fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/kthread.h>
  25 #include <linux/slab.h>
  26 #include "compat.h"
  27 #include "hash.h"
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "print-tree.h"
  31 #include "transaction.h"
  32 #include "volumes.h"
  33 #include "locking.h"
  34 #include "free-space-cache.h"
  35
  36 static int update_block_group(struct btrfs_trans_handle *trans,
  37                               struct btrfs_root *root,
  38                               u64 bytenr, u64 num_bytes, int alloc);
  39 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
  40                                  u64 num_bytes, int reserve, int sinfo);
  41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  42                                 struct btrfs_root *root,
  43                                 u64 bytenr, u64 num_bytes, u64 parent,
  44                                 u64 root_objectid, u64 owner_objectid,
  45                                 u64 owner_offset, int refs_to_drop,
  46                                 struct btrfs_delayed_extent_op *extra_op);
  47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  48                                     struct extent_buffer *leaf,
  49                                     struct btrfs_extent_item *ei);
  50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  51                                       struct btrfs_root *root,
  52                                       u64 parent, u64 root_objectid,
  53                                       u64 flags, u64 owner, u64 offset,
  54                                       struct btrfs_key *ins, int ref_mod);
  55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  56                                      struct btrfs_root *root,
  57                                      u64 parent, u64 root_objectid,
  58                                      u64 flags, struct btrfs_disk_key *key,
  59                                      int level, struct btrfs_key *ins);
  60 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  61                           struct btrfs_root *extent_root, u64 alloc_bytes,
  62                           u64 flags, int force);
  63 static int find_next_key(struct btrfs_path *path, int level,
  64                          struct btrfs_key *key);
  65 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  66                             int dump_block_groups);
  67
  68 static noinline int
  69 block_group_cache_done(struct btrfs_block_group_cache *cache)
  70 {
  71         smp_mb();
  72         return cache->cached == BTRFS_CACHE_FINISHED;
  73 }
  74
  75 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  76 {
  77         return (cache->flags & bits) == bits;
  78 }
  79
  80 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
  81 {
  82         atomic_inc(&cache->count);
  83 }
  84
  85 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
  86 {
  87         if (atomic_dec_and_test(&cache->count)) {
  88                 WARN_ON(cache->pinned > 0);
  89                 WARN_ON(cache->reserved > 0);
  90                 WARN_ON(cache->reserved_pinned > 0);
  91                 kfree(cache);
  92         }
  93 }
  94
  95 /*
  96  * this adds the block group to the fs_info rb tree for the block group
  97  * cache
  98  */
  99 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 100                                 struct btrfs_block_group_cache *block_group)
 101 {
 102         struct rb_node **p;
 103         struct rb_node *parent = NULL;
 104         struct btrfs_block_group_cache *cache;
 105
 106         spin_lock(&info->block_group_cache_lock);
 107         p = &info->block_group_cache_tree.rb_node;
 108
 109         while (*p) {
 110                 parent = *p;
 111                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 112                                  cache_node);
 113                 if (block_group->key.objectid < cache->key.objectid) {
 114                         p = &(*p)->rb_left;
 115                 } else if (block_group->key.objectid > cache->key.objectid) {
 116                         p = &(*p)->rb_right;
 117                 } else {
 118                         spin_unlock(&info->block_group_cache_lock);
 119                         return -EEXIST;
 120                 }
 121         }
 122
 123         rb_link_node(&block_group->cache_node, parent, p);
 124         rb_insert_color(&block_group->cache_node,
 125                         &info->block_group_cache_tree);
 126         spin_unlock(&info->block_group_cache_lock);
 127
 128         return 0;
 129 }
 130
 131 /*
 132  * This will return the block group at or after bytenr if contains is 0, else
 133  * it will return the block group that contains the bytenr
 134  */
 135 static struct btrfs_block_group_cache *
 136 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 137                               int contains)
 138 {
 139         struct btrfs_block_group_cache *cache, *ret = NULL;
 140         struct rb_node *n;
 141         u64 end, start;
 142
 143         spin_lock(&info->block_group_cache_lock);
 144         n = info->block_group_cache_tree.rb_node;
 145
 146         while (n) {
 147                 cache = rb_entry(n, struct btrfs_block_group_cache,
 148                                  cache_node);
 149                 end = cache->key.objectid + cache->key.offset - 1;
 150                 start = cache->key.objectid;
 151
 152                 if (bytenr < start) {
 153                         if (!contains && (!ret || start < ret->key.objectid))
 154                                 ret = cache;
 155                         n = n->rb_left;
 156                 } else if (bytenr > start) {
 157                         if (contains && bytenr <= end) {
 158                                 ret = cache;
 159                                 break;
 160                         }
 161                         n = n->rb_right;
 162                 } else {
 163                         ret = cache;
 164                         break;
 165                 }
 166         }
 167         if (ret)
 168                 btrfs_get_block_group(ret);
 169         spin_unlock(&info->block_group_cache_lock);
 170
 171         return ret;
 172 }
 173
 174 static int add_excluded_extent(struct btrfs_root *root,
 175                                u64 start, u64 num_bytes)
 176 {
 177         u64 end = start + num_bytes - 1;
 178         set_extent_bits(&root->fs_info->freed_extents[0],
 179                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 180         set_extent_bits(&root->fs_info->freed_extents[1],
 181                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 182         return 0;
 183 }
 184
 185 static void free_excluded_extents(struct btrfs_root *root,
 186                                   struct btrfs_block_group_cache *cache)
 187 {
 188         u64 start, end;
 189
 190         start = cache->key.objectid;
 191         end = start + cache->key.offset - 1;
 192
 193         clear_extent_bits(&root->fs_info->freed_extents[0],
 194                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 195         clear_extent_bits(&root->fs_info->freed_extents[1],
 196                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 197 }
 198
 199 static int exclude_super_stripes(struct btrfs_root *root,
 200                                  struct btrfs_block_group_cache *cache)
 201 {
 202         u64 bytenr;
 203         u64 *logical;
 204         int stripe_len;
 205         int i, nr, ret;
 206
 207         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 208                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 209                 cache->bytes_super += stripe_len;
 210                 ret = add_excluded_extent(root, cache->key.objectid,
 211                                           stripe_len);
 212                 BUG_ON(ret);
 213         }
 214
 215         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 216                 bytenr = btrfs_sb_offset(i);
 217                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 218                                        cache->key.objectid, bytenr,
 219                                        0, &logical, &nr, &stripe_len);
 220                 BUG_ON(ret);
 221
 222                 while (nr--) {
 223                         cache->bytes_super += stripe_len;
 224                         ret = add_excluded_extent(root, logical[nr],
 225                                                   stripe_len);
 226                         BUG_ON(ret);
 227                 }
 228
 229                 kfree(logical);
 230         }
 231         return 0;
 232 }
 233
 234 static struct btrfs_caching_control *
 235 get_caching_control(struct btrfs_block_group_cache *cache)
 236 {
 237         struct btrfs_caching_control *ctl;
 238
 239         spin_lock(&cache->lock);
 240         if (cache->cached != BTRFS_CACHE_STARTED) {
 241                 spin_unlock(&cache->lock);
 242                 return NULL;
 243         }
 244
 245         /* We're loading it the fast way, so we don't have a caching_ctl. */
 246         if (!cache->caching_ctl) {
 247                 spin_unlock(&cache->lock);
 248                 return NULL;
 249         }
 250
 251         ctl = cache->caching_ctl;
 252         atomic_inc(&ctl->count);
 253         spin_unlock(&cache->lock);
 254         return ctl;
 255 }
 256
 257 static void put_caching_control(struct btrfs_caching_control *ctl)
 258 {
 259         if (atomic_dec_and_test(&ctl->count))
 260                 kfree(ctl);
 261 }
 262
 263 /*
 264  * this is only called by cache_block_group, since we could have freed extents
 265  * we need to check the pinned_extents for any extents that can't be used yet
 266  * since their free space will be released as soon as the transaction commits.
 267  */
 268 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 269                               struct btrfs_fs_info *info, u64 start, u64 end)
 270 {
 271         u64 extent_start, extent_end, size, total_added = 0;
 272         int ret;
 273
 274         while (start < end) {
 275                 ret = find_first_extent_bit(info->pinned_extents, start,
 276                                             &extent_start, &extent_end,
 277                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 278                 if (ret)
 279                         break;
 280
 281                 if (extent_start <= start) {
 282                         start = extent_end + 1;
 283                 } else if (extent_start > start && extent_start < end) {
 284                         size = extent_start - start;
 285                         total_added += size;
 286                         ret = btrfs_add_free_space(block_group, start,
 287                                                    size);
 288                         BUG_ON(ret);
 289                         start = extent_end + 1;
 290                 } else {
 291                         break;
 292                 }
 293         }
 294
 295         if (start < end) {
 296                 size = end - start;
 297                 total_added += size;
 298                 ret = btrfs_add_free_space(block_group, start, size);
 299                 BUG_ON(ret);
 300         }
 301
 302         return total_added;
 303 }
 304
 305 static int caching_kthread(void *data)
 306 {
 307         struct btrfs_block_group_cache *block_group = data;
 308         struct btrfs_fs_info *fs_info = block_group->fs_info;
 309         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 310         struct btrfs_root *extent_root = fs_info->extent_root;
 311         struct btrfs_path *path;
 312         struct extent_buffer *leaf;
 313         struct btrfs_key key;
 314         u64 total_found = 0;
 315         u64 last = 0;
 316         u32 nritems;
 317         int ret = 0;
 318
 319         path = btrfs_alloc_path();
 320         if (!path)
 321                 return -ENOMEM;
 322
 323         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 324
 325         /*
 326          * We don't want to deadlock with somebody trying to allocate a new
 327          * extent for the extent root while also trying to search the extent
 328          * root to add free space.  So we skip locking and search the commit
 329          * root, since its read-only
 330          */
 331         path->skip_locking = 1;
 332         path->search_commit_root = 1;
 333         path->reada = 2;
 334
 335         key.objectid = last;
 336         key.offset = 0;
 337         key.type = BTRFS_EXTENT_ITEM_KEY;
 338 again:
 339         mutex_lock(&caching_ctl->mutex);
 340         /* need to make sure the commit_root doesn't disappear */
 341         down_read(&fs_info->extent_commit_sem);
 342
 343         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 344         if (ret < 0)
 345                 goto err;
 346
 347         leaf = path->nodes[0];
 348         nritems = btrfs_header_nritems(leaf);
 349
 350         while (1) {
 351                 smp_mb();
 352                 if (fs_info->closing > 1) {
 353                         last = (u64)-1;
 354                         break;
 355                 }
 356
 357                 if (path->slots[0] < nritems) {
 358                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 359                 } else {
 360                         ret = find_next_key(path, 0, &key);
 361                         if (ret)
 362                                 break;
 363
 364                         caching_ctl->progress = last;
 365                         btrfs_release_path(extent_root, path);
 366                         up_read(&fs_info->extent_commit_sem);
 367                         mutex_unlock(&caching_ctl->mutex);
 368                         if (btrfs_transaction_in_commit(fs_info))
 369                                 schedule_timeout(1);
 370                         else
 371                                 cond_resched();
 372                         goto again;
 373                 }
 374
 375                 if (key.objectid < block_group->key.objectid) {
 376                         path->slots[0]++;
 377                         continue;
 378                 }
 379
 380                 if (key.objectid >= block_group->key.objectid +
 381                     block_group->key.offset)
 382                         break;
 383
 384                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 385                         total_found += add_new_free_space(block_group,
 386                                                           fs_info, last,
 387                                                           key.objectid);
 388                         last = key.objectid + key.offset;
 389
 390                         if (total_found > (1024 * 1024 * 2)) {
 391                                 total_found = 0;
 392                                 wake_up(&caching_ctl->wait);
 393                         }
 394                 }
 395                 path->slots[0]++;
 396         }
 397         ret = 0;
 398
 399         total_found += add_new_free_space(block_group, fs_info, last,
 400                                           block_group->key.objectid +
 401                                           block_group->key.offset);
 402         caching_ctl->progress = (u64)-1;
 403
 404         spin_lock(&block_group->lock);
 405         block_group->caching_ctl = NULL;
 406         block_group->cached = BTRFS_CACHE_FINISHED;
 407         spin_unlock(&block_group->lock);
 408
 409 err:
 410         btrfs_free_path(path);
 411         up_read(&fs_info->extent_commit_sem);
 412
 413         free_excluded_extents(extent_root, block_group);
 414
 415         mutex_unlock(&caching_ctl->mutex);
 416         wake_up(&caching_ctl->wait);
 417
 418         put_caching_control(caching_ctl);
 419         atomic_dec(&block_group->space_info->caching_threads);
 420         btrfs_put_block_group(block_group);
 421
 422         return 0;
 423 }
 424
 425 static int cache_block_group(struct btrfs_block_group_cache *cache,
 426                              struct btrfs_trans_handle *trans,
 427                              struct btrfs_root *root,
 428                              int load_cache_only)
 429 {
 430         struct btrfs_fs_info *fs_info = cache->fs_info;
 431         struct btrfs_caching_control *caching_ctl;
 432         struct task_struct *tsk;
 433         int ret = 0;
 434
 435         smp_mb();
 436         if (cache->cached != BTRFS_CACHE_NO)
 437                 return 0;
 438
 439         /*
 440          * We can't do the read from on-disk cache during a commit since we need
 441          * to have the normal tree locking.  Also if we are currently trying to
 442          * allocate blocks for the tree root we can't do the fast caching since
 443          * we likely hold important locks.
