093         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6094
6095         while (1) {
6096                 wret = walk_down_tree(trans, root, path, wc);
6097                 if (wret < 0) {
6098                         ret = wret;
6099                         break;
6100                 }
6101
6102                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6103                 if (wret < 0)
6104                         ret = wret;
6105                 if (wret != 0)
6106                         break;
6107         }
6108
6109         kfree(wc);
6110         btrfs_free_path(path);
6111         return ret;
6112 }
6113
6114 #if 0
6115 static unsigned long calc_ra(unsigned long start, unsigned long last,
6116                              unsigned long nr)
6117 {
6118         return min(last, start + nr - 1);
6119 }
6120
6121 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6122                                          u64 len)
6123 {
6124         u64 page_start;
6125         u64 page_end;
6126         unsigned long first_index;
6127         unsigned long last_index;
6128         unsigned long i;
6129         struct page *page;
6130         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6131         struct file_ra_state *ra;
6132         struct btrfs_ordered_extent *ordered;
6133         unsigned int total_read = 0;
6134         unsigned int total_dirty = 0;
6135         int ret = 0;
6136
6137         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6138
6139         mutex_lock(&inode->i_mutex);
6140         first_index = start >> PAGE_CACHE_SHIFT;
6141         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6142
6143         /* make sure the dirty trick played by the caller work */
6144         ret = invalidate_inode_pages2_range(inode->i_mapping,
6145                                             first_index, last_index);
6146         if (ret)
6147                 goto out_unlock;
6148
6149         file_ra_state_init(ra, inode->i_mapping);
6150
6151         for (i = first_index ; i <= last_index; i++) {
6152                 if (total_read % ra->ra_pages == 0) {
6153                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6154                                        calc_ra(i, last_index, ra->ra_pages));
6155                 }
6156                 total_read++;
6157 again:
6158                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6159                         BUG_ON(1);
6160                 page = grab_cache_page(inode->i_mapping, i);
6161                 if (!page) {
6162                         ret = -ENOMEM;
6163                         goto out_unlock;
6164                 }
6165                 if (!PageUptodate(page)) {
6166                         btrfs_readpage(NULL, page);
6167                         lock_page(page);
6168                         if (!PageUptodate(page)) {
6169                                 unlock_page(page);
6170                                 page_cache_release(page);
6171                                 ret = -EIO;
6172                                 goto out_unlock;
6173                         }
6174                 }
6175                 wait_on_page_writeback(page);
6176
6177                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6178                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6179                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6180
6181                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6182                 if (ordered) {
6183                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6184                         unlock_page(page);
6185                         page_cache_release(page);
6186                         btrfs_start_ordered_extent(inode, ordered, 1);
6187                         btrfs_put_ordered_extent(ordered);
6188                         goto again;
6189                 }
6190                 set_page_extent_mapped(page);
6191
6192                 if (i == first_index)
6193                         set_extent_bits(io_tree, page_start, page_end,
6194                                         EXTENT_BOUNDARY, GFP_NOFS);
6195                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6196
6197                 set_page_dirty(page);
6198                 total_dirty++;
6199
6200                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6201                 unlock_page(page);
6202                 page_cache_release(page);
6203         }
6204
6205 out_unlock:
6206         kfree(ra);
6207         mutex_unlock(&inode->i_mutex);
6208         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6209         return ret;
6210 }
6211
6212 static noinline int relocate_data_extent(struct inode *reloc_inode,
6213                                          struct btrfs_key *extent_key,
6214                                          u64 offset)
6215 {
6216         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6217         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6218         struct extent_map *em;
6219         u64 start = extent_key->objectid - offset;
6220         u64 end = start + extent_key->offset - 1;
6221
6222         em = alloc_extent_map(GFP_NOFS);
6223         BUG_ON(!em || IS_ERR(em));
6224
6225         em->start = start;
6226         em->len = extent_key->offset;
6227         em->block_len = extent_key->offset;
6228         em->block_start = extent_key->objectid;
6229         em->bdev = root->fs_info->fs_devices->latest_bdev;
6230         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6231
6232         /* setup extent map to cheat btrfs_readpage */
6233         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6234         while (1) {
6235                 int ret;
6236                 write_lock(&em_tree->lock);
6237                 ret = add_extent_mapping(em_tree, em);
6238                 write_unlock(&em_tree->lock);
6239                 if (ret != -EEXIST) {
6240                         free_extent_map(em);
6241                         break;
6242                 }
6243                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6244         }
6245         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6246
6247         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6248 }
6249
6250 struct btrfs_ref_path {
6251         u64 extent_start;
6252         u64 nodes[BTRFS_MAX_LEVEL];
6253         u64 root_objectid;
6254         u64 root_generation;
6255         u64 owner_objectid;
6256         u32 num_refs;
6257         int lowest_level;
6258         int current_level;
6259         int shared_level;
6260
6261         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6262         u64 new_nodes[BTRFS_MAX_LEVEL];
6263 };
6264
6265 struct disk_extent {
6266         u64 ram_bytes;
6267         u64 disk_bytenr;
6268         u64 disk_num_bytes;
6269         u64 offset;
6270         u64 num_bytes;
6271         u8 compression;
6272         u8 encryption;
6273         u16 other_encoding;
6274 };
6275
6276 static int is_cowonly_root(u64 root_objectid)
6277 {
6278         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6279             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6280             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6281             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6282             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6283             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6284                 return 1;
6285         return 0;
6286 }
6287
6288 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6289                                     struct btrfs_root *extent_root,
6290                                     struct btrfs_ref_path *ref_path,
6291                                     int first_time)
6292 {
6293         struct extent_buffer *leaf;
6294         struct btrfs_path *path;
6295         struct btrfs_extent_ref *ref;
6296         struct btrfs_key key;
6297         struct btrfs_key found_key;
6298         u64 bytenr;
6299         u32 nritems;
6300         int level;
6301         int ret = 1;
6302
6303         path = btrfs_alloc_path();
6304         if (!path)
6305                 return -ENOMEM;
6306
6307         if (first_time) {
6308                 ref_path->lowest_level = -1;
6309                 ref_path->current_level = -1;
6310                 ref_path->shared_level = -1;
6311                 goto walk_up;
6312         }
6313 walk_down:
6314         level = ref_path->current_level - 1;
6315         while (level >= -1) {
6316                 u64 parent;
6317                 if (level < ref_path->lowest_level)
6318                         break;
6319
6320                 if (level >= 0)
6321                         bytenr = ref_path->nodes[level];
6322                 else
6323                         bytenr = ref_path->extent_start;
6324                 BUG_ON(bytenr == 0);
6325
6326                 parent = ref_path->nodes[level + 1];
6327                 ref_path->nodes[level + 1] = 0;
6328                 ref_path->current_level = level;
6329                 BUG_ON(parent == 0);
6330
6331                 key.objectid = bytenr;
6332                 key.offset = parent + 1;
6333                 key.type = BTRFS_EXTENT_REF_KEY;
6334
6335                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6336                 if (ret < 0)
6337                         goto out;
6338                 BUG_ON(ret == 0);
6339
6340                 leaf = path->nodes[0];
6341                 nritems = btrfs_header_nritems(leaf);
6342                 if (path->slots[0] >= nritems) {
6343                         ret = btrfs_next_leaf(extent_root, path);
6344                         if (ret < 0)
6345                                 goto out;
6346                         if (ret > 0)
6347                                 goto next;
6348                         leaf = path->nodes[0];
6349                 }
6350
6351                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6352                 if (found_key.objectid == bytenr &&
6353                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6354                         if (level < ref_path->shared_level)
6355                                 ref_path->shared_level = level;
6356                         goto found;
6357                 }
6358 next:
6359                 level--;
6360                 btrfs_release_path(extent_root, path);
6361                 cond_resched();
6362         }
6363         /* reached lowest level */
6364         ret = 1;
6365         goto out;
6366 walk_up:
6367         level = ref_path->current_level;
6368         while (level < BTRFS_MAX_LEVEL - 1) {
6369                 u64 ref_objectid;
6370
6371                 if (level >= 0)
6372                         bytenr = ref_path->nodes[level];
6373                 else
6374                         bytenr = ref_path->extent_start;
6375
6376                 BUG_ON(bytenr == 0);
6377
6378                 key.objectid = bytenr;
6379                 key.offset = 0;
6380                 key.type = BTRFS_EXTENT_REF_KEY;
6381
6382                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6383                 if (ret < 0)
6384                         goto out;
6385
6386                 leaf = path->nodes[0];
6387                 nritems = btrfs_header_nritems(leaf);
6388                 if (path->slots[0] >= nritems) {
6389                         ret = btrfs_next_leaf(extent_root, path);
6390                         if (ret < 0)
6391                                 goto out;
6392                         if (ret > 0) {
6393                                 /* the extent was freed by someone */
6394                                 if (ref_path->lowest_level == level)
6395                                         goto out;
6396                                 btrfs_release_path(extent_root, path);
6397                                 goto walk_down;
6398                         }
6399                         leaf = path->nodes[0];
6400                 }
6401
6402                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6403                 if (found_key.objectid != bytenr ||
6404                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6405                         /* the extent was freed by someone */
6406                         if (ref_path->lowest_level == level) {
6407                                 ret = 1;
6408                                 goto out;
6409                         }
6410                         btrfs_release_path(extent_root, path);
6411                         goto walk_down;
6412                 }
6413 found:
6414                 ref = btrfs_item_ptr(leaf, path->slots[0],
6415                                 struct btrfs_extent_ref);
6416                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6417                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6418                         if (first_time) {
6419                                 level = (int)ref_objectid;
6420                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6421                                 ref_path->lowest_level = level;
6422                                 ref_path->current_level = level;
6423                                 ref_path->nodes[level] = bytenr;
6424                         } else {
6425                                 WARN_ON(ref_objectid != level);
6426                         }
6427                 } else {
6428                         WARN_ON(level != -1);
6429                 }
6430                 first_time = 0;
6431
6432                 if (ref_path->lowest_level == level) {
6433                         ref_path->owner_objectid = ref_objectid;
6434                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6435                 }
6436
6437                 /*
6438                  * the block is tree root or the block isn't in reference
6439                  * counted tree.
6440                  */
6441                 if (found_key.objectid == found_key.offset ||
6442                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6443                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6444                         ref_path->root_generation =
6445                                 btrfs_ref_generation(leaf, ref);
6446                         if (level < 0) {
6447                                 /* special reference from the tree log */
6448                                 ref_path->nodes[0] = found_key.offset;
6449                                 ref_path->current_level = 0;
6450                         }
6451                         ret = 0;
6452                         goto out;
6453                 }
6454
6455                 level++;
6456                 BUG_ON(ref_path->nodes[level] != 0);
6457                 ref_path->nodes[level] = found_key.offset;
6458                 ref_path->current_level = level;
6459
6460                 /*
6461                  * the reference was created in the running transaction,
6462                  * no need to continue walking up.
6463                  */
6464                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6465                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6466                         ref_path->root_generation =
6467                                 btrfs_ref_generation(leaf, ref);
6468                         ret = 0;
6469                         goto out;
6470                 }
6471
6472                 btrfs_release_path(extent_root, path);
6473                 cond_resched();
6474         }
6475         /* reached max tree level, but no tree root found. */
6476         BUG();
6477 out:
6478         btrfs_free_path(path);
6479         return ret;
6480 }
6481
6482 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6483                                 struct btrfs_root *extent_root,
6484                                 struct btrfs_ref_path *ref_path,
6485                                 u64 extent_start)
6486 {
6487         memset(ref_path, 0, sizeof(*ref_path));
6488         ref_path->extent_start = extent_start;
6489
6490         return __next_ref_path(trans, extent_root, ref_path, 1);
6491 }
6492
6493 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6494                                struct btrfs_root *extent_root,
6495                                struct btrfs_ref_path *ref_path)
6496 {
6497         return __next_ref_path(trans, extent_root, ref_path, 0);
6498 }
6499
6500 static noinline int get_new_locations(struct inode *reloc_inode,
6501                                       struct btrfs_key *extent_key,
6502                                       u64 offset, int no_fragment,
6503                                       struct disk_extent **extents,
6504                                       int *nr_extents)
6505 {
6506         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6507         struct btrfs_path *path;
6508         struct btrfs_file_extent_item *fi;
6509         struct extent_buffer *leaf;
6510         struct disk_extent *exts = *extents;
6511         struct btrfs_key found_key;
6512         u64 cur_pos;
6513         u64 last_byte;
6514         u32 nritems;
6515         int nr = 0;
6516         int max = *nr_extents;
6517         int ret;
6518
6519         WARN_ON(!no_fragment && *extents);
6520         if (!exts) {
6521                 max = 1;
6522                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6523                 if (!exts)
6524                         return -ENOMEM;
6525         }
6526
6527         path = btrfs_alloc_path();
6528         BUG_ON(!path);
6529
6530         cur_pos = extent_key->objectid - offset;
6531         last_byte = extent_key->objectid + extent_key->offset;
6532         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6533                                        cur_pos, 0);
6534         if (ret < 0)
6535                 goto out;
6536         if (ret > 0) {
6537                 ret = -ENOENT;
6538                 goto out;
6539         }
6540
6541         while (1) {
6542                 leaf = path->nodes[0];
6543                 nritems = btrfs_header_nritems(leaf);
6544                 if (path->slots[0] >= nritems) {
6545                         ret = btrfs_next_leaf(root, path);
6546                         if (ret < 0)
6547                                 goto out;
6548                         if (ret > 0)
6549                                 break;
6550                         leaf = path->nodes[0];
6551                 }
6552
6553                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6554                 if (found_key.offset != cur_pos ||
6555                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6556                     found_key.objectid != reloc_inode->i_ino)
6557                         break;
6558
6559                 fi = btrfs_item_ptr(leaf, path->slots[0],
6560                                     struct btrfs_file_extent_item);
6561                 if (btrfs_file_extent_type(leaf, fi) !=
6562                     BTRFS_FILE_EXTENT_REG ||
6563                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6564                         break;
6565
6566                 if (nr == max) {
6567                         struct disk_extent *old = exts;
6568                         max *= 2;
6569                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6570                         memcpy(exts, old, sizeof(*exts) * nr);
6571                         if (old != *extents)
6572                                 kfree(old);
6573                 }
6574
6575                 exts[nr].disk_bytenr =
6576                         btrfs_file_extent_disk_bytenr(leaf, fi);
6577                 exts[nr].disk_num_bytes =
6578                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6579                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6580                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6581                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6582                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6583                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6584                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6585                                                                            fi);
6586                 BUG_ON(exts[nr].offset > 0);
6587                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6588                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6589
6590                 cur_pos += exts[nr].num_bytes;
6591                 nr++;
6592
6593                 if (cur_pos + offset >= last_byte)
6594                         break;
6595
6596                 if (no_fragment) {
6597                         ret = 1;
6598                         goto out;
6599                 }
6600                 path->slots[0]++;
6601         }
6602
6603         BUG_ON(cur_pos + offset > last_byte);
6604         if (cur_pos + offset < last_byte) {
6605                 ret = -ENOENT;
6606                 goto out;
6607         }
6608         ret = 0;
6609 out:
6610         btrfs_free_path(path);
6611         if (ret) {
6612                 if (exts != *extents)
6613                         kfree(exts);
6614         } else {
6615                 *extents = exts;
6616                 *nr_extents = nr;
6617         }
6618         return ret;
6619 }
6620
6621 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6622                                         struct btrfs_root *root,
6623                                         struct btrfs_path *path,
6624                                         struct btrfs_key *extent_key,
6625                                         struct btrfs_key *leaf_key,
6626                                         struct btrfs_ref_path *ref_path,
6627                                         struct disk_extent *new_extents,
6628                                         int nr_extents)
6629 {
6630         struct extent_buffer *leaf;
6631         struct btrfs_file_extent_item *fi;
6632         struct inode *inode = NULL;
6633         struct btrfs_key key;
6634         u64 lock_start = 0;
6635         u64 lock_end = 0;
6636         u64 num_bytes;
6637         u64 ext_offset;
6638         u64 search_end = (u64)-1;
6639         u32 nritems;
6640         int nr_scaned = 0;
6641         int extent_locked = 0;
6642         int extent_type;
6643         int ret;
6644
6645         memcpy(&key, leaf_key, sizeof(key));
6646         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6647                 if (key.objectid < ref_path->owner_objectid ||
6648                     (key.objectid == ref_path->owner_objectid &&
6649                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6650                         key.objectid = ref_path->owner_objectid;
6651                         key.type = BTRFS_EXTENT_DATA_KEY;
6652                         key.offset = 0;
6653                 }
6654         }
6655
6656         while (1) {
6657                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6658                 if (ret < 0)
6659                         goto out;
6660
6661                 leaf = path->nodes[0];
6662                 nritems = btrfs_header_nritems(leaf);
6663 next:
6664                 if (extent_locked && ret > 0) {
6665                         /*
6666                          * the file extent item was modified by someone
6667                          * before the extent got locked.