 444          */
 445         if (!trans->transaction->in_commit &&
 446             (root && root != root->fs_info->tree_root)) {
 447                 spin_lock(&cache->lock);
 448                 if (cache->cached != BTRFS_CACHE_NO) {
 449                         spin_unlock(&cache->lock);
 450                         return 0;
 451                 }
 452                 cache->cached = BTRFS_CACHE_STARTED;
 453                 spin_unlock(&cache->lock);
 454
 455                 ret = load_free_space_cache(fs_info, cache);
 456
 457                 spin_lock(&cache->lock);
 458                 if (ret == 1) {
 459                         cache->cached = BTRFS_CACHE_FINISHED;
 460                         cache->last_byte_to_unpin = (u64)-1;
 461                 } else {
 462                         cache->cached = BTRFS_CACHE_NO;
 463                 }
 464                 spin_unlock(&cache->lock);
 465                 if (ret == 1) {
 466                         free_excluded_extents(fs_info->extent_root, cache);
 467                         return 0;
 468                 }
 469         }
 470
 471         if (load_cache_only)
 472                 return 0;
 473
 474         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
 475         BUG_ON(!caching_ctl);
 476
 477         INIT_LIST_HEAD(&caching_ctl->list);
 478         mutex_init(&caching_ctl->mutex);
 479         init_waitqueue_head(&caching_ctl->wait);
 480         caching_ctl->block_group = cache;
 481         caching_ctl->progress = cache->key.objectid;
 482         /* one for caching kthread, one for caching block group list */
 483         atomic_set(&caching_ctl->count, 2);
 484
 485         spin_lock(&cache->lock);
 486         if (cache->cached != BTRFS_CACHE_NO) {
 487                 spin_unlock(&cache->lock);
 488                 kfree(caching_ctl);
 489                 return 0;
 490         }
 491         cache->caching_ctl = caching_ctl;
 492         cache->cached = BTRFS_CACHE_STARTED;
 493         spin_unlock(&cache->lock);
 494
 495         down_write(&fs_info->extent_commit_sem);
 496         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 497         up_write(&fs_info->extent_commit_sem);
 498
 499         atomic_inc(&cache->space_info->caching_threads);
 500         btrfs_get_block_group(cache);
 501
 502         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 503                           cache->key.objectid);
 504         if (IS_ERR(tsk)) {
 505                 ret = PTR_ERR(tsk);
 506                 printk(KERN_ERR "error running thread %d\n", ret);
 507                 BUG();
 508         }
 509
 510         return ret;
 511 }
 512
 513 /*
 514  * return the block group that starts at or after bytenr
 515  */
 516 static struct btrfs_block_group_cache *
 517 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 518 {
 519         struct btrfs_block_group_cache *cache;
 520
 521         cache = block_group_cache_tree_search(info, bytenr, 0);
 522
 523         return cache;
 524 }
 525
 526 /*
 527  * return the block group that contains the given bytenr
 528  */
 529 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 530                                                  struct btrfs_fs_info *info,
 531                                                  u64 bytenr)
 532 {
 533         struct btrfs_block_group_cache *cache;
 534
 535         cache = block_group_cache_tree_search(info, bytenr, 1);
 536
 537         return cache;
 538 }
 539
 540 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 541                                                   u64 flags)
 542 {
 543         struct list_head *head = &info->space_info;
 544         struct btrfs_space_info *found;
 545
 546         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 547                  BTRFS_BLOCK_GROUP_METADATA;
 548
 549         rcu_read_lock();
 550         list_for_each_entry_rcu(found, head, list) {
 551                 if (found->flags & flags) {
 552                         rcu_read_unlock();
 553                         return found;
 554                 }
 555         }
 556         rcu_read_unlock();
 557         return NULL;
 558 }
 559
 560 /*
 561  * after adding space to the filesystem, we need to clear the full flags
 562  * on all the space infos.
 563  */
 564 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 565 {
 566         struct list_head *head = &info->space_info;
 567         struct btrfs_space_info *found;
 568
 569         rcu_read_lock();
 570         list_for_each_entry_rcu(found, head, list)
 571                 found->full = 0;
 572         rcu_read_unlock();
 573 }
 574
 575 static u64 div_factor(u64 num, int factor)
 576 {
 577         if (factor == 10)
 578                 return num;
 579         num *= factor;
 580         do_div(num, 10);
 581         return num;
 582 }
 583
 584 static u64 div_factor_fine(u64 num, int factor)
 585 {
 586         if (factor == 100)
 587                 return num;
 588         num *= factor;
 589         do_div(num, 100);
 590         return num;
 591 }
 592
 593 u64 btrfs_find_block_group(struct btrfs_root *root,
 594                            u64 search_start, u64 search_hint, int owner)
 595 {
 596         struct btrfs_block_group_cache *cache;
 597         u64 used;
 598         u64 last = max(search_hint, search_start);
 599         u64 group_start = 0;
 600         int full_search = 0;
 601         int factor = 9;
 602         int wrapped = 0;
 603 again:
 604         while (1) {
 605                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 606                 if (!cache)
 607                         break;
 608
 609                 spin_lock(&cache->lock);
 610                 last = cache->key.objectid + cache->key.offset;
 611                 used = btrfs_block_group_used(&cache->item);
 612
 613                 if ((full_search || !cache->ro) &&
 614                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 615                         if (used + cache->pinned + cache->reserved <
 616                             div_factor(cache->key.offset, factor)) {
 617                                 group_start = cache->key.objectid;
 618                                 spin_unlock(&cache->lock);
 619                                 btrfs_put_block_group(cache);
 620                                 goto found;
 621                         }
 622                 }
 623                 spin_unlock(&cache->lock);
 624                 btrfs_put_block_group(cache);
 625                 cond_resched();
 626         }
 627         if (!wrapped) {
 628                 last = search_start;
 629                 wrapped = 1;
 630                 goto again;
 631         }
 632         if (!full_search && factor < 10) {
 633                 last = search_start;
 634                 full_search = 1;
 635                 factor = 10;
 636                 goto again;
 637         }
 638 found:
 639         return group_start;
 640 }
 641
 642 /* simple helper to search for an existing extent at a given offset */
 643 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 644 {
 645         int ret;
 646         struct btrfs_key key;
 647         struct btrfs_path *path;
 648
 649         path = btrfs_alloc_path();
 650         BUG_ON(!path);
 651         key.objectid = start;
 652         key.offset = len;
 653         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 654         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 655                                 0, 0);
 656         btrfs_free_path(path);
 657         return ret;
 658 }
 659
 660 /*
 661  * helper function to lookup reference count and flags of extent.
 662  *
 663  * the head node for delayed ref is used to store the sum of all the
 664  * reference count modifications queued up in the rbtree. the head
 665  * node may also store the extent flags to set. This way you can check
 666  * to see what the reference count and extent flags would be if all of
 667  * the delayed refs are not processed.
 668  */
 669 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 670                              struct btrfs_root *root, u64 bytenr,
 671                              u64 num_bytes, u64 *refs, u64 *flags)
 672 {
 673         struct btrfs_delayed_ref_head *head;
 674         struct btrfs_delayed_ref_root *delayed_refs;
 675         struct btrfs_path *path;
 676         struct btrfs_extent_item *ei;
 677         struct extent_buffer *leaf;
 678         struct btrfs_key key;
 679         u32 item_size;
 680         u64 num_refs;
 681         u64 extent_flags;
 682         int ret;
 683
 684         path = btrfs_alloc_path();
 685         if (!path)
 686                 return -ENOMEM;
 687
 688         key.objectid = bytenr;
 689         key.type = BTRFS_EXTENT_ITEM_KEY;
 690         key.offset = num_bytes;
 691         if (!trans) {
 692                 path->skip_locking = 1;
 693                 path->search_commit_root = 1;
 694         }
 695 again:
 696         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 697                                 &key, path, 0, 0);
 698         if (ret < 0)
 699                 goto out_free;
 700
 701         if (ret == 0) {
 702                 leaf = path->nodes[0];
 703                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 704                 if (item_size >= sizeof(*ei)) {
 705                         ei = btrfs_item_ptr(leaf, path->slots[0],
 706                                             struct btrfs_extent_item);
 707                         num_refs = btrfs_extent_refs(leaf, ei);
 708                         extent_flags = btrfs_extent_flags(leaf, ei);
 709                 } else {
 710 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 711                         struct btrfs_extent_item_v0 *ei0;
 712                         BUG_ON(item_size != sizeof(*ei0));
 713                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 714                                              struct btrfs_extent_item_v0);
 715                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 716                         /* FIXME: this isn't correct for data */
 717                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 718 #else
 719                         BUG();
 720 #endif
 721                 }
 722                 BUG_ON(num_refs == 0);
 723         } else {
 724                 num_refs = 0;
 725                 extent_flags = 0;
 726                 ret = 0;
 727         }
 728
 729         if (!trans)
 730                 goto out;
 731
 732         delayed_refs = &trans->transaction->delayed_refs;
 733         spin_lock(&delayed_refs->lock);
 734         head = btrfs_find_delayed_ref_head(trans, bytenr);
 735         if (head) {
 736                 if (!mutex_trylock(&head->mutex)) {
 737                         atomic_inc(&head->node.refs);
 738                         spin_unlock(&delayed_refs->lock);
 739
 740                         btrfs_release_path(root->fs_info->extent_root, path);
 741
 742                         mutex_lock(&head->mutex);
 743                         mutex_unlock(&head->mutex);
 744                         btrfs_put_delayed_ref(&head->node);
 745                         goto again;
 746                 }
 747                 if (head->extent_op && head->extent_op->update_flags)
 748                         extent_flags |= head->extent_op->flags_to_set;
 749                 else
 750                         BUG_ON(num_refs == 0);
 751
 752                 num_refs += head->node.ref_mod;
 753                 mutex_unlock(&head->mutex);
 754         }
 755         spin_unlock(&delayed_refs->lock);
 756 out:
 757         WARN_ON(num_refs == 0);
 758         if (refs)
 759                 *refs = num_refs;
 760         if (flags)
 761                 *flags = extent_flags;
 762 out_free:
 763         btrfs_free_path(path);
 764         return ret;
 765 }
 766
 767 /*
 768  * Back reference rules.  Back refs have three main goals:
 769  *
 770  * 1) differentiate between all holders of references to an extent so that
 771  *    when a reference is dropped we can make sure it was a valid reference
 772  *    before freeing the extent.