6668                          */
6669                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6670                                       lock_end, GFP_NOFS);
6671                         extent_locked = 0;
6672                 }
6673
6674                 if (path->slots[0] >= nritems) {
6675                         if (++nr_scaned > 2)
6676                                 break;
6677
6678                         BUG_ON(extent_locked);
6679                         ret = btrfs_next_leaf(root, path);
6680                         if (ret < 0)
6681                                 goto out;
6682                         if (ret > 0)
6683                                 break;
6684                         leaf = path->nodes[0];
6685                         nritems = btrfs_header_nritems(leaf);
6686                 }
6687
6688                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6689
6690                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6691                         if ((key.objectid > ref_path->owner_objectid) ||
6692                             (key.objectid == ref_path->owner_objectid &&
6693                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6694                             key.offset >= search_end)
6695                                 break;
6696                 }
6697
6698                 if (inode && key.objectid != inode->i_ino) {
6699                         BUG_ON(extent_locked);
6700                         btrfs_release_path(root, path);
6701                         mutex_unlock(&inode->i_mutex);
6702                         iput(inode);
6703                         inode = NULL;
6704                         continue;
6705                 }
6706
6707                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6708                         path->slots[0]++;
6709                         ret = 1;
6710                         goto next;
6711                 }
6712                 fi = btrfs_item_ptr(leaf, path->slots[0],
6713                                     struct btrfs_file_extent_item);
6714                 extent_type = btrfs_file_extent_type(leaf, fi);
6715                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6716                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6717                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6718                      extent_key->objectid)) {
6719                         path->slots[0]++;
6720                         ret = 1;
6721                         goto next;
6722                 }
6723
6724                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6725                 ext_offset = btrfs_file_extent_offset(leaf, fi);
6726
6727                 if (search_end == (u64)-1) {
6728                         search_end = key.offset - ext_offset +
6729                                 btrfs_file_extent_ram_bytes(leaf, fi);
6730                 }
6731
6732                 if (!extent_locked) {
6733                         lock_start = key.offset;
6734                         lock_end = lock_start + num_bytes - 1;
6735                 } else {
6736                         if (lock_start > key.offset ||
6737                             lock_end + 1 < key.offset + num_bytes) {
6738                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6739                                               lock_start, lock_end, GFP_NOFS);
6740                                 extent_locked = 0;
6741                         }
6742                 }
6743
6744                 if (!inode) {
6745                         btrfs_release_path(root, path);
6746
6747                         inode = btrfs_iget_locked(root->fs_info->sb,
6748                                                   key.objectid, root);
6749                         if (inode->i_state & I_NEW) {
6750                                 BTRFS_I(inode)->root = root;
6751                                 BTRFS_I(inode)->location.objectid =
6752                                         key.objectid;
6753                                 BTRFS_I(inode)->location.type =
6754                                         BTRFS_INODE_ITEM_KEY;
6755                                 BTRFS_I(inode)->location.offset = 0;
6756                                 btrfs_read_locked_inode(inode);
6757                                 unlock_new_inode(inode);
6758                         }
6759                         /*
6760                          * some code call btrfs_commit_transaction while
6761                          * holding the i_mutex, so we can't use mutex_lock
6762                          * here.
6763                          */
6764                         if (is_bad_inode(inode) ||
6765                             !mutex_trylock(&inode->i_mutex)) {
6766                                 iput(inode);
6767                                 inode = NULL;
6768                                 key.offset = (u64)-1;
6769                                 goto skip;
6770                         }
6771                 }
6772
6773                 if (!extent_locked) {
6774                         struct btrfs_ordered_extent *ordered;
6775
6776                         btrfs_release_path(root, path);
6777
6778                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6779                                     lock_end, GFP_NOFS);
6780                         ordered = btrfs_lookup_first_ordered_extent(inode,
6781                                                                     lock_end);
6782                         if (ordered &&
6783                             ordered->file_offset <= lock_end &&
6784                             ordered->file_offset + ordered->len > lock_start) {
6785                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6786                                               lock_start, lock_end, GFP_NOFS);
6787                                 btrfs_start_ordered_extent(inode, ordered, 1);
6788                                 btrfs_put_ordered_extent(ordered);
6789                                 key.offset += num_bytes;
6790                                 goto skip;
6791                         }
6792                         if (ordered)
6793                                 btrfs_put_ordered_extent(ordered);
6794
6795                         extent_locked = 1;
6796                         continue;
6797                 }
6798
6799                 if (nr_extents == 1) {
6800                         /* update extent pointer in place */
6801                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
6802                                                 new_extents[0].disk_bytenr);
6803                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6804                                                 new_extents[0].disk_num_bytes);
6805                         btrfs_mark_buffer_dirty(leaf);
6806
6807                         btrfs_drop_extent_cache(inode, key.offset,
6808                                                 key.offset + num_bytes - 1, 0);
6809
6810                         ret = btrfs_inc_extent_ref(trans, root,
6811                                                 new_extents[0].disk_bytenr,
6812                                                 new_extents[0].disk_num_bytes,
6813                                                 leaf->start,
6814                                                 root->root_key.objectid,
6815                                                 trans->transid,
6816                                                 key.objectid);
6817                         BUG_ON(ret);
6818
6819                         ret = btrfs_free_extent(trans, root,
6820                                                 extent_key->objectid,
6821                                                 extent_key->offset,
6822                                                 leaf->start,
6823                                                 btrfs_header_owner(leaf),
6824                                                 btrfs_header_generation(leaf),
6825                                                 key.objectid, 0);
6826                         BUG_ON(ret);
6827
6828                         btrfs_release_path(root, path);
6829                         key.offset += num_bytes;
6830                 } else {
6831                         BUG_ON(1);
6832 #if 0
6833                         u64 alloc_hint;
6834                         u64 extent_len;
6835                         int i;
6836                         /*
6837                          * drop old extent pointer at first, then insert the
6838                          * new pointers one bye one
6839                          */
6840                         btrfs_release_path(root, path);
6841                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
6842                                                  key.offset + num_bytes,
6843                                                  key.offset, &alloc_hint);
6844                         BUG_ON(ret);
6845
6846                         for (i = 0; i < nr_extents; i++) {
6847                                 if (ext_offset >= new_extents[i].num_bytes) {
6848                                         ext_offset -= new_extents[i].num_bytes;
6849                                         continue;
6850                                 }
6851                                 extent_len = min(new_extents[i].num_bytes -
6852                                                  ext_offset, num_bytes);
6853
6854                                 ret = btrfs_insert_empty_item(trans, root,
6855                                                               path, &key,
6856                                                               sizeof(*fi));
6857                                 BUG_ON(ret);
6858
6859                                 leaf = path->nodes[0];
6860                                 fi = btrfs_item_ptr(leaf, path->slots[0],
6861                                                 struct btrfs_file_extent_item);
6862                                 btrfs_set_file_extent_generation(leaf, fi,
6863                                                         trans->transid);
6864                                 btrfs_set_file_extent_type(leaf, fi,
6865                                                         BTRFS_FILE_EXTENT_REG);
6866                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6867                                                 new_extents[i].disk_bytenr);
6868                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6869                                                 new_extents[i].disk_num_bytes);
6870                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
6871                                                 new_extents[i].ram_bytes);
6872
6873                                 btrfs_set_file_extent_compression(leaf, fi,
6874                                                 new_extents[i].compression);
6875                                 btrfs_set_file_extent_encryption(leaf, fi,