 773  *
 774  * 2) Provide enough information to quickly find the holders of an extent
 775  *    if we notice a given block is corrupted or bad.
 776  *
 777  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 778  *    maintenance.  This is actually the same as #2, but with a slightly
 779  *    different use case.
 780  *
 781  * There are two kinds of back refs. The implicit back refs is optimized
 782  * for pointers in non-shared tree blocks. For a given pointer in a block,
 783  * back refs of this kind provide information about the block's owner tree
 784  * and the pointer's key. These information allow us to find the block by
 785  * b-tree searching. The full back refs is for pointers in tree blocks not
 786  * referenced by their owner trees. The location of tree block is recorded
 787  * in the back refs. Actually the full back refs is generic, and can be
 788  * used in all cases the implicit back refs is used. The major shortcoming
 789  * of the full back refs is its overhead. Every time a tree block gets
 790  * COWed, we have to update back refs entry for all pointers in it.
 791  *
 792  * For a newly allocated tree block, we use implicit back refs for
 793  * pointers in it. This means most tree related operations only involve
 794  * implicit back refs. For a tree block created in old transaction, the
 795  * only way to drop a reference to it is COW it. So we can detect the
 796  * event that tree block loses its owner tree's reference and do the
 797  * back refs conversion.
 798  *
 799  * When a tree block is COW'd through a tree, there are four cases:
 800  *
 801  * The reference count of the block is one and the tree is the block's
 802  * owner tree. Nothing to do in this case.
 803  *
 804  * The reference count of the block is one and the tree is not the
 805  * block's owner tree. In this case, full back refs is used for pointers
 806  * in the block. Remove these full back refs, add implicit back refs for
 807  * every pointers in the new block.
 808  *
 809  * The reference count of the block is greater than one and the tree is
 810  * the block's owner tree. In this case, implicit back refs is used for
 811  * pointers in the block. Add full back refs for every pointers in the
 812  * block, increase lower level extents' reference counts. The original
 813  * implicit back refs are entailed to the new block.
 814  *
 815  * The reference count of the block is greater than one and the tree is
 816  * not the block's owner tree. Add implicit back refs for every pointer in
 817  * the new block, increase lower level extents' reference count.
 818  *
 819  * Back Reference Key composing:
 820  *
 821  * The key objectid corresponds to the first byte in the extent,
 822  * The key type is used to differentiate between types of back refs.
 823  * There are different meanings of the key offset for different types
 824  * of back refs.
 825  *
 826  * File extents can be referenced by:
 827  *
 828  * - multiple snapshots, subvolumes, or different generations in one subvol
 829  * - different files inside a single subvolume
 830  * - different offsets inside a file (bookend extents in file.c)
 831  *
 832  * The extent ref structure for the implicit back refs has fields for:
 833  *
 834  * - Objectid of the subvolume root
 835  * - objectid of the file holding the reference
 836  * - original offset in the file
 837  * - how many bookend extents
 838  *
 839  * The key offset for the implicit back refs is hash of the first
 840  * three fields.
 841  *
 842  * The extent ref structure for the full back refs has field for:
 843  *
 844  * - number of pointers in the tree leaf
 845  *
 846  * The key offset for the implicit back refs is the first byte of
 847  * the tree leaf
 848  *
 849  * When a file extent is allocated, The implicit back refs is used.
 850  * the fields are filled in:
 851  *
 852  *     (root_key.objectid, inode objectid, offset in file, 1)
 853  *
 854  * When a file extent is removed file truncation, we find the
 855  * corresponding implicit back refs and check the following fields:
 856  *
 857  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 858  *
 859  * Btree extents can be referenced by:
 860  *
 861  * - Different subvolumes
 862  *
 863  * Both the implicit back refs and the full back refs for tree blocks
 864  * only consist of key. The key offset for the implicit back refs is
 865  * objectid of block's owner tree. The key offset for the full back refs
 866  * is the first byte of parent block.
 867  *
 868  * When implicit back refs is used, information about the lowest key and
 869  * level of the tree block are required. These information are stored in
 870  * tree block info structure.
 871  */
 872
 873 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 874 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 875                                   struct btrfs_root *root,
 876                                   struct btrfs_path *path,
 877                                   u64 owner, u32 extra_size)
 878 {
 879         struct btrfs_extent_item *item;
 880         struct btrfs_extent_item_v0 *ei0;
 881         struct btrfs_extent_ref_v0 *ref0;
 882         struct btrfs_tree_block_info *bi;
 883         struct extent_buffer *leaf;
 884         struct btrfs_key key;
 885         struct btrfs_key found_key;
 886         u32 new_size = sizeof(*item);
 887         u64 refs;
 888         int ret;
 889
 890         leaf = path->nodes[0];
 891         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 892
 893         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 894         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 895                              struct btrfs_extent_item_v0);
 896         refs = btrfs_extent_refs_v0(leaf, ei0);
 897
 898         if (owner == (u64)-1) {
 899                 while (1) {
 900                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 901                                 ret = btrfs_next_leaf(root, path);
 902                                 if (ret < 0)
 903                                         return ret;
 904                                 BUG_ON(ret > 0);
 905                                 leaf = path->nodes[0];
 906                         }
 907                         btrfs_item_key_to_cpu(leaf, &found_key,
 908                                               path->slots[0]);
 909                         BUG_ON(key.objectid != found_key.objectid);
 910                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 911                                 path->slots[0]++;
 912                                 continue;
 913                         }
 914                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 915                                               struct btrfs_extent_ref_v0);
 916                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 917                         break;
 918                 }
 919         }
 920         btrfs_release_path(root, path);
 921
 922         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 923                 new_size += sizeof(*bi);
 924
 925         new_size -= sizeof(*ei0);
 926         ret = btrfs_search_slot(trans, root, &key, path,
 927                                 new_size + extra_size, 1);
 928         if (ret < 0)
 929                 return ret;
 930         BUG_ON(ret);
 931
 932         ret = btrfs_extend_item(trans, root, path, new_size);
 933         BUG_ON(ret);
 934
 935         leaf = path->nodes[0];
 936         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 937         btrfs_set_extent_refs(leaf, item, refs);
 938         /* FIXME: get real generation */
 939         btrfs_set_extent_generation(leaf, item, 0);
 940         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 941                 btrfs_set_extent_flags(leaf, item,
 942                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 943                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 944                 bi = (struct btrfs_tree_block_info *)(item + 1);
 945                 /* FIXME: get first key of the block */
 946                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 947                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 948         } else {
 949                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 950         }
 951         btrfs_mark_buffer_dirty(leaf);
 952         return 0;
 953 }
 954 #endif
 955
 956 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 957 {
 958         u32 high_crc = ~(u32)0;
 959         u32 low_crc = ~(u32)0;
 960         __le64 lenum;
 961
 962         lenum = cpu_to_le64(root_objectid);
 963         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 964         lenum = cpu_to_le64(owner);
 965         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 966         lenum = cpu_to_le64(offset);
 967         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 968
 969         return ((u64)high_crc << 31) ^ (u64)low_crc;
 970 }
 971
 972 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 973                                      struct btrfs_extent_data_ref *ref)
 974 {
 975         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 976                                     btrfs_extent_data_ref_objectid(leaf, ref),
 977                                     btrfs_extent_data_ref_offset(leaf, ref));
 978 }
 979
 980 static int match_extent_data_ref(struct extent_buffer *leaf,
 981                                  struct btrfs_extent_data_ref *ref,
 982                                  u64 root_objectid, u64 owner, u64 offset)
 983 {
 984         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 985             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 986             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 987                 return 0;
 988         return 1;
 989 }
 990
 991 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 992                                            struct btrfs_root *root,
 993                                            struct btrfs_path *path,
 994                                            u64 bytenr, u64 parent,
 995                                            u64 root_objectid,
 996                                            u64 owner, u64 offset)
 997 {
 998         struct btrfs_key key;
 999         struct btrfs_extent_data_ref *ref;
1000         struct extent_buffer *leaf;
1001         u32 nritems;
1002         int ret;
1003         int recow;
1004         int err = -ENOENT;
1005
1006         key.objectid = bytenr;
1007         if (parent) {
1008                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1009                 key.offset = parent;
1010         } else {
1011                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1012                 key.offset = hash_extent_data_ref(root_objectid,
1013                                                   owner, offset);
1014         }
1015 again:
1016         recow = 0;
1017         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1018         if (ret < 0) {
1019                 err = ret;
1020                 goto fail;
1021         }
1022
1023         if (parent) {
1024                 if (!ret)
1025                         return 0;
1026 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1027                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1028                 btrfs_release_path(root, path);
1029                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1030                 if (ret < 0) {
1031                         err = ret;
1032                         goto fail;
1033                 }
1034                 if (!ret)
1035                         return 0;
1036 #endif
1037                 goto fail;
1038         }
1039
1040         leaf = path->nodes[0];
1041         nritems = btrfs_header_nritems(leaf);
1042         while (1) {
1043                 if (path->slots[0] >= nritems) {
1044                         ret = btrfs_next_leaf(root, path);
1045                         if (ret < 0)
1046                                 err = ret;
1047                         if (ret)
1048                                 goto fail;
1049
1050                         leaf = path->nodes[0];
1051                         nritems = btrfs_header_nritems(leaf);
1052                         recow = 1;
1053                 }
1054
1055                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1056                 if (key.objectid != bytenr ||
1057                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1058                         goto fail;
1059
1060                 ref = btrfs_item_ptr(leaf, path->slots[0],
1061                                      struct btrfs_extent_data_ref);
1062
1063                 if (match_extent_data_ref(leaf, ref, root_objectid,
1064                                           owner, offset)) {
1065                         if (recow) {
1066                                 btrfs_release_path(root, path);
1067                                 goto again;
1068                         }
1069                         err = 0;
1070                         break;
1071                 }
1072                 path->slots[0]++;
1073         }
1074 fail:
1075         return err;
1076 }
1077
1078 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1079                                            struct btrfs_root *root,
1080                                            struct btrfs_path *path,
1081                                            u64 bytenr, u64 parent,
1082                                            u64 root_objectid, u64 owner,
1083                                            u64 offset, int refs_to_add)
1084 {
1085         struct btrfs_key key;
1086         struct extent_buffer *leaf;
1087         u32 size;
1088         u32 num_refs;
1089         int ret;
1090
1091         key.objectid = bytenr;
1092         if (parent) {
1093                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1094                 key.offset = parent;
1095                 size = sizeof(struct btrfs_shared_data_ref);
1096         } else {
1097                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1098                 key.offset = hash_extent_data_ref(root_objectid,
1099                                                   owner, offset);
1100                 size = sizeof(struct btrfs_extent_data_ref);
1101         }
1102
1103         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1104         if (ret && ret != -EEXIST)
1105                 goto fail;
1106
1107         leaf = path->nodes[0];
1108         if (parent) {
1109                 struct btrfs_shared_data_ref *ref;
1110                 ref = btrfs_item_ptr(leaf, path->slots[0],
1111                                      struct btrfs_shared_data_ref);
1112                 if (ret == 0) {
1113                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1114                 } else {
1115                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1116                         num_refs += refs_to_add;
1117                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1118                 }
1119         } else {
1120                 struct btrfs_extent_data_ref *ref;
1121                 while (ret == -EEXIST) {
1122                         ref = btrfs_item_ptr(leaf, path->slots[0],
1123                                              struct btrfs_extent_data_ref);
1124                         if (match_extent_data_ref(leaf, ref, root_objectid,
1125                                                   owner, offset))
1126                                 break;
1127                         btrfs_release_path(root, path);
1128                         key.offset++;