6876                                                 new_extents[i].encryption);
6877                                 btrfs_set_file_extent_other_encoding(leaf, fi,
6878                                                 new_extents[i].other_encoding);
6879
6880                                 btrfs_set_file_extent_num_bytes(leaf, fi,
6881                                                         extent_len);
6882                                 ext_offset += new_extents[i].offset;
6883                                 btrfs_set_file_extent_offset(leaf, fi,
6884                                                         ext_offset);
6885                                 btrfs_mark_buffer_dirty(leaf);
6886
6887                                 btrfs_drop_extent_cache(inode, key.offset,
6888                                                 key.offset + extent_len - 1, 0);
6889
6890                                 ret = btrfs_inc_extent_ref(trans, root,
6891                                                 new_extents[i].disk_bytenr,
6892                                                 new_extents[i].disk_num_bytes,
6893                                                 leaf->start,
6894                                                 root->root_key.objectid,
6895                                                 trans->transid, key.objectid);
6896                                 BUG_ON(ret);
6897                                 btrfs_release_path(root, path);
6898
6899                                 inode_add_bytes(inode, extent_len);
6900
6901                                 ext_offset = 0;
6902                                 num_bytes -= extent_len;
6903                                 key.offset += extent_len;
6904
6905                                 if (num_bytes == 0)
6906                                         break;
6907                         }
6908                         BUG_ON(i >= nr_extents);
6909 #endif
6910                 }
6911
6912                 if (extent_locked) {
6913                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6914                                       lock_end, GFP_NOFS);
6915                         extent_locked = 0;
6916                 }
6917 skip:
6918                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6919                     key.offset >= search_end)
6920                         break;
6921
6922                 cond_resched();
6923         }
6924         ret = 0;
6925 out:
6926         btrfs_release_path(root, path);
6927         if (inode) {
6928                 mutex_unlock(&inode->i_mutex);
6929                 if (extent_locked) {
6930                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6931                                       lock_end, GFP_NOFS);
6932                 }
6933                 iput(inode);
6934         }
6935         return ret;
6936 }
6937
6938 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6939                                struct btrfs_root *root,
6940                                struct extent_buffer *buf, u64 orig_start)
6941 {
6942         int level;
6943         int ret;
6944
6945         BUG_ON(btrfs_header_generation(buf) != trans->transid);
6946         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6947
6948         level = btrfs_header_level(buf);
6949         if (level == 0) {
6950                 struct btrfs_leaf_ref *ref;
6951                 struct btrfs_leaf_ref *orig_ref;
6952
6953                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6954                 if (!orig_ref)
6955                         return -ENOENT;
6956
6957                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6958                 if (!ref) {
6959                         btrfs_free_leaf_ref(root, orig_ref);
6960                         return -ENOMEM;
6961                 }
6962
6963                 ref->nritems = orig_ref->nritems;
6964                 memcpy(ref->extents, orig_ref->extents,
6965                         sizeof(ref->extents[0]) * ref->nritems);
6966
6967                 btrfs_free_leaf_ref(root, orig_ref);
6968
6969                 ref->root_gen = trans->transid;
6970                 ref->bytenr = buf->start;
6971                 ref->owner = btrfs_header_owner(buf);
6972                 ref->generation = btrfs_header_generation(buf);
6973
6974                 ret = btrfs_add_leaf_ref(root, ref, 0);
6975                 WARN_ON(ret);
6976                 btrfs_free_leaf_ref(root, ref);
6977         }
6978         return 0;
6979 }
6980
6981 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6982                                         struct extent_buffer *leaf,
6983                                         struct btrfs_block_group_cache *group,
6984                                         struct btrfs_root *target_root)
6985 {
6986         struct btrfs_key key;
6987         struct inode *inode = NULL;
6988         struct btrfs_file_extent_item *fi;
6989         struct extent_state *cached_state = NULL;
6990         u64 num_bytes;
6991         u64 skip_objectid = 0;
6992         u32 nritems;
6993         u32 i;
6994
6995         nritems = btrfs_header_nritems(leaf);
6996         for (i = 0; i < nritems; i++) {
6997                 btrfs_item_key_to_cpu(leaf, &key, i);
6998                 if (key.objectid == skip_objectid ||
6999                     key.type != BTRFS_EXTENT_DATA_KEY)
7000                         continue;
7001                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7002                 if (btrfs_file_extent_type(leaf, fi) ==
7003                     BTRFS_FILE_EXTENT_INLINE)
7004                         continue;
7005                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7006                         continue;
7007                 if (!inode || inode->i_ino != key.objectid) {
7008                         iput(inode);
7009                         inode = btrfs_ilookup(target_root->fs_info->sb,
7010                                               key.objectid, target_root, 1);
7011                 }
7012                 if (!inode) {
7013                         skip_objectid = key.objectid;
7014                         continue;
7015                 }
7016                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7017
7018                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7019                                  key.offset + num_bytes - 1, 0, &cached_state,
7020                                  GFP_NOFS);
7021                 btrfs_drop_extent_cache(inode, key.offset,
7022                                         key.offset + num_bytes - 1, 1);
7023                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7024                                      key.offset + num_bytes - 1, &cached_state,
7025                                      GFP_NOFS);
7026                 cond_resched();
7027         }
7028         iput(inode);
7029         return 0;
7030 }
7031
7032 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7033                                         struct btrfs_root *root,
7034                                         struct extent_buffer *leaf,
7035                                         struct btrfs_block_group_cache *group,
7036                                         struct inode *reloc_inode)
7037 {
7038         struct btrfs_key key;
7039         struct btrfs_key extent_key;
7040         struct btrfs_file_extent_item *fi;
7041         struct btrfs_leaf_ref *ref;
7042         struct disk_extent *new_extent;
7043         u64 bytenr;
7044         u64 num_bytes;
7045         u32 nritems;
7046         u32 i;
7047         int ext_index;
7048         int nr_extent;
7049         int ret;
7050
7051         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7052         BUG_ON(!new_extent);
7053
7054         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7055         BUG_ON(!ref);
7056
7057         ext_index = -1;
7058         nritems = btrfs_header_nritems(leaf);
7059         for (i = 0; i < nritems; i++) {
7060                 btrfs_item_key_to_cpu(leaf, &key, i);
7061                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7062                         continue;
7063                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7064                 if (btrfs_file_extent_type(leaf, fi) ==
7065                     BTRFS_FILE_EXTENT_INLINE)
7066                         continue;
7067                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7068                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7069                 if (bytenr == 0)
7070                         continue;
7071
7072                 ext_index++;
7073                 if (bytenr >= group->key.objectid + group->key.offset ||
7074                     bytenr + num_bytes <= group->key.objectid)
7075                         continue;
7076
7077                 extent_key.objectid = bytenr;
7078                 extent_key.offset = num_bytes;
7079                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7080                 nr_extent = 1;
7081                 ret = get_new_locations(reloc_inode, &extent_key,
7082                                         group->key.objectid, 1,
7083                                         &new_extent, &nr_extent);
7084                 if (ret > 0)
7085                         continue;
7086                 BUG_ON(ret < 0);
7087
7088                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7089                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7090                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7091                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7092
7093                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7094                                                 new_extent->disk_bytenr);
7095                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7096                                                 new_extent->disk_num_bytes);
7097                 btrfs_mark_buffer_dirty(leaf);
7098
7099                 ret = btrfs_inc_extent_ref(trans, root,
7100                                         new_extent->disk_bytenr,
7101                                         new_extent->disk_num_bytes,
7102                                         leaf->start,
7103                                         root->root_key.objectid,
7104                                         trans->transid, key.objectid);
7105                 BUG_ON(ret);
7106
7107                 ret = btrfs_free_extent(trans, root,
7108                                         bytenr, num_bytes, leaf->start,
7109                                         btrfs_header_owner(leaf),
7110                                         btrfs_header_generation(leaf),
7111                                         key.objectid, 0);
7112                 BUG_ON(ret);
7113                 cond_resched();
7114         }
7115         kfree(new_extent);
7116         BUG_ON(ext_index + 1 != ref->nritems);
7117         btrfs_free_leaf_ref(root, ref);
7118         return 0;
7119 }
7120
7121 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7122                           struct btrfs_root *root)
7123 {
7124         struct btrfs_root *reloc_root;
7125         int ret;
7126
7127         if (root->reloc_root) {
7128                 reloc_root = root->reloc_root;
7129                 root->reloc_root = NULL;
7130                 list_add(&reloc_root->dead_list,
7131                          &root->fs_info->dead_reloc_roots);
7132
7133                 btrfs_set_root_bytenr(&reloc_root->root_item,
7134                                       reloc_root->node->start);
7135                 btrfs_set_root_level(&root->root_item,
7136                                      btrfs_header_level(reloc_root->node));
7137                 memset(&reloc_root->root_item.drop_progress, 0,
7138                         sizeof(struct btrfs_disk_key));
7139                 reloc_root->root_item.drop_level = 0;
7140
7141                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7142                                         &reloc_root->root_key,
7143                                         &reloc_root->root_item);
7144                 BUG_ON(ret);
7145         }
7146         return 0;
7147 }
7148
7149 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7150 {
7151         struct btrfs_trans_handle *trans;
7152         struct btrfs_root *reloc_root;
7153         struct btrfs_root *prev_root = NULL;
7154         struct list_head dead_roots;
7155         int ret;
7156         unsigned long nr;
7157
7158         INIT_LIST_HEAD(&dead_roots);
7159         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7160
7161         while (!list_empty(&dead_roots)) {
7162                 reloc_root = list_entry(dead_roots.prev,
7163                                         struct btrfs_root, dead_list);
7164                 list_del_init(&reloc_root->dead_list);
7165
7166                 BUG_ON(reloc_root->commit_root != NULL);
7167                 while (1) {
7168                         trans = btrfs_join_transaction(root, 1);
7169                         BUG_ON(!trans);
7170
7171                         mutex_lock(&root->fs_info->drop_mutex);
7172                         ret = btrfs_drop_snapshot(trans, reloc_root);
7173                         if (ret != -EAGAIN)
7174                                 break;
7175                         mutex_unlock(&root->fs_info->drop_mutex);
7176
7177                         nr = trans->blocks_used;
7178                         ret = btrfs_end_transaction(trans, root);
7179                         BUG_ON(ret);
7180                         btrfs_btree_balance_dirty(root, nr);
7181                 }
7182
7183                 free_extent_buffer(reloc_root->node);
7184
7185                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7186                                      &reloc_root->root_key);
7187                 BUG_ON(ret);
7188                 mutex_unlock(&root->fs_info->drop_mutex);
7189
7190                 nr = trans->blocks_used;
7191                 ret = btrfs_end_transaction(trans, root);
7192                 BUG_ON(ret);
7193                 btrfs_btree_balance_dirty(root, nr);
7194
7195                 kfree(prev_root);
7196                 prev_root = reloc_root;
7197         }
7198         if (prev_root) {
7199                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7200                 kfree(prev_root);
7201         }
7202         return 0;
7203 }
7204
7205 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7206 {
7207         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7208         return 0;
7209 }
7210
7211 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7212 {
7213         struct btrfs_root *reloc_root;
7214         struct btrfs_trans_handle *trans;
7215         struct btrfs_key location;
7216         int found;
7217         int ret;
7218
7219         mutex_lock(&root->fs_info->tree_reloc_mutex);
7220         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7221         BUG_ON(ret);
7222         found = !list_empty(&root->fs_info->dead_reloc_roots);
7223         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7224
7225         if (found) {
7226                 trans = btrfs_start_transaction(root, 1);
7227                 BUG_ON(!trans);
7228                 ret = btrfs_commit_transaction(trans, root);
7229                 BUG_ON(ret);
7230         }
7231
7232         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7233         location.offset = (u64)-1;
7234         location.type = BTRFS_ROOT_ITEM_KEY;
7235
7236         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7237         BUG_ON(!reloc_root);
7238         btrfs_orphan_cleanup(reloc_root);
7239         return 0;
7240 }
7241
7242 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7243                                     struct btrfs_root *root)
7244 {
7245         struct btrfs_root *reloc_root;
7246         struct extent_buffer *eb;
7247         struct btrfs_root_item *root_item;
7248         struct btrfs_key root_key;
7249         int ret;
7250
7251         BUG_ON(!root->ref_cows);
7252         if (root->reloc_root)
7253                 return 0;
7254
7255         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7256         BUG_ON(!root_item);
7257
7258         ret = btrfs_copy_root(trans, root, root->commit_root,
7259                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7260         BUG_ON(ret);
7261
7262         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7263         root_key.offset = root->root_key.objectid;
7264         root_key.type = BTRFS_ROOT_ITEM_KEY;
7265
7266         memcpy(root_item, &root->root_item, sizeof(root_item));
7267         btrfs_set_root_refs(root_item, 0);
7268         btrfs_set_root_bytenr(root_item, eb->start);
7269         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7270         btrfs_set_root_generation(root_item, trans->transid);
7271
7272         btrfs_tree_unlock(eb);
7273         free_extent_buffer(eb);
7274
7275         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7276                                 &root_key, root_item);
7277         BUG_ON(ret);
7278         kfree(root_item);
7279
7280         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7281                                                  &root_key);
7282         BUG_ON(!reloc_root);
7283         reloc_root->last_trans = trans->transid;
7284         reloc_root->commit_root = NULL;
7285         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7286
7287         root->reloc_root = reloc_root;
7288         return 0;
7289 }
7290
7291 /*
7292  * Core function of space balance.
7293  *
7294  * The idea is using reloc trees to relocate tree blocks in reference
7295  * counted roots. There is one reloc tree for each subvol, and all
7296  * reloc trees share same root key objectid. Reloc trees are snapshots
7297  * of the latest committed roots of subvols (root->commit_root).
7298  *
7299  * To relocate a tree block referenced by a subvol, there are two steps.
7300  * COW the block through subvol's reloc tree, then update block pointer
7301  * in the subvol to point to the new block. Since all reloc trees share
7302  * same root key objectid, doing special handing for tree blocks owned
7303  * by them is easy. Once a tree block has been COWed in one reloc tree,
7304  * we can use the resulting new block directly when the same block is
7305  * required to COW again through other reloc trees. By this way, relocated
7306  * tree blocks are shared between reloc trees, so they are also shared
7307  * between subvols.
7308  */
7309 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7310                                       struct btrfs_root *root,
7311                                       struct btrfs_path *path,
7312                                       struct btrfs_key *first_key,
7313                                       struct btrfs_ref_path *ref_path,
7314                                       struct btrfs_block_group_cache *group,
7315                                       struct inode *reloc_inode)
7316 {
7317         struct btrfs_root *reloc_root;
7318         struct extent_buffer *eb = NULL;
7319         struct btrfs_key *keys;
7320         u64 *nodes;
7321         int level;
7322         int shared_level;
7323         int lowest_level = 0;
7324         int ret;
7325
7326         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7327                 lowest_level = ref_path->owner_objectid;
7328
7329         if (!root->ref_cows) {
7330                 path->lowest_level = lowest_level;
7331                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7332                 BUG_ON(ret < 0);
7333                 path->lowest_level = 0;
7334                 btrfs_release_path(root, path);
7335                 return 0;
7336         }
7337
7338         mutex_lock(&root->fs_info->tree_reloc_mutex);
7339         ret = init_reloc_tree(trans, root);
7340         BUG_ON(ret);
7341         reloc_root = root->reloc_root;
7342
7343         shared_level = ref_path->shared_level;
7344         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7345
7346         keys = ref_path->node_keys;
7347         nodes = ref_path->new_nodes;
7348         memset(&keys[shared_level + 1], 0,
7349                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7350         memset(&nodes[shared_level + 1], 0,
7351                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7352
7353         if (nodes[lowest_level] == 0) {
7354                 path->lowest_level = lowest_level;
7355                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7356                                         0, 1);
7357                 BUG_ON(ret);
7358                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7359                         eb = path->nodes[level];
7360                         if (!eb || eb == reloc_root->node)
7361                                 break;
7362                         nodes[level] = eb->start;
7363                         if (level == 0)
7364                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7365                         else
7366                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7367                 }
7368                 if (nodes[0] &&
7369                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7370                         eb = path->nodes[0];
7371                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7372                                                       group, reloc_inode);
7373                         BUG_ON(ret);
7374                 }
7375                 btrfs_release_path(reloc_root, path);
7376         } else {
7377                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7378                                        lowest_level);
7379                 BUG_ON(ret);
7380         }
7381
7382         /*
7383          * replace tree blocks in the fs tree with tree blocks in
7384          * the reloc tree.