1129                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1130                                                       size);
1131                         if (ret && ret != -EEXIST)
1132                                 goto fail;
1133
1134                         leaf = path->nodes[0];
1135                 }
1136                 ref = btrfs_item_ptr(leaf, path->slots[0],
1137                                      struct btrfs_extent_data_ref);
1138                 if (ret == 0) {
1139                         btrfs_set_extent_data_ref_root(leaf, ref,
1140                                                        root_objectid);
1141                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1142                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1143                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1144                 } else {
1145                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1146                         num_refs += refs_to_add;
1147                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1148                 }
1149         }
1150         btrfs_mark_buffer_dirty(leaf);
1151         ret = 0;
1152 fail:
1153         btrfs_release_path(root, path);
1154         return ret;
1155 }
1156
1157 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1158                                            struct btrfs_root *root,
1159                                            struct btrfs_path *path,
1160                                            int refs_to_drop)
1161 {
1162         struct btrfs_key key;
1163         struct btrfs_extent_data_ref *ref1 = NULL;
1164         struct btrfs_shared_data_ref *ref2 = NULL;
1165         struct extent_buffer *leaf;
1166         u32 num_refs = 0;
1167         int ret = 0;
1168
1169         leaf = path->nodes[0];
1170         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1171
1172         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1173                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1174                                       struct btrfs_extent_data_ref);
1175                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1176         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1177                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1178                                       struct btrfs_shared_data_ref);
1179                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1180 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1181         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1182                 struct btrfs_extent_ref_v0 *ref0;
1183                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1184                                       struct btrfs_extent_ref_v0);
1185                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1186 #endif
1187         } else {
1188                 BUG();
1189         }
1190
1191         BUG_ON(num_refs < refs_to_drop);
1192         num_refs -= refs_to_drop;
1193
1194         if (num_refs == 0) {
1195                 ret = btrfs_del_item(trans, root, path);
1196         } else {
1197                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1198                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1199                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1200                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1201 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1202                 else {
1203                         struct btrfs_extent_ref_v0 *ref0;
1204                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1205                                         struct btrfs_extent_ref_v0);
1206                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1207                 }
1208 #endif
1209                 btrfs_mark_buffer_dirty(leaf);
1210         }
1211         return ret;
1212 }
1213
1214 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1215                                           struct btrfs_path *path,
1216                                           struct btrfs_extent_inline_ref *iref)
1217 {
1218         struct btrfs_key key;
1219         struct extent_buffer *leaf;
1220         struct btrfs_extent_data_ref *ref1;
1221         struct btrfs_shared_data_ref *ref2;
1222         u32 num_refs = 0;
1223
1224         leaf = path->nodes[0];
1225         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1226         if (iref) {
1227                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1228                     BTRFS_EXTENT_DATA_REF_KEY) {
1229                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1230                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1231                 } else {
1232                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1233                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1234                 }
1235         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1236                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1237                                       struct btrfs_extent_data_ref);
1238                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1239         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1240                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1241                                       struct btrfs_shared_data_ref);
1242                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1243 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1244         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1245                 struct btrfs_extent_ref_v0 *ref0;
1246                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1247                                       struct btrfs_extent_ref_v0);
1248                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1249 #endif
1250         } else {
1251                 WARN_ON(1);
1252         }
1253         return num_refs;
1254 }
1255
1256 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1257                                           struct btrfs_root *root,
1258                                           struct btrfs_path *path,
1259                                           u64 bytenr, u64 parent,
1260                                           u64 root_objectid)
1261 {
1262         struct btrfs_key key;
1263         int ret;
1264
1265         key.objectid = bytenr;
1266         if (parent) {
1267                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1268                 key.offset = parent;
1269         } else {
1270                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1271                 key.offset = root_objectid;
1272         }
1273
1274         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1275         if (ret > 0)
1276                 ret = -ENOENT;
1277 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1278         if (ret == -ENOENT && parent) {
1279                 btrfs_release_path(root, path);
1280                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1281                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1282                 if (ret > 0)
1283                         ret = -ENOENT;
1284         }
1285 #endif
1286         return ret;
1287 }
1288
1289 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1290                                           struct btrfs_root *root,
1291                                           struct btrfs_path *path,
1292                                           u64 bytenr, u64 parent,
1293                                           u64 root_objectid)
1294 {
1295         struct btrfs_key key;
1296         int ret;
1297
1298         key.objectid = bytenr;
1299         if (parent) {
1300                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1301                 key.offset = parent;
1302         } else {
1303                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1304                 key.offset = root_objectid;
1305         }
1306
1307         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1308         btrfs_release_path(root, path);
1309         return ret;
1310 }
1311
1312 static inline int extent_ref_type(u64 parent, u64 owner)
1313 {
1314         int type;
1315         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1316                 if (parent > 0)
1317                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1318                 else
1319                         type = BTRFS_TREE_BLOCK_REF_KEY;
1320         } else {
1321                 if (parent > 0)
1322                         type = BTRFS_SHARED_DATA_REF_KEY;
1323                 else
1324                         type = BTRFS_EXTENT_DATA_REF_KEY;
1325         }
1326         return type;
1327 }
1328
1329 static int find_next_key(struct btrfs_path *path, int level,
1330                          struct btrfs_key *key)
1331
1332 {
1333         for (; level < BTRFS_MAX_LEVEL; level++) {
1334                 if (!path->nodes[level])
1335                         break;
1336                 if (path->slots[level] + 1 >=
1337                     btrfs_header_nritems(path->nodes[level]))
1338                         continue;
1339                 if (level == 0)
1340                         btrfs_item_key_to_cpu(path->nodes[level], key,
1341                                               path->slots[level] + 1);
1342                 else
1343                         btrfs_node_key_to_cpu(path->nodes[level], key,
1344                                               path->slots[level] + 1);
1345                 return 0;
1346         }
1347         return 1;
1348 }
1349
1350 /*
1351  * look for inline back ref. if back ref is found, *ref_ret is set
1352  * to the address of inline back ref, and 0 is returned.
1353  *
1354  * if back ref isn't found, *ref_ret is set to the address where it
1355  * should be inserted, and -ENOENT is returned.
1356  *
1357  * if insert is true and there are too many inline back refs, the path
1358  * points to the extent item, and -EAGAIN is returned.
1359  *
1360  * NOTE: inline back refs are ordered in the same way that back ref
1361  *       items in the tree are ordered.
1362  */
1363 static noinline_for_stack
1364 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1365                                  struct btrfs_root *root,
1366                                  struct btrfs_path *path,
1367                                  struct btrfs_extent_inline_ref **ref_ret,
1368                                  u64 bytenr, u64 num_bytes,
1369                                  u64 parent, u64 root_objectid,
1370                                  u64 owner, u64 offset, int insert)
1371 {
1372         struct btrfs_key key;
1373         struct extent_buffer *leaf;
1374         struct btrfs_extent_item *ei;
1375         struct btrfs_extent_inline_ref *iref;
1376         u64 flags;
1377         u64 item_size;
1378         unsigned long ptr;
1379         unsigned long end;
1380         int extra_size;
1381         int type;
1382         int want;
1383         int ret;
1384         int err = 0;
1385
1386         key.objectid = bytenr;
1387         key.type = BTRFS_EXTENT_ITEM_KEY;
1388         key.offset = num_bytes;
1389
1390         want = extent_ref_type(parent, owner);
1391         if (insert) {
1392                 extra_size = btrfs_extent_inline_ref_size(want);
1393                 path->keep_locks = 1;
1394         } else
1395                 extra_size = -1;
1396         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1397         if (ret < 0) {
1398                 err = ret;
1399                 goto out;
1400         }
1401         BUG_ON(ret);
1402
1403         leaf = path->nodes[0];
1404         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1405 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1406         if (item_size < sizeof(*ei)) {
1407                 if (!insert) {
1408                         err = -ENOENT;
1409                         goto out;
1410                 }
1411                 ret = convert_extent_item_v0(trans, root, path, owner,
1412                                              extra_size);
1413                 if (ret < 0) {
1414                         err = ret;
1415                         goto out;
1416                 }
1417                 leaf = path->nodes[0];
1418                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1419         }
1420 #endif
1421         BUG_ON(item_size < sizeof(*ei));
1422
1423         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1424         flags = btrfs_extent_flags(leaf, ei);
1425
1426         ptr = (unsigned long)(ei + 1);
1427         end = (unsigned long)ei + item_size;
1428
1429         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1430                 ptr += sizeof(struct btrfs_tree_block_info);
1431                 BUG_ON(ptr > end);
1432         } else {
1433                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1434         }
1435
1436         err = -ENOENT;
1437         while (1) {
1438                 if (ptr >= end) {
1439                         WARN_ON(ptr > end);
1440                         break;
1441                 }
1442                 iref = (struct btrfs_extent_inline_ref *)ptr;
1443                 type = btrfs_extent_inline_ref_type(leaf, iref);
1444                 if (want < type)
1445                         break;
1446                 if (want > type) {
1447                         ptr += btrfs_extent_inline_ref_size(type);
1448                         continue;
1449                 }
1450
1451                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1452                         struct btrfs_extent_data_ref *dref;
1453                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1454                         if (match_extent_data_ref(leaf, dref, root_objectid,
1455                                                   owner, offset)) {
1456                                 err = 0;
1457                                 break;
1458                         }
1459                         if (hash_extent_data_ref_item(leaf, dref) <
1460                             hash_extent_data_ref(root_objectid, owner, offset))
1461                                 break;
1462                 } else {
1463                         u64 ref_offset;
1464                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1465                         if (parent > 0) {
1466                                 if (parent == ref_offset) {
1467                                         err = 0;
1468                                         break;
1469                                 }
1470                                 if (ref_offset < parent)
1471                                         break;
1472                         } else {
1473                                 if (root_objectid == ref_offset) {
1474                                         err = 0;
1475                                         break;
1476                                 }
1477                                 if (ref_offset < root_objectid)
1478                                         break;
1479                         }
1480                 }
1481                 ptr += btrfs_extent_inline_ref_size(type);
1482         }
1483         if (err == -ENOENT && insert) {
1484                 if (item_size + extra_size >=
1485                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1486                         err = -EAGAIN;
1487                         goto out;
1488                 }
1489                 /*
1490                  * To add new inline back ref, we have to make sure
1491                  * there is no corresponding back ref item.