7385          */
7386         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7387         BUG_ON(ret < 0);
7388
7389         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7390                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7391                                         0, 0);
7392                 BUG_ON(ret);
7393                 extent_buffer_get(path->nodes[0]);
7394                 eb = path->nodes[0];
7395                 btrfs_release_path(reloc_root, path);
7396                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7397                 BUG_ON(ret);
7398                 free_extent_buffer(eb);
7399         }
7400
7401         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7402         path->lowest_level = 0;
7403         return 0;
7404 }
7405
7406 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7407                                         struct btrfs_root *root,
7408                                         struct btrfs_path *path,
7409                                         struct btrfs_key *first_key,
7410                                         struct btrfs_ref_path *ref_path)
7411 {
7412         int ret;
7413
7414         ret = relocate_one_path(trans, root, path, first_key,
7415                                 ref_path, NULL, NULL);
7416         BUG_ON(ret);
7417
7418         return 0;
7419 }
7420
7421 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7422                                     struct btrfs_root *extent_root,
7423                                     struct btrfs_path *path,
7424                                     struct btrfs_key *extent_key)
7425 {
7426         int ret;
7427
7428         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7429         if (ret)
7430                 goto out;
7431         ret = btrfs_del_item(trans, extent_root, path);
7432 out:
7433         btrfs_release_path(extent_root, path);
7434         return ret;
7435 }
7436
7437 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7438                                                 struct btrfs_ref_path *ref_path)
7439 {
7440         struct btrfs_key root_key;
7441
7442         root_key.objectid = ref_path->root_objectid;
7443         root_key.type = BTRFS_ROOT_ITEM_KEY;
7444         if (is_cowonly_root(ref_path->root_objectid))
7445                 root_key.offset = 0;
7446         else
7447                 root_key.offset = (u64)-1;
7448
7449         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7450 }
7451
7452 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7453                                         struct btrfs_path *path,
7454                                         struct btrfs_key *extent_key,
7455                                         struct btrfs_block_group_cache *group,
7456                                         struct inode *reloc_inode, int pass)
7457 {
7458         struct btrfs_trans_handle *trans;
7459         struct btrfs_root *found_root;
7460         struct btrfs_ref_path *ref_path = NULL;
7461         struct disk_extent *new_extents = NULL;
7462         int nr_extents = 0;
7463         int loops;
7464         int ret;
7465         int level;
7466         struct btrfs_key first_key;
7467         u64 prev_block = 0;
7468
7469
7470         trans = btrfs_start_transaction(extent_root, 1);
7471         BUG_ON(!trans);
7472
7473         if (extent_key->objectid == 0) {
7474                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7475                 goto out;
7476         }
7477
7478         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7479         if (!ref_path) {
7480                 ret = -ENOMEM;
7481                 goto out;
7482         }
7483
7484         for (loops = 0; ; loops++) {
7485                 if (loops == 0) {
7486                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7487                                                    extent_key->objectid);
7488                 } else {
7489                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7490                 }
7491                 if (ret < 0)
7492                         goto out;
7493                 if (ret > 0)
7494                         break;
7495
7496                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7497                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7498                         continue;
7499
7500                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7501                 BUG_ON(!found_root);
7502                 /*
7503                  * for reference counted tree, only process reference paths
7504                  * rooted at the latest committed root.
7505                  */
7506                 if (found_root->ref_cows &&
7507                     ref_path->root_generation != found_root->root_key.offset)
7508                         continue;
7509
7510                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7511                         if (pass == 0) {
7512                                 /*
7513                                  * copy data extents to new locations
7514                                  */
7515                                 u64 group_start = group->key.objectid;
7516                                 ret = relocate_data_extent(reloc_inode,
7517                                                            extent_key,
7518                                                            group_start);
7519                                 if (ret < 0)
7520                                         goto out;
7521                                 break;
7522                         }
7523                         level = 0;
7524                 } else {
7525                         level = ref_path->owner_objectid;
7526                 }
7527
7528                 if (prev_block != ref_path->nodes[level]) {
7529                         struct extent_buffer *eb;
7530                         u64 block_start = ref_path->nodes[level];
7531                         u64 block_size = btrfs_level_size(found_root, level);
7532
7533                         eb = read_tree_block(found_root, block_start,
7534                                              block_size, 0);
7535                         btrfs_tree_lock(eb);
7536                         BUG_ON(level != btrfs_header_level(eb));
7537
7538                         if (level == 0)
7539                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7540                         else
7541                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7542
7543                         btrfs_tree_unlock(eb);
7544                         free_extent_buffer(eb);
7545                         prev_block = block_start;
7546                 }
7547
7548                 mutex_lock(&extent_root->fs_info->trans_mutex);
7549                 btrfs_record_root_in_trans(found_root);
7550                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7551                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7552                         /*
7553                          * try to update data extent references while
7554                          * keeping metadata shared between snapshots.
7555                          */
7556                         if (pass == 1) {
7557                                 ret = relocate_one_path(trans, found_root,
7558                                                 path, &first_key, ref_path,
7559                                                 group, reloc_inode);
7560                                 if (ret < 0)
7561                                         goto out;
7562                                 continue;
7563                         }
7564                         /*
7565                          * use fallback method to process the remaining
7566                          * references.
7567                          */
7568                         if (!new_extents) {
7569                                 u64 group_start = group->key.objectid;
7570                                 new_extents = kmalloc(sizeof(*new_extents),
7571                                                       GFP_NOFS);
7572                                 nr_extents = 1;
7573                                 ret = get_new_locations(reloc_inode,
7574                                                         extent_key,
7575                                                         group_start, 1,
7576                                                         &new_extents,
7577                                                         &nr_extents);
7578                                 if (ret)
7579                                         goto out;
7580                         }
7581                         ret = replace_one_extent(trans, found_root,
7582                                                 path, extent_key,
7583                                                 &first_key, ref_path,
7584                                                 new_extents, nr_extents);
7585                 } else {
7586                         ret = relocate_tree_block(trans, found_root, path,
7587                                                   &first_key, ref_path);
7588                 }
7589                 if (ret < 0)
7590                         goto out;
7591         }
7592         ret = 0;
7593 out:
7594         btrfs_end_transaction(trans, extent_root);
7595         kfree(new_extents);
7596         kfree(ref_path);
7597         return ret;
7598 }
7599 #endif
7600
7601 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7602 {
7603         u64 num_devices;
7604         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7605                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7606
7607         num_devices = root->fs_info->fs_devices->rw_devices;
7608         if (num_devices == 1) {
7609                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7610                 stripped = flags & ~stripped;
7611
7612                 /* turn raid0 into single device chunks */
7613                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7614                         return stripped;
7615
7616                 /* turn mirroring into duplication */
7617                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7618                              BTRFS_BLOCK_GROUP_RAID10))
7619                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7620                 return flags;
7621         } else {
7622                 /* they already had raid on here, just return */
7623                 if (flags & stripped)
7624                         return flags;
7625
7626                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7627                 stripped = flags & ~stripped;
7628
7629                 /* switch duplicated blocks with raid1 */
7630                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7631                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7632
7633                 /* turn single device chunks into raid0 */
7634                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7635         }
7636         return flags;
7637 }
7638
7639 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
7640 {
7641         struct btrfs_space_info *sinfo = cache->space_info;
7642         u64 num_bytes;
7643         int ret = -ENOSPC;
7644
7645         if (cache->ro)
7646                 return 0;
7647
7648         spin_lock(&sinfo->lock);
7649         spin_lock(&cache->lock);
7650         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7651                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7652
7653         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7654             sinfo->bytes_may_use + sinfo->bytes_readonly +
7655             cache->reserved_pinned + num_bytes < sinfo->total_bytes) {
7656                 sinfo->bytes_readonly += num_bytes;
7657                 sinfo->bytes_reserved += cache->reserved_pinned;
7658                 cache->reserved_pinned = 0;
7659                 cache->ro = 1;
7660                 ret = 0;
7661         }
7662         spin_unlock(&cache->lock);
7663         spin_unlock(&sinfo->lock);
7664         return ret;
7665 }
7666
7667 int btrfs_set_block_group_ro(struct btrfs_root *root,
7668                              struct btrfs_block_group_cache *cache)
7669
7670 {
7671         struct btrfs_trans_handle *trans;
7672         u64 alloc_flags;
7673         int ret;
7674
7675         BUG_ON(cache->ro);
7676
7677         trans = btrfs_join_transaction(root, 1);
7678         BUG_ON(IS_ERR(trans));
7679
7680         alloc_flags = update_block_group_flags(root, cache->flags);
7681         if (alloc_flags != cache->flags)
7682                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7683
7684         ret = set_block_group_ro(cache);
7685         if (!ret)
7686                 goto out;
7687         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7688         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7689         if (ret < 0)
7690                 goto out;
7691         ret = set_block_group_ro(cache);
7692 out:
7693         btrfs_end_transaction(trans, root);
7694         return ret;
7695 }
7696
7697 int btrfs_set_block_group_rw(struct btrfs_root *root,
7698                               struct btrfs_block_group_cache *cache)
7699 {
7700         struct btrfs_space_info *sinfo = cache->space_info;
7701         u64 num_bytes;
7702
7703         BUG_ON(!cache->ro);
7704
7705         spin_lock(&sinfo->lock);
7706         spin_lock(&cache->lock);
7707         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7708                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7709         sinfo->bytes_readonly -= num_bytes;
7710         cache->ro = 0;
7711         spin_unlock(&cache->lock);
7712         spin_unlock(&sinfo->lock);
7713         return 0;
7714 }
7715
7716 /*
7717  * checks to see if its even possible to relocate this block group.
7718  *
7719  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7720  * ok to go ahead and try.
7721  */
7722 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7723 {
7724         struct btrfs_block_group_cache *block_group;
7725         struct btrfs_space_info *space_info;
7726         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7727         struct btrfs_device *device;
7728         int full = 0;
7729         int ret = 0;
7730
7731         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7732
7733         /* odd, couldn't find the block group, leave it alone */
7734         if (!block_group)
7735                 return -1;
7736
7737         /* no bytes used, we're good */
7738         if (!btrfs_block_group_used(&block_group->item))
7739                 goto out;
7740
7741         space_info = block_group->space_info;
7742         spin_lock(&space_info->lock);
7743
7744         full = space_info->full;
7745
7746         /*
7747          * if this is the last block group we have in this space, we can't
7748          * relocate it unless we're able to allocate a new chunk below.
7749          *
7750          * Otherwise, we need to make sure we have room in the space to handle
7751          * all of the extents from this block group.  If we can, we're good
7752          */
7753         if ((space_info->total_bytes != block_group->key.offset) &&
7754            (space_info->bytes_used + space_info->bytes_reserved +
7755             space_info->bytes_pinned + space_info->bytes_readonly +
7756             btrfs_block_group_used(&block_group->item) <
7757             space_info->total_bytes)) {
7758                 spin_unlock(&space_info->lock);
7759                 goto out;
7760         }
7761         spin_unlock(&space_info->lock);
7762
7763         /*
7764          * ok we don't have enough space, but maybe we have free space on our
7765          * devices to allocate new chunks for relocation, so loop through our
7766          * alloc devices and guess if we have enough space.  However, if we
7767          * were marked as full, then we know there aren't enough chunks, and we
7768          * can just return.
7769          */
7770         ret = -1;
7771         if (full)
7772                 goto out;
7773
7774         mutex_lock(&root->fs_info->chunk_mutex);
7775         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7776                 u64 min_free = btrfs_block_group_used(&block_group->item);
7777                 u64 dev_offset, max_avail;
7778
7779                 /*
7780                  * check to make sure we can actually find a chunk with enough
7781                  * space to fit our block group in.
7782                  */
7783                 if (device->total_bytes > device->bytes_used + min_free) {
7784                         ret = find_free_dev_extent(NULL, device, min_free,
7785                                                    &dev_offset, &max_avail);
7786                         if (!ret)
7787                                 break;
7788                         ret = -1;
7789                 }
7790         }
7791         mutex_unlock(&root->fs_info->chunk_mutex);
7792 out:
7793         btrfs_put_block_group(block_group);
7794         return ret;
7795 }
7796
7797 static int find_first_block_group(struct btrfs_root *root,
7798                 struct btrfs_path *path, struct btrfs_key *key)
7799 {
7800         int ret = 0;
7801         struct btrfs_key found_key;
7802         struct extent_buffer *leaf;
7803         int slot;
7804
7805         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7806         if (ret < 0)
7807                 goto out;
7808
7809         while (1) {
7810                 slot = path->slots[0];
7811                 leaf = path->nodes[0];
7812                 if (slot >= btrfs_header_nritems(leaf)) {
7813                         ret = btrfs_next_leaf(root, path);
7814                         if (ret == 0)
7815                                 continue;
7816                         if (ret < 0)
7817                                 goto out;
7818                         break;
7819                 }
7820                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7821
7822                 if (found_key.objectid >= key->objectid &&
7823                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7824                         ret = 0;
7825                         goto out;
7826                 }
7827                 path->slots[0]++;
7828         }
7829 out:
7830         return ret;
7831 }
7832
7833 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7834 {
7835         struct btrfs_block_group_cache *block_group;
7836         struct btrfs_space_info *space_info;
7837         struct btrfs_caching_control *caching_ctl;
7838         struct rb_node *n;
7839
7840         down_write(&info->extent_commit_sem);
7841         while (!list_empty(&info->caching_block_groups)) {
7842                 caching_ctl = list_entry(info->caching_block_groups.next,
7843                                          struct btrfs_caching_control, list);
7844                 list_del(&caching_ctl->list);
7845                 put_caching_control(caching_ctl);
7846         }
7847         up_write(&info->extent_commit_sem);
7848
7849         spin_lock(&info->block_group_cache_lock);
7850         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7851                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7852                                        cache_node);
7853                 rb_erase(&block_group->cache_node,
7854                          &info->block_group_cache_tree);
7855                 spin_unlock(&info->block_group_cache_lock);
7856
7857                 down_write(&block_group->space_info->groups_sem);
7858                 list_del(&block_group->list);
7859                 up_write(&block_group->space_info->groups_sem);
7860
7861                 if (block_group->cached == BTRFS_CACHE_STARTED)
7862                         wait_block_group_cache_done(block_group);
7863
7864                 btrfs_remove_free_space_cache(block_group);
7865                 btrfs_put_block_group(block_group);
7866
7867                 spin_lock(&info->block_group_cache_lock);
7868         }
7869         spin_unlock(&info->block_group_cache_lock);
7870
7871         /* now that all the block groups are freed, go through and
7872          * free all the space_info structs.  This is only called during
7873          * the final stages of unmount, and so we know nobody is
7874          * using them.  We call synchronize_rcu() once before we start,
7875          * just to be on the safe side.
7876          */
7877         synchronize_rcu();
7878
7879         release_global_block_rsv(info);
7880
7881         while(!list_empty(&info->space_info)) {
7882                 space_info = list_entry(info->space_info.next,
7883                                         struct btrfs_space_info,
7884                                         list);
7885                 if (space_info->bytes_pinned > 0 ||
7886                     space_info->bytes_reserved > 0) {
7887                         WARN_ON(1);
7888                         dump_space_info(space_info, 0, 0);
7889                 }
7890                 list_del(&space_info->list);
7891                 kfree(space_info);
7892         }
7893         return 0;
7894 }
7895
7896 static void __link_block_group(struct btrfs_space_info *space_info,
7897                                struct btrfs_block_group_cache *cache)
7898 {
7899         int index = get_block_group_index(cache);
7900
7901         down_write(&space_info->groups_sem);
7902         list_add_tail(&cache->list, &space_info->block_groups[index]);
7903         up_write(&space_info->groups_sem);
7904 }
7905
7906 int btrfs_read_block_groups(struct btrfs_root *root)
7907 {
7908         struct btrfs_path *path;
7909         int ret;
7910         struct btrfs_block_group_cache *cache;
7911         struct btrfs_fs_info *info = root->fs_info;
7912         struct btrfs_space_info *space_info;
7913         struct btrfs_key key;
7914         struct btrfs_key found_key;
7915         struct extent_buffer *leaf;
7916
7917         root = info->extent_root;
7918         key.objectid = 0;
7919         key.offset = 0;
7920         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7921         path = btrfs_alloc_path();
7922         if (!path)
7923                 return -ENOMEM;
7924
7925         while (1) {
7926                 ret = find_first_block_group(root, path, &key);
7927                 if (ret > 0)
7928                         break;
7929                 if (ret != 0)
7930                         goto error;
7931
7932                 leaf = path->nodes[0];
7933                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7934                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7935                 if (!cache) {
7936                         ret = -ENOMEM;
7937                         goto error;
7938                 }
7939
7940                 atomic_set(&cache->count, 1);
7941                 spin_lock_init(&cache->lock);
7942                 spin_lock_init(&cache->tree_lock);
7943                 cache->fs_info = info;
7944                 INIT_LIST_HEAD(&cache->list);
7945                 INIT_LIST_HEAD(&cache->cluster_list);
7946
7947                 /*
7948                  * we only want to have 32k of ram per block group for keeping
7949                  * track of free space, and if we pass 1/2 of that we want to
7950                  * start converting things over to using bitmaps
7951                  */
7952                 cache->extents_thresh = ((1024 * 32) / 2) /
7953                         sizeof(struct btrfs_free_space);
7954
7955                 read_extent_buffer(leaf, &cache->item,
7956                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7957                                    sizeof(cache->item));
7958                 memcpy(&cache->key, &found_key, sizeof(found_key));
7959
7960                 key.objectid = found_key.objectid + found_key.offset;
7961                 btrfs_release_path(root, path);
7962                 cache->flags = btrfs_block_group_flags(&cache->item);
7963                 cache->sectorsize = root->sectorsize;
7964
7965                 /*
7966                  * check for two cases, either we are full, and therefore
7967                  * don't need to bother with the caching work since we won't
7968                  * find any space, or we are empty, and we can just add all
7969                  * the space in and be done with it.  This saves us _alot_ of
7970                  * time, particularly in the full case.