1492                  * For simplicity, we just do not add new inline back
1493                  * ref if there is any kind of item for this block
1494                  */
1495                 if (find_next_key(path, 0, &key) == 0 &&
1496                     key.objectid == bytenr &&
1497                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1498                         err = -EAGAIN;
1499                         goto out;
1500                 }
1501         }
1502         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1503 out:
1504         if (insert) {
1505                 path->keep_locks = 0;
1506                 btrfs_unlock_up_safe(path, 1);
1507         }
1508         return err;
1509 }
1510
1511 /*
1512  * helper to add new inline back ref
1513  */
1514 static noinline_for_stack
1515 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1516                                 struct btrfs_root *root,
1517                                 struct btrfs_path *path,
1518                                 struct btrfs_extent_inline_ref *iref,
1519                                 u64 parent, u64 root_objectid,
1520                                 u64 owner, u64 offset, int refs_to_add,
1521                                 struct btrfs_delayed_extent_op *extent_op)
1522 {
1523         struct extent_buffer *leaf;
1524         struct btrfs_extent_item *ei;
1525         unsigned long ptr;
1526         unsigned long end;
1527         unsigned long item_offset;
1528         u64 refs;
1529         int size;
1530         int type;
1531         int ret;
1532
1533         leaf = path->nodes[0];
1534         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1535         item_offset = (unsigned long)iref - (unsigned long)ei;
1536
1537         type = extent_ref_type(parent, owner);
1538         size = btrfs_extent_inline_ref_size(type);
1539
1540         ret = btrfs_extend_item(trans, root, path, size);
1541         BUG_ON(ret);
1542
1543         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1544         refs = btrfs_extent_refs(leaf, ei);
1545         refs += refs_to_add;
1546         btrfs_set_extent_refs(leaf, ei, refs);
1547         if (extent_op)
1548                 __run_delayed_extent_op(extent_op, leaf, ei);
1549
1550         ptr = (unsigned long)ei + item_offset;
1551         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1552         if (ptr < end - size)
1553                 memmove_extent_buffer(leaf, ptr + size, ptr,
1554                                       end - size - ptr);
1555
1556         iref = (struct btrfs_extent_inline_ref *)ptr;
1557         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1558         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1559                 struct btrfs_extent_data_ref *dref;
1560                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1561                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1562                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1563                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1564                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1565         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1566                 struct btrfs_shared_data_ref *sref;
1567                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1568                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1569                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1570         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1571                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1572         } else {
1573                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1574         }
1575         btrfs_mark_buffer_dirty(leaf);
1576         return 0;
1577 }
1578
1579 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1580                                  struct btrfs_root *root,
1581                                  struct btrfs_path *path,
1582                                  struct btrfs_extent_inline_ref **ref_ret,
1583                                  u64 bytenr, u64 num_bytes, u64 parent,
1584                                  u64 root_objectid, u64 owner, u64 offset)
1585 {
1586         int ret;
1587
1588         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1589                                            bytenr, num_bytes, parent,
1590                                            root_objectid, owner, offset, 0);
1591         if (ret != -ENOENT)
1592                 return ret;
1593
1594         btrfs_release_path(root, path);
1595         *ref_ret = NULL;
1596
1597         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1598                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1599                                             root_objectid);
1600         } else {
1601                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1602                                              root_objectid, owner, offset);
1603         }
1604         return ret;
1605 }
1606
1607 /*
1608  * helper to update/remove inline back ref
1609  */
1610 static noinline_for_stack
1611 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1612                                  struct btrfs_root *root,
1613                                  struct btrfs_path *path,
1614                                  struct btrfs_extent_inline_ref *iref,
1615                                  int refs_to_mod,
1616                                  struct btrfs_delayed_extent_op *extent_op)
1617 {
1618         struct extent_buffer *leaf;
1619         struct btrfs_extent_item *ei;
1620         struct btrfs_extent_data_ref *dref = NULL;
1621         struct btrfs_shared_data_ref *sref = NULL;
1622         unsigned long ptr;
1623         unsigned long end;
1624         u32 item_size;
1625         int size;
1626         int type;
1627         int ret;
1628         u64 refs;
1629
1630         leaf = path->nodes[0];
1631         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1632         refs = btrfs_extent_refs(leaf, ei);
1633         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1634         refs += refs_to_mod;
1635         btrfs_set_extent_refs(leaf, ei, refs);
1636         if (extent_op)
1637                 __run_delayed_extent_op(extent_op, leaf, ei);
1638
1639         type = btrfs_extent_inline_ref_type(leaf, iref);
1640
1641         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1642                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1643                 refs = btrfs_extent_data_ref_count(leaf, dref);
1644         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1645                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1646                 refs = btrfs_shared_data_ref_count(leaf, sref);
1647         } else {
1648                 refs = 1;
1649                 BUG_ON(refs_to_mod != -1);
1650         }
1651
1652         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1653         refs += refs_to_mod;
1654
1655         if (refs > 0) {
1656                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1657                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1658                 else
1659                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1660         } else {
1661                 size =  btrfs_extent_inline_ref_size(type);
1662                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1663                 ptr = (unsigned long)iref;
1664                 end = (unsigned long)ei + item_size;
1665                 if (ptr + size < end)
1666                         memmove_extent_buffer(leaf, ptr, ptr + size,
1667                                               end - ptr - size);
1668                 item_size -= size;
1669                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1670                 BUG_ON(ret);
1671         }
1672         btrfs_mark_buffer_dirty(leaf);
1673         return 0;
1674 }
1675
1676 static noinline_for_stack
1677 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1678                                  struct btrfs_root *root,
1679                                  struct btrfs_path *path,
1680                                  u64 bytenr, u64 num_bytes, u64 parent,
1681                                  u64 root_objectid, u64 owner,
1682                                  u64 offset, int refs_to_add,
1683                                  struct btrfs_delayed_extent_op *extent_op)
1684 {
1685         struct btrfs_extent_inline_ref *iref;
1686         int ret;
1687
1688         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1689                                            bytenr, num_bytes, parent,
1690                                            root_objectid, owner, offset, 1);
1691         if (ret == 0) {
1692                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1693                 ret = update_inline_extent_backref(trans, root, path, iref,
1694                                                    refs_to_add, extent_op);
1695         } else if (ret == -ENOENT) {
1696                 ret = setup_inline_extent_backref(trans, root, path, iref,
1697                                                   parent, root_objectid,
1698                                                   owner, offset, refs_to_add,
1699                                                   extent_op);
1700         }
1701         return ret;
1702 }
1703
1704 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1705                                  struct btrfs_root *root,
1706                                  struct btrfs_path *path,
1707                                  u64 bytenr, u64 parent, u64 root_objectid,
1708                                  u64 owner, u64 offset, int refs_to_add)
1709 {
1710         int ret;
1711         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1712                 BUG_ON(refs_to_add != 1);
1713                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1714                                             parent, root_objectid);
1715         } else {
1716                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1717                                              parent, root_objectid,
1718                                              owner, offset, refs_to_add);
1719         }
1720         return ret;
1721 }
1722
1723 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1724                                  struct btrfs_root *root,
1725                                  struct btrfs_path *path,
1726                                  struct btrfs_extent_inline_ref *iref,
1727                                  int refs_to_drop, int is_data)
1728 {
1729         int ret;
1730
1731         BUG_ON(!is_data && refs_to_drop != 1);
1732         if (iref) {
1733                 ret = update_inline_extent_backref(trans, root, path, iref,
1734                                                    -refs_to_drop, NULL);
1735         } else if (is_data) {
1736                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1737         } else {
1738                 ret = btrfs_del_item(trans, root, path);
1739         }
1740         return ret;
1741 }
1742
1743 static void btrfs_issue_discard(struct block_device *bdev,
1744                                 u64 start, u64 len)
1745 {
1746         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL, 0);
1747 }
1748
1749 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1750                                 u64 num_bytes)
1751 {
1752         int ret;
1753         u64 map_length = num_bytes;
1754         struct btrfs_multi_bio *multi = NULL;
1755
1756         if (!btrfs_test_opt(root, DISCARD))
1757                 return 0;
1758
1759         /* Tell the block device(s) that the sectors can be discarded */
1760         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1761                               bytenr, &map_length, &multi, 0);
1762         if (!ret) {
1763                 struct btrfs_bio_stripe *stripe = multi->stripes;
1764                 int i;
1765
1766                 if (map_length > num_bytes)
1767                         map_length = num_bytes;
1768
1769                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1770                         btrfs_issue_discard(stripe->dev->bdev,
1771                                             stripe->physical,
1772                                             map_length);
1773                 }
1774                 kfree(multi);
1775         }
1776
1777         return ret;
1778 }
1779
1780 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1781                          struct btrfs_root *root,
1782                          u64 bytenr, u64 num_bytes, u64 parent,
1783                          u64 root_objectid, u64 owner, u64 offset)
1784 {
1785         int ret;
1786         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1787                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1788
1789         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1790                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1791                                         parent, root_objectid, (int)owner,
1792                                         BTRFS_ADD_DELAYED_REF, NULL);
1793         } else {
1794                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1795                                         parent, root_objectid, owner, offset,
1796                                         BTRFS_ADD_DELAYED_REF, NULL);
1797         }
1798         return ret;
1799 }
1800
1801 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1802                                   struct btrfs_root *root,
1803                                   u64 bytenr, u64 num_bytes,
1804                                   u64 parent, u64 root_objectid,
1805                                   u64 owner, u64 offset, int refs_to_add,
1806                                   struct btrfs_delayed_extent_op *extent_op)
1807 {
1808         struct btrfs_path *path;
1809         struct extent_buffer *leaf;
1810         struct btrfs_extent_item *item;
1811         u64 refs;
1812         int ret;
1813         int err = 0;
1814
1815         path = btrfs_alloc_path();
1816         if (!path)
1817                 return -ENOMEM;
1818
1819         path->reada = 1;
1820         path->leave_spinning = 1;
1821         /* this will setup the path even if it fails to insert the back ref */
1822         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1823                                            path, bytenr, num_bytes, parent,
1824                                            root_objectid, owner, offset,
1825                                            refs_to_add, extent_op);
1826         if (ret == 0)
1827                 goto out;
1828
1829         if (ret != -EAGAIN) {
1830                 err = ret;
1831                 goto out;
1832         }
1833
1834         leaf = path->nodes[0];
1835         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1836         refs = btrfs_extent_refs(leaf, item);
1837         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1838         if (extent_op)
1839                 __run_delayed_extent_op(extent_op, leaf, item);
1840
1841         btrfs_mark_buffer_dirty(leaf);
1842         btrfs_release_path(root->fs_info->extent_root, path);
1843
1844         path->reada = 1;
1845         path->leave_spinning = 1;
1846
1847         /* now insert the actual backref */
1848         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1849                                     path, bytenr, parent, root_objectid,
1850                                     owner, offset, refs_to_add);
1851         BUG_ON(ret);
1852 out:
1853         btrfs_free_path(path);
1854         return err;
1855 }
1856
1857 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1858                                 struct btrfs_root *root,
1859                                 struct btrfs_delayed_ref_node *node,
1860                                 struct btrfs_delayed_extent_op *extent_op,
1861                                 int insert_reserved)
1862 {
1863         int ret = 0;
1864         struct btrfs_delayed_data_ref *ref;
1865         struct btrfs_key ins;
1866         u64 parent = 0;
1867         u64 ref_root = 0;
1868         u64 flags = 0;
1869
1870         ins.objectid = node->bytenr;
1871         ins.offset = node->num_bytes;
1872         ins.type = BTRFS_EXTENT_ITEM_KEY;
1873
1874         ref = btrfs_delayed_node_to_data_ref(node);
1875         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1876                 parent = ref->parent;
1877         else
1878                 ref_root = ref->root;
1879
1880         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1881                 if (extent_op) {
1882                         BUG_ON(extent_op->update_key);
1883                         flags |= extent_op->flags_to_set;
1884                 }
1885                 ret = alloc_reserved_file_extent(trans, root,
1886                                                  parent, ref_root, flags,
1887                                                  ref->objectid, ref->offset,
1888                                                  &ins, node->ref_mod);
1889         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1890                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1891                                              node->num_bytes, parent,
1892                                              ref_root, ref->objectid,
1893                                              ref->offset, node->ref_mod,
1894                                              extent_op);
1895         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1896                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1897                                           node->num_bytes, parent,
1898                                           ref_root, ref->objectid,
1899                                           ref->offset, node->ref_mod,
1900                                           extent_op);
1901         } else {
1902                 BUG();
1903         }
1904         return ret;
1905 }
1906
1907 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1908                                     struct extent_buffer *leaf,
1909                                     struct btrfs_extent_item *ei)
1910 {
1911         u64 flags = btrfs_extent_flags(leaf, ei);
1912         if (extent_op->update_flags) {
1913                 flags |= extent_op->flags_to_set;
1914                 btrfs_set_extent_flags(leaf, ei, flags);
1915         }
1916
1917         if (extent_op->update_key) {
1918                 struct btrfs_tree_block_info *bi;
1919                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1920                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1921                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1922         }
1923 }
1924
1925 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1926                                  struct btrfs_root *root,
1927                                  struct btrfs_delayed_ref_node *node,
1928                                  struct btrfs_delayed_extent_op *extent_op)
1929 {
1930         struct btrfs_key key;
1931         struct btrfs_path *path;
1932         struct btrfs_extent_item *ei;
1933         struct extent_buffer *leaf;
1934         u32 item_size;
1935         int ret;
1936         int err = 0;
1937
1938         path = btrfs_alloc_path();
1939         if (!path)
1940                 return -ENOMEM;
1941
1942         key.objectid = node->bytenr;
1943         key.type = BTRFS_EXTENT_ITEM_KEY;
1944         key.offset = node->num_bytes;
1945
1946         path->reada = 1;
1947         path->leave_spinning = 1;
1948         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1949                                 path, 0, 1);
1950         if (ret < 0) {
1951                 err = ret;
1952                 goto out;
1953         }
1954         if (ret > 0) {
1955                 err = -EIO;
1956                 goto out;
1957         }
1958
1959         leaf = path->nodes[0];
1960         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1961 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1962         if (item_size < sizeof(*ei)) {
1963                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1964                                              path, (u64)-1, 0);
1965                 if (ret < 0) {
1966                         err = ret;
1967                         goto out;
1968                 }
1969                 leaf = path->nodes[0];
1970                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1971         }
1972 #endif
1973         BUG_ON(item_size < sizeof(*ei));
1974         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1975         __run_delayed_extent_op(extent_op, leaf, ei);
1976
1977         btrfs_mark_buffer_dirty(leaf);
1978 out:
1979         btrfs_free_path(path);
1980         return err;
1981 }
1982
1983 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1984                                 struct btrfs_root *root,
1985                                 struct btrfs_delayed_ref_node *node,
1986                                 struct btrfs_delayed_extent_op *extent_op,
1987                                 int insert_reserved)
1988 {
1989         int ret = 0;
1990         struct btrfs_delayed_tree_ref *ref;
1991         struct btrfs_key ins;
1992         u64 parent = 0;
1993         u64 ref_root = 0;
1994
1995         ins.objectid = node->bytenr;
1996         ins.offset = node->num_bytes;
1997         ins.type = BTRFS_EXTENT_ITEM_KEY;
1998
1999         ref = btrfs_delayed_node_to_tree_ref(node);
2000         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2001                 parent = ref->parent;
2002         else
2003                 ref_root = ref->root;
2004
2005         BUG_ON(node->ref_mod != 1);
2006         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2007                 BUG_ON(!extent_op || !extent_op->update_flags ||
2008                        !extent_op->update_key);
2009                 ret = alloc_reserved_tree_block(trans, root,
2010                                                 parent, ref_root,
2011                                                 extent_op->flags_to_set,
2012                                                 &extent_op->key,
2013                                                 ref->level, &ins);
2014         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2015                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2016                                              node->num_bytes, parent, ref_root,
2017                                              ref->level, 0, 1, extent_op);
2018         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2019                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2020                                           node->num_bytes, parent, ref_root,
2021                                           ref->level, 0, 1, extent_op);
2022         } else {
2023                 BUG();
2024         }
2025         return ret;
2026 }
2027
2028 /* helper function to actually process a single delayed ref entry */
2029 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2030                                struct btrfs_root *root,
2031                                struct btrfs_delayed_ref_node *node,
2032                                struct btrfs_delayed_extent_op *extent_op,
2033                                int insert_reserved)
2034 {
2035         int ret;
2036         if (btrfs_delayed_ref_is_head(node)) {
2037                 struct btrfs_delayed_ref_head *head;
2038                 /*
2039                  * we've hit the end of the chain and we were supposed
2040                  * to insert this extent into the tree.  But, it got
2041                  * deleted before we ever needed to insert it, so all
2042                  * we have to do is clean up the accounting
2043                  */
2044                 BUG_ON(extent_op);
2045                 head = btrfs_delayed_node_to_head(node);
2046                 if (insert_reserved) {
2047                         btrfs_pin_extent(root, node->bytenr,
2048                                          node->num_bytes, 1);
2049                         if (head->is_data) {
2050                                 ret = btrfs_del_csums(trans, root,
2051                                                       node->bytenr,
2052                                                       node->num_bytes);
2053                                 BUG_ON(ret);
2054                         }
2055                 }
2056                 mutex_unlock(&head->mutex);
2057                 return 0;
2058         }
2059
2060         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2061             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2062                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2063                                            insert_reserved);
2064         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2065                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2066                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2067                                            insert_reserved);
2068         else
2069                 BUG();
2070         return ret;
2071 }
2072
2073 static noinline struct btrfs_delayed_ref_node *
2074 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2075 {
2076         struct rb_node *node;
2077         struct btrfs_delayed_ref_node *ref;
2078         int action = BTRFS_ADD_DELAYED_REF;
2079 again:
2080         /*
2081          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2082          * this prevents ref count from going down to zero when
2083          * there still are pending delayed ref.
2084          */
2085         node = rb_prev(&head->node.rb_node);
2086         while (1) {
2087                 if (!node)
2088                         break;
2089                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2090                                 rb_node);
2091                 if (ref->bytenr != head->node.bytenr)
2092                         break;
2093                 if (ref->action == action)
2094                         return ref;
2095                 node = rb_prev(node);
2096         }
2097         if (action == BTRFS_ADD_DELAYED_REF) {
2098                 action = BTRFS_DROP_DELAYED_REF;
2099                 goto again;
2100         }
2101         return NULL;
2102 }
2103
2104 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2105                                        struct btrfs_root *root,
2106                                        struct list_head *cluster)
2107 {
2108         struct btrfs_delayed_ref_root *delayed_refs;
2109         struct btrfs_delayed_ref_node *ref;
2110         struct btrfs_delayed_ref_head *locked_ref = NULL;
2111         struct btrfs_delayed_extent_op *extent_op;
2112         int ret;
2113         int count = 0;
2114         int must_insert_reserved = 0;
2115
2116         delayed_refs = &trans->transaction->delayed_refs;
2117         while (1) {
2118                 if (!locked_ref) {
2119                         /* pick a new head ref from the cluster list */
2120                         if (list_empty(cluster))
2121                                 break;
2122
2123                         locked_ref = list_entry(cluster->next,
2124                                      struct btrfs_delayed_ref_head, cluster);
2125
2126                         /* grab the lock that says we are going to process
2127                          * all the refs for this head */
2128                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2129
2130                         /*
2131                          * we may have dropped the spin lock to get the head
2132                          * mutex lock, and that might have given someone else
2133                          * time to free the head.  If that's true, it has been
2134                          * removed from our list and we can move on.
2135                          */
2136                         if (ret == -EAGAIN) {
2137                                 locked_ref = NULL;
2138                                 count++;
2139                                 continue;
2140                         }
2141                 }
2142
2143                 /*
2144                  * record the must insert reserved flag before we
2145                  * drop the spin lock.
2146                  */
2147                 must_insert_reserved = locked_ref->must_insert_reserved;
2148                 locked_ref->must_insert_reserved = 0;
2149
2150                 extent_op = locked_ref->extent_op;
2151                 locked_ref->extent_op = NULL;
2152
2153                 /*
2154                  * locked_ref is the head node, so we have to go one
2155                  * node back for any delayed ref updates
2156                  */
2157                 ref = select_delayed_ref(locked_ref);
2158                 if (!ref) {
2159                         /* All delayed refs have been processed, Go ahead
2160                          * and send the head node to run_one_delayed_ref,
2161                          * so that any accounting fixes can happen
2162                          */
2163                         ref = &locked_ref->node;
2164
2165                         if (extent_op && must_insert_reserved) {
2166                                 kfree(extent_op);
2167                                 extent_op = NULL;
2168                         }
2169
2170                         if (extent_op) {
2171                                 spin_unlock(&delayed_refs->lock);
2172
2173                                 ret = run_delayed_extent_op(trans, root,
2174                                                             ref, extent_op);
2175                                 BUG_ON(ret);
2176                                 kfree(extent_op);
2177
2178                                 cond_resched();
2179                                 spin_lock(&delayed_refs->lock);
2180                                 continue;
2181                         }
2182
2183                         list_del_init(&locked_ref->cluster);
2184                         locked_ref = NULL;
2185                 }
2186
2187                 ref->in_tree = 0;
2188                 rb_erase(&ref->rb_node, &delayed_refs->root);
2189                 delayed_refs->num_entries--;
2190
2191                 spin_unlock(&delayed_refs->lock);
2192
2193                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2194                                           must_insert_reserved);
2195                 BUG_ON(ret);
2196
2197                 btrfs_put_delayed_ref(ref);
2198                 kfree(extent_op);
2199                 count++;
2200
2201                 cond_resched();
2202                 spin_lock(&delayed_refs->lock);
2203         }
2204         return count;
2205 }
2206
2207 /*
2208  * this starts processing the delayed reference count updates and
2209  * extent insertions we have queued up so far.  count can be
2210  * 0, which means to process everything in the tree at the start
2211  * of the run (but not newly added entries), or it can be some target
2212  * number you'd like to process.