7971                  */
7972                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7973                         exclude_super_stripes(root, cache);
7974                         cache->last_byte_to_unpin = (u64)-1;
7975                         cache->cached = BTRFS_CACHE_FINISHED;
7976                         free_excluded_extents(root, cache);
7977                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7978                         exclude_super_stripes(root, cache);
7979                         cache->last_byte_to_unpin = (u64)-1;
7980                         cache->cached = BTRFS_CACHE_FINISHED;
7981                         add_new_free_space(cache, root->fs_info,
7982                                            found_key.objectid,
7983                                            found_key.objectid +
7984                                            found_key.offset);
7985                         free_excluded_extents(root, cache);
7986                 }
7987
7988                 ret = update_space_info(info, cache->flags, found_key.offset,
7989                                         btrfs_block_group_used(&cache->item),
7990                                         &space_info);
7991                 BUG_ON(ret);
7992                 cache->space_info = space_info;
7993                 spin_lock(&cache->space_info->lock);
7994                 cache->space_info->bytes_readonly += cache->bytes_super;
7995                 spin_unlock(&cache->space_info->lock);
7996
7997                 __link_block_group(space_info, cache);
7998
7999                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8000                 BUG_ON(ret);
8001
8002                 set_avail_alloc_bits(root->fs_info, cache->flags);
8003                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8004                         set_block_group_ro(cache);
8005         }
8006
8007         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8008                 if (!(get_alloc_profile(root, space_info->flags) &
8009                       (BTRFS_BLOCK_GROUP_RAID10 |
8010                        BTRFS_BLOCK_GROUP_RAID1 |
8011                        BTRFS_BLOCK_GROUP_DUP)))
8012                         continue;
8013                 /*
8014                  * avoid allocating from un-mirrored block group if there are
8015                  * mirrored block groups.
8016                  */
8017                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8018                         set_block_group_ro(cache);
8019                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8020                         set_block_group_ro(cache);
8021         }
8022
8023         init_global_block_rsv(info);
8024         ret = 0;
8025 error:
8026         btrfs_free_path(path);
8027         return ret;
8028 }
8029
8030 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8031                            struct btrfs_root *root, u64 bytes_used,
8032                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8033                            u64 size)
8034 {
8035         int ret;
8036         struct btrfs_root *extent_root;
8037         struct btrfs_block_group_cache *cache;
8038
8039         extent_root = root->fs_info->extent_root;
8040
8041         root->fs_info->last_trans_log_full_commit = trans->transid;
8042
8043         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8044         if (!cache)
8045                 return -ENOMEM;
8046
8047         cache->key.objectid = chunk_offset;
8048         cache->key.offset = size;
8049         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8050         cache->sectorsize = root->sectorsize;
8051
8052         /*
8053          * we only want to have 32k of ram per block group for keeping track
8054          * of free space, and if we pass 1/2 of that we want to start
8055          * converting things over to using bitmaps
8056          */
8057         cache->extents_thresh = ((1024 * 32) / 2) /
8058                 sizeof(struct btrfs_free_space);
8059         atomic_set(&cache->count, 1);
8060         spin_lock_init(&cache->lock);
8061         spin_lock_init(&cache->tree_lock);
8062         INIT_LIST_HEAD(&cache->list);
8063         INIT_LIST_HEAD(&cache->cluster_list);
8064
8065         btrfs_set_block_group_used(&cache->item, bytes_used);
8066         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8067         cache->flags = type;
8068         btrfs_set_block_group_flags(&cache->item, type);
8069
8070         cache->last_byte_to_unpin = (u64)-1;
8071         cache->cached = BTRFS_CACHE_FINISHED;
8072         exclude_super_stripes(root, cache);
8073
8074         add_new_free_space(cache, root->fs_info, chunk_offset,
8075                            chunk_offset + size);
8076
8077         free_excluded_extents(root, cache);
8078
8079         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8080                                 &cache->space_info);
8081         BUG_ON(ret);
8082
8083         spin_lock(&cache->space_info->lock);
8084         cache->space_info->bytes_readonly += cache->bytes_super;
8085         spin_unlock(&cache->space_info->lock);
8086
8087         __link_block_group(cache->space_info, cache);
8088
8089         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8090         BUG_ON(ret);
8091
8092         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8093                                 sizeof(cache->item));
8094         BUG_ON(ret);
8095
8096         set_avail_alloc_bits(extent_root->fs_info, type);
8097
8098         return 0;
8099 }
8100
8101 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8102                              struct btrfs_root *root, u64 group_start)
8103 {
8104         struct btrfs_path *path;
8105         struct btrfs_block_group_cache *block_group;
8106         struct btrfs_free_cluster *cluster;
8107         struct btrfs_key key;
8108         int ret;
8109         int factor;
8110
8111         root = root->fs_info->extent_root;
8112
8113         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8114         BUG_ON(!block_group);
8115         BUG_ON(!block_group->ro);
8116
8117         memcpy(&key, &block_group->key, sizeof(key));
8118         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8119                                   BTRFS_BLOCK_GROUP_RAID1 |
8120                                   BTRFS_BLOCK_GROUP_RAID10))
8121                 factor = 2;
8122         else
8123                 factor = 1;
8124
8125         /* make sure this block group isn't part of an allocation cluster */
8126         cluster = &root->fs_info->data_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         /*
8132          * make sure this block group isn't part of a metadata
8133          * allocation cluster
8134          */
8135         cluster = &root->fs_info->meta_alloc_cluster;
8136         spin_lock(&cluster->refill_lock);
8137         btrfs_return_cluster_to_free_space(block_group, cluster);
8138         spin_unlock(&cluster->refill_lock);
8139
8140         path = btrfs_alloc_path();
8141         BUG_ON(!path);
8142
8143         spin_lock(&root->fs_info->block_group_cache_lock);
8144         rb_erase(&block_group->cache_node,
8145                  &root->fs_info->block_group_cache_tree);
8146         spin_unlock(&root->fs_info->block_group_cache_lock);
8147
8148         down_write(&block_group->space_info->groups_sem);
8149         /*
8150          * we must use list_del_init so people can check to see if they
8151          * are still on the list after taking the semaphore
8152          */
8153         list_del_init(&block_group->list);
8154         up_write(&block_group->space_info->groups_sem);
8155
8156         if (block_group->cached == BTRFS_CACHE_STARTED)
8157                 wait_block_group_cache_done(block_group);
8158
8159         btrfs_remove_free_space_cache(block_group);
8160
8161         spin_lock(&block_group->space_info->lock);
8162         block_group->space_info->total_bytes -= block_group->key.offset;
8163         block_group->space_info->bytes_readonly -= block_group->key.offset;
8164         block_group->space_info->disk_total -= block_group->key.offset * factor;
8165         spin_unlock(&block_group->space_info->lock);
8166
8167         btrfs_clear_space_info_full(root->fs_info);
8168
8169         btrfs_put_block_group(block_group);
8170         btrfs_put_block_group(block_group);
8171
8172         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8173         if (ret > 0)
8174                 ret = -EIO;
8175         if (ret < 0)
8176                 goto out;
8177
8178         ret = btrfs_del_item(trans, root, path);
8179 out:
8180         btrfs_free_path(path);
8181         return ret;
8182 }