2213  */
2214 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2215                            struct btrfs_root *root, unsigned long count)
2216 {
2217         struct rb_node *node;
2218         struct btrfs_delayed_ref_root *delayed_refs;
2219         struct btrfs_delayed_ref_node *ref;
2220         struct list_head cluster;
2221         int ret;
2222         int run_all = count == (unsigned long)-1;
2223         int run_most = 0;
2224
2225         if (root == root->fs_info->extent_root)
2226                 root = root->fs_info->tree_root;
2227
2228         delayed_refs = &trans->transaction->delayed_refs;
2229         INIT_LIST_HEAD(&cluster);
2230 again:
2231         spin_lock(&delayed_refs->lock);
2232         if (count == 0) {
2233                 count = delayed_refs->num_entries * 2;
2234                 run_most = 1;
2235         }
2236         while (1) {
2237                 if (!(run_all || run_most) &&
2238                     delayed_refs->num_heads_ready < 64)
2239                         break;
2240
2241                 /*
2242                  * go find something we can process in the rbtree.  We start at
2243                  * the beginning of the tree, and then build a cluster
2244                  * of refs to process starting at the first one we are able to
2245                  * lock
2246                  */
2247                 ret = btrfs_find_ref_cluster(trans, &cluster,
2248                                              delayed_refs->run_delayed_start);
2249                 if (ret)
2250                         break;
2251
2252                 ret = run_clustered_refs(trans, root, &cluster);
2253                 BUG_ON(ret < 0);
2254
2255                 count -= min_t(unsigned long, ret, count);
2256
2257                 if (count == 0)
2258                         break;
2259         }
2260
2261         if (run_all) {
2262                 node = rb_first(&delayed_refs->root);
2263                 if (!node)
2264                         goto out;
2265                 count = (unsigned long)-1;
2266
2267                 while (node) {
2268                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2269                                        rb_node);
2270                         if (btrfs_delayed_ref_is_head(ref)) {
2271                                 struct btrfs_delayed_ref_head *head;
2272
2273                                 head = btrfs_delayed_node_to_head(ref);
2274                                 atomic_inc(&ref->refs);
2275
2276                                 spin_unlock(&delayed_refs->lock);
2277                                 mutex_lock(&head->mutex);
2278                                 mutex_unlock(&head->mutex);
2279
2280                                 btrfs_put_delayed_ref(ref);
2281                                 cond_resched();
2282                                 goto again;
2283                         }
2284                         node = rb_next(node);
2285                 }
2286                 spin_unlock(&delayed_refs->lock);
2287                 schedule_timeout(1);
2288                 goto again;
2289         }
2290 out:
2291         spin_unlock(&delayed_refs->lock);
2292         return 0;
2293 }
2294
2295 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2296                                 struct btrfs_root *root,
2297                                 u64 bytenr, u64 num_bytes, u64 flags,
2298                                 int is_data)
2299 {
2300         struct btrfs_delayed_extent_op *extent_op;
2301         int ret;
2302
2303         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2304         if (!extent_op)
2305                 return -ENOMEM;
2306
2307         extent_op->flags_to_set = flags;
2308         extent_op->update_flags = 1;
2309         extent_op->update_key = 0;
2310         extent_op->is_data = is_data ? 1 : 0;
2311
2312         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2313         if (ret)
2314                 kfree(extent_op);
2315         return ret;
2316 }
2317
2318 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2319                                       struct btrfs_root *root,
2320                                       struct btrfs_path *path,
2321                                       u64 objectid, u64 offset, u64 bytenr)
2322 {
2323         struct btrfs_delayed_ref_head *head;
2324         struct btrfs_delayed_ref_node *ref;
2325         struct btrfs_delayed_data_ref *data_ref;
2326         struct btrfs_delayed_ref_root *delayed_refs;
2327         struct rb_node *node;
2328         int ret = 0;
2329
2330         ret = -ENOENT;
2331         delayed_refs = &trans->transaction->delayed_refs;
2332         spin_lock(&delayed_refs->lock);
2333         head = btrfs_find_delayed_ref_head(trans, bytenr);
2334         if (!head)
2335                 goto out;
2336
2337         if (!mutex_trylock(&head->mutex)) {
2338                 atomic_inc(&head->node.refs);
2339                 spin_unlock(&delayed_refs->lock);
2340
2341                 btrfs_release_path(root->fs_info->extent_root, path);
2342
2343                 mutex_lock(&head->mutex);
2344                 mutex_unlock(&head->mutex);
2345                 btrfs_put_delayed_ref(&head->node);
2346                 return -EAGAIN;
2347         }
2348
2349         node = rb_prev(&head->node.rb_node);
2350         if (!node)
2351                 goto out_unlock;
2352
2353         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2354
2355         if (ref->bytenr != bytenr)
2356                 goto out_unlock;
2357
2358         ret = 1;
2359         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2360                 goto out_unlock;
2361
2362         data_ref = btrfs_delayed_node_to_data_ref(ref);
2363
2364         node = rb_prev(node);
2365         if (node) {
2366                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2367                 if (ref->bytenr == bytenr)
2368                         goto out_unlock;
2369         }
2370
2371         if (data_ref->root != root->root_key.objectid ||
2372             data_ref->objectid != objectid || data_ref->offset != offset)
2373                 goto out_unlock;
2374
2375         ret = 0;
2376 out_unlock:
2377         mutex_unlock(&head->mutex);
2378 out:
2379         spin_unlock(&delayed_refs->lock);
2380         return ret;
2381 }
2382
2383 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2384                                         struct btrfs_root *root,
2385                                         struct btrfs_path *path,
2386                                         u64 objectid, u64 offset, u64 bytenr)
2387 {
2388         struct btrfs_root *extent_root = root->fs_info->extent_root;
2389         struct extent_buffer *leaf;
2390         struct btrfs_extent_data_ref *ref;
2391         struct btrfs_extent_inline_ref *iref;
2392         struct btrfs_extent_item *ei;
2393         struct btrfs_key key;
2394         u32 item_size;
2395         int ret;
2396
2397         key.objectid = bytenr;
2398         key.offset = (u64)-1;
2399         key.type = BTRFS_EXTENT_ITEM_KEY;
2400
2401         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2402         if (ret < 0)
2403                 goto out;
2404         BUG_ON(ret == 0);
2405
2406         ret = -ENOENT;
2407         if (path->slots[0] == 0)
2408                 goto out;
2409
2410         path->slots[0]--;
2411         leaf = path->nodes[0];
2412         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2413
2414         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2415                 goto out;
2416
2417         ret = 1;
2418         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2419 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2420         if (item_size < sizeof(*ei)) {
2421                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2422                 goto out;
2423         }
2424 #endif
2425         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2426
2427         if (item_size != sizeof(*ei) +
2428             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2429                 goto out;
2430
2431         if (btrfs_extent_generation(leaf, ei) <=
2432             btrfs_root_last_snapshot(&root->root_item))
2433                 goto out;
2434
2435         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2436         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2437             BTRFS_EXTENT_DATA_REF_KEY)
2438                 goto out;
2439
2440         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2441         if (btrfs_extent_refs(leaf, ei) !=
2442             btrfs_extent_data_ref_count(leaf, ref) ||
2443             btrfs_extent_data_ref_root(leaf, ref) !=
2444             root->root_key.objectid ||
2445             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2446             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2447                 goto out;
2448
2449         ret = 0;
2450 out:
2451         return ret;
2452 }
2453
2454 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2455                           struct btrfs_root *root,
2456                           u64 objectid, u64 offset, u64 bytenr)
2457 {
2458         struct btrfs_path *path;
2459         int ret;
2460         int ret2;
2461
2462         path = btrfs_alloc_path();
2463         if (!path)
2464                 return -ENOENT;
2465
2466         do {
2467                 ret = check_committed_ref(trans, root, path, objectid,
2468                                           offset, bytenr);
2469                 if (ret && ret != -ENOENT)
2470                         goto out;
2471
2472                 ret2 = check_delayed_ref(trans, root, path, objectid,
2473                                          offset, bytenr);
2474         } while (ret2 == -EAGAIN);
2475
2476         if (ret2 && ret2 != -ENOENT) {
2477                 ret = ret2;
2478                 goto out;
2479         }
2480
2481         if (ret != -ENOENT || ret2 != -ENOENT)
2482                 ret = 0;
2483 out:
2484         btrfs_free_path(path);
2485         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2486                 WARN_ON(ret > 0);
2487         return ret;
2488 }
2489
2490 #if 0
2491 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2492                     struct extent_buffer *buf, u32 nr_extents)
2493 {
2494         struct btrfs_key key;
2495         struct btrfs_file_extent_item *fi;
2496         u64 root_gen;
2497         u32 nritems;
2498         int i;
2499         int level;
2500         int ret = 0;
2501         int shared = 0;
2502
2503         if (!root->ref_cows)
2504                 return 0;
2505
2506         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2507                 shared = 0;
2508                 root_gen = root->root_key.offset;
2509         } else {
2510                 shared = 1;
2511                 root_gen = trans->transid - 1;
2512         }
2513
2514         level = btrfs_header_level(buf);
2515         nritems = btrfs_header_nritems(buf);
2516
2517         if (level == 0) {
2518                 struct btrfs_leaf_ref *ref;
2519                 struct btrfs_extent_info *info;
2520
2521                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2522                 if (!ref) {
2523                         ret = -ENOMEM;
2524                         goto out;
2525                 }
2526
2527                 ref->root_gen = root_gen;
2528                 ref->bytenr = buf->start;
2529                 ref->owner = btrfs_header_owner(buf);
2530                 ref->generation = btrfs_header_generation(buf);
2531                 ref->nritems = nr_extents;
2532                 info = ref->extents;
2533
2534                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2535                         u64 disk_bytenr;
2536                         btrfs_item_key_to_cpu(buf, &key, i);
2537                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2538                                 continue;
2539                         fi = btrfs_item_ptr(buf, i,
2540                                             struct btrfs_file_extent_item);
2541                         if (btrfs_file_extent_type(buf, fi) ==
2542                             BTRFS_FILE_EXTENT_INLINE)
2543                                 continue;
2544                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2545                         if (disk_bytenr == 0)
2546                                 continue;
2547
2548                         info->bytenr = disk_bytenr;
2549                         info->num_bytes =
2550                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2551                         info->objectid = key.objectid;
2552                         info->offset = key.offset;
2553                         info++;
2554                 }
2555
2556                 ret = btrfs_add_leaf_ref(root, ref, shared);
2557                 if (ret == -EEXIST && shared) {
2558                         struct btrfs_leaf_ref *old;
2559                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2560                         BUG_ON(!old);
2561                         btrfs_remove_leaf_ref(root, old);
2562                         btrfs_free_leaf_ref(root, old);
2563                         ret = btrfs_add_leaf_ref(root, ref, shared);
2564                 }
2565                 WARN_ON(ret);
2566                 btrfs_free_leaf_ref(root, ref);
2567         }
2568 out:
2569         return ret;
2570 }
2571
2572 /* when a block goes through cow, we update the reference counts of
2573  * everything that block points to.  The internal pointers of the block
2574  * can be in just about any order, and it is likely to have clusters of
2575  * things that are close together and clusters of things that are not.
2576  *
2577  * To help reduce the seeks that come with updating all of these reference
2578  * counts, sort them by byte number before actual updates are done.
2579  *
2580  * struct refsort is used to match byte number to slot in the btree block.
2581  * we sort based on the byte number and then use the slot to actually
2582  * find the item.
2583  *
2584  * struct refsort is smaller than strcut btrfs_item and smaller than
2585  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2586  * for a btree block, there's no way for a kmalloc of refsorts for a
2587  * single node to be bigger than a page.
2588  */
2589 struct refsort {
2590         u64 bytenr;
2591         u32 slot;
2592 };
2593
2594 /*
2595  * for passing into sort()
2596  */
2597 static int refsort_cmp(const void *a_void, const void *b_void)
2598 {
2599         const struct refsort *a = a_void;
2600         const struct refsort *b = b_void;
2601
2602         if (a->bytenr < b->bytenr)
2603                 return -1;
2604         if (a->bytenr > b->bytenr)
2605                 return 1;
2606         return 0;
2607 }
2608 #endif
2609
2610 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2611                            struct btrfs_root *root,
2612                            struct extent_buffer *buf,
2613                            int full_backref, int inc)
2614 {
2615         u64 bytenr;
2616         u64 num_bytes;
2617         u64 parent;
2618         u64 ref_root;
2619         u32 nritems;
2620         struct btrfs_key key;
2621         struct btrfs_file_extent_item *fi;
2622         int i;
2623         int level;
2624         int ret = 0;
2625         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2626                             u64, u64, u64, u64, u64, u64);
2627
2628         ref_root = btrfs_header_owner(buf);
2629         nritems = btrfs_header_nritems(buf);
2630         level = btrfs_header_level(buf);
2631
2632         if (!root->ref_cows && level == 0)
2633                 return 0;
2634
2635         if (inc)
2636                 process_func = btrfs_inc_extent_ref;
2637         else
2638                 process_func = btrfs_free_extent;
2639
2640         if (full_backref)
2641                 parent = buf->start;
2642         else
2643                 parent = 0;
2644
2645         for (i = 0; i < nritems; i++) {
2646                 if (level == 0) {
2647                         btrfs_item_key_to_cpu(buf, &key, i);
2648                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2649                                 continue;
2650                         fi = btrfs_item_ptr(buf, i,
2651                                             struct btrfs_file_extent_item);
2652                         if (btrfs_file_extent_type(buf, fi) ==
2653                             BTRFS_FILE_EXTENT_INLINE)
2654                                 continue;
2655                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2656                         if (bytenr == 0)
2657                                 continue;
2658
2659                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2660                         key.offset -= btrfs_file_extent_offset(buf, fi);
2661                         ret = process_func(trans, root, bytenr, num_bytes,
2662                                            parent, ref_root, key.objectid,
2663                                            key.offset);
2664                         if (ret)
2665                                 goto fail;
2666                 } else {
2667                         bytenr = btrfs_node_blockptr(buf, i);
2668                         num_bytes = btrfs_level_size(root, level - 1);
2669                         ret = process_func(trans, root, bytenr, num_bytes,
2670                                            parent, ref_root, level - 1, 0);
2671                         if (ret)
2672                                 goto fail;
2673                 }
2674         }
2675         return 0;
2676 fail:
2677         BUG();
2678         return ret;
2679 }
2680
2681 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2682                   struct extent_buffer *buf, int full_backref)
2683 {
2684         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2685 }
2686
2687 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2688                   struct extent_buffer *buf, int full_backref)
2689 {
2690         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2691 }
2692
2693 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2694                                  struct btrfs_root *root,
2695                                  struct btrfs_path *path,
2696                                  struct btrfs_block_group_cache *cache)
2697 {
2698         int ret;
2699         struct btrfs_root *extent_root = root->fs_info->extent_root;
2700         unsigned long bi;
2701         struct extent_buffer *leaf;
2702
2703         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2704         if (ret < 0)
2705                 goto fail;
2706         BUG_ON(ret);
2707
2708         leaf = path->nodes[0];
2709         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2710         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2711         btrfs_mark_buffer_dirty(leaf);
2712         btrfs_release_path(extent_root, path);
2713 fail:
2714         if (ret)
2715                 return ret;
2716         return 0;
2717
2718 }
2719
2720 static struct btrfs_block_group_cache *
2721 next_block_group(struct btrfs_root *root,
2722                  struct btrfs_block_group_cache *cache)
2723 {
2724         struct rb_node *node;
2725         spin_lock(&root->fs_info->block_group_cache_lock);
2726         node = rb_next(&cache->cache_node);
2727         btrfs_put_block_group(cache);
2728         if (node) {
2729                 cache = rb_entry(node, struct btrfs_block_group_cache,
2730                                  cache_node);
2731                 btrfs_get_block_group(cache);
2732         } else
2733                 cache = NULL;
2734         spin_unlock(&root->fs_info->block_group_cache_lock);
2735         return cache;
2736 }
2737
2738 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2739                             struct btrfs_trans_handle *trans,
2740                             struct btrfs_path *path)
2741 {
2742         struct btrfs_root *root = block_group->fs_info->tree_root;
2743         struct inode *inode = NULL;
2744         u64 alloc_hint = 0;
2745         int dcs = BTRFS_DC_ERROR;
2746         int num_pages = 0;
2747         int retries = 0;
2748         int ret = 0;
2749
2750         /*
2751          * If this block group is smaller than 100 megs don't bother caching the
2752          * block group.
2753          */
2754         if (block_group->key.offset < (100 * 1024 * 1024)) {
2755                 spin_lock(&block_group->lock);
2756                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2757                 spin_unlock(&block_group->lock);
2758                 return 0;
2759         }
2760
2761 again:
2762         inode = lookup_free_space_inode(root, block_group, path);
2763         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2764                 ret = PTR_ERR(inode);
2765                 btrfs_release_path(root, path);
2766                 goto out;
2767         }
2768
2769         if (IS_ERR(inode)) {
2770                 BUG_ON(retries);
2771                 retries++;
2772
2773                 if (block_group->ro)
2774                         goto out_free;
2775
2776                 ret = create_free_space_inode(root, trans, block_group, path);
2777                 if (ret)
2778                         goto out_free;
2779                 goto again;
2780         }
2781
2782         /*
2783          * We want to set the generation to 0, that way if anything goes wrong
2784          * from here on out we know not to trust this cache when we load up next
2785          * time.
2786          */
2787         BTRFS_I(inode)->generation = 0;
2788         ret = btrfs_update_inode(trans, root, inode);
2789         WARN_ON(ret);
2790
2791         if (i_size_read(inode) > 0) {
2792                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2793                                                       inode);
2794                 if (ret)
2795                         goto out_put;
2796         }
2797
2798         spin_lock(&block_group->lock);
2799         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2800                 /* We're not cached, don't bother trying to write stuff out */
2801                 dcs = BTRFS_DC_WRITTEN;
2802                 spin_unlock(&block_group->lock);
2803                 goto out_put;
2804         }
2805         spin_unlock(&block_group->lock);
2806
2807         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2808         if (!num_pages)
2809                 num_pages = 1;
2810
2811         /*
2812          * Just to make absolutely sure we have enough space, we're going to
2813          * preallocate 12 pages worth of space for each block group.  In
2814          * practice we ought to use at most 8, but we need extra space so we can
2815          * add our header and have a terminator between the extents and the
2816          * bitmaps.
2817          */
2818         num_pages *= 16;
2819         num_pages *= PAGE_CACHE_SIZE;
2820
2821         ret = btrfs_check_data_free_space(inode, num_pages);
2822         if (ret)
2823                 goto out_put;
2824
2825         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2826                                               num_pages, num_pages,
2827                                               &alloc_hint);
2828         if (!ret)
2829                 dcs = BTRFS_DC_SETUP;
2830         btrfs_free_reserved_data_space(inode, num_pages);
2831 out_put:
2832         iput(inode);
2833 out_free:
2834         btrfs_release_path(root, path);
2835 out:
2836         spin_lock(&block_group->lock);
2837         block_group->disk_cache_state = dcs;
2838         spin_unlock(&block_group->lock);
2839
2840         return ret;
2841 }
2842
2843 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2844                                    struct btrfs_root *root)
2845 {
2846         struct btrfs_block_group_cache *cache;
2847         int err = 0;
2848         struct btrfs_path *path;
2849         u64 last = 0;
2850
2851         path = btrfs_alloc_path();
2852         if (!path)
2853                 return -ENOMEM;
2854
2855 again:
2856         while (1) {
2857                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2858                 while (cache) {
2859                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2860                                 break;
2861                         cache = next_block_group(root, cache);
2862                 }
2863                 if (!cache) {
2864                         if (last == 0)
2865                                 break;
2866                         last = 0;
2867                         continue;
2868                 }
2869                 err = cache_save_setup(cache, trans, path);
2870                 last = cache->key.objectid + cache->key.offset;
2871                 btrfs_put_block_group(cache);
2872         }
2873
2874         while (1) {
2875                 if (last == 0) {
2876                         err = btrfs_run_delayed_refs(trans, root,
2877                                                      (unsigned long)-1);
2878                         BUG_ON(err);
2879                 }
2880
2881                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2882                 while (cache) {
2883                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2884                                 btrfs_put_block_group(cache);
2885                                 goto again;
2886                         }
2887
2888                         if (cache->dirty)
2889                                 break;
2890                         cache = next_block_group(root, cache);
2891                 }
2892                 if (!cache) {
2893                         if (last == 0)
2894                                 break;
2895                         last = 0;
2896                         continue;
2897                 }
2898
2899                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2900                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2901                 cache->dirty = 0;
2902                 last = cache->key.objectid + cache->key.offset;
2903
2904                 err = write_one_cache_group(trans, root, path, cache);
2905                 BUG_ON(err);
2906                 btrfs_put_block_group(cache);
2907         }
2908
2909         while (1) {
2910                 /*
2911                  * I don't think this is needed since we're just marking our
2912                  * preallocated extent as written, but just in case it can't
2913                  * hurt.
2914                  */
2915                 if (last == 0) {
2916                         err = btrfs_run_delayed_refs(trans, root,
2917                                                      (unsigned long)-1);
2918                         BUG_ON(err);
2919                 }
2920
2921                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2922                 while (cache) {
2923                         /*
2924                          * Really this shouldn't happen, but it could if we
2925                          * couldn't write the entire preallocated extent and
2926                          * splitting the extent resulted in a new block.
2927                          */
2928                         if (cache->dirty) {
2929                                 btrfs_put_block_group(cache);
2930                                 goto again;
2931                         }
2932                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2933                                 break;
2934                         cache = next_block_group(root, cache);
2935                 }
2936                 if (!cache) {
2937                         if (last == 0)
2938                                 break;
2939                         last = 0;
2940                         continue;
2941                 }
2942
2943                 btrfs_write_out_cache(root, trans, cache, path);
2944
2945                 /*
2946                  * If we didn't have an error then the cache state is still
2947                  * NEED_WRITE, so we can set it to WRITTEN.
2948                  */
2949                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2950                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2951                 last = cache->key.objectid + cache->key.offset;
2952                 btrfs_put_block_group(cache);
2953         }
2954
2955         btrfs_free_path(path);
2956         return 0;
2957 }
2958
2959 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2960 {
2961         struct btrfs_block_group_cache *block_group;
2962         int readonly = 0;
2963
2964         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2965         if (!block_group || block_group->ro)
2966                 readonly = 1;
2967         if (block_group)
2968                 btrfs_put_block_group(block_group);
2969         return readonly;
2970 }
2971
2972 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2973                              u64 total_bytes, u64 bytes_used,
2974                              struct btrfs_space_info **space_info)
2975 {
2976         struct btrfs_space_info *found;
2977         int i;
2978         int factor;
2979
2980         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2981                      BTRFS_BLOCK_GROUP_RAID10))
2982                 factor = 2;
2983         else
2984                 factor = 1;
2985
2986         found = __find_space_info(info, flags);
2987         if (found) {
2988                 spin_lock(&found->lock);
2989                 found->total_bytes += total_bytes;
2990                 found->disk_total += total_bytes * factor;
2991                 found->bytes_used += bytes_used;
2992                 found->disk_used += bytes_used * factor;
2993                 found->full = 0;
2994                 spin_unlock(&found->lock);
2995                 *space_info = found;
2996                 return 0;
2997         }
2998         found = kzalloc(sizeof(*found), GFP_NOFS);
2999         if (!found)
3000                 return -ENOMEM;
3001
3002         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3003                 INIT_LIST_HEAD(&found->block_groups[i]);
3004         init_rwsem(&found->groups_sem);
3005         spin_lock_init(&found->lock);
3006         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3007                                 BTRFS_BLOCK_GROUP_SYSTEM |
3008                                 BTRFS_BLOCK_GROUP_METADATA);
3009         found->total_bytes = total_bytes;
3010         found->disk_total = total_bytes * factor;
3011         found->bytes_used = bytes_used;
3012         found->disk_used = bytes_used * factor;
3013         found->bytes_pinned = 0;
3014         found->bytes_reserved = 0;
3015         found->bytes_readonly = 0;
3016         found->bytes_may_use = 0;
3017         found->full = 0;
3018         found->force_alloc = 0;
3019         *space_info = found;
3020         list_add_rcu(&found->list, &info->space_info);
3021         atomic_set(&found->caching_threads, 0);
3022         return 0;
3023 }
3024
3025 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3026 {
3027         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3028                                    BTRFS_BLOCK_GROUP_RAID1 |
3029                                    BTRFS_BLOCK_GROUP_RAID10 |
3030                                    BTRFS_BLOCK_GROUP_DUP);
3031         if (extra_flags) {
3032                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3033                         fs_info->avail_data_alloc_bits |= extra_flags;
3034                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3035                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3036                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3037                         fs_info->avail_system_alloc_bits |= extra_flags;
3038         }
3039 }
3040
3041 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3042 {
3043         /*
3044          * we add in the count of missing devices because we want
3045          * to make sure that any RAID levels on a degraded FS
3046          * continue to be honored.
3047          */
3048         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3049                 root->fs_info->fs_devices->missing_devices;
3050
3051         if (num_devices == 1)
3052                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3053         if (num_devices < 4)
3054                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3055
3056         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3057             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3058                       BTRFS_BLOCK_GROUP_RAID10))) {
3059                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3060         }
3061
3062         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3063             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3064                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3065         }
3066
3067         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3068             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3069              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3070              (flags & BTRFS_BLOCK_GROUP_DUP)))
3071                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3072         return flags;
3073 }
3074
3075 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3076 {
3077         if (flags & BTRFS_BLOCK_GROUP_DATA)
3078                 flags |= root->fs_info->avail_data_alloc_bits &
3079                          root->fs_info->data_alloc_profile;
3080         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3081                 flags |= root->fs_info->avail_system_alloc_bits &
3082                          root->fs_info->system_alloc_profile;
3083         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3084                 flags |= root->fs_info->avail_metadata_alloc_bits &
3085                          root->fs_info->metadata_alloc_profile;
3086         return btrfs_reduce_alloc_profile(root, flags);
3087 }
3088
3089 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3090 {
3091         u64 flags;
3092
3093         if (data)
3094                 flags = BTRFS_BLOCK_GROUP_DATA;
3095         else if (root == root->fs_info->chunk_root)
3096                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3097         else
3098                 flags = BTRFS_BLOCK_GROUP_METADATA;
3099
3100         return get_alloc_profile(root, flags);
3101 }
3102
3103 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3104 {
3105         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3106                                                        BTRFS_BLOCK_GROUP_DATA);
3107 }
3108
3109 /*
3110  * This will check the space that the inode allocates from to make sure we have
3111  * enough space for bytes.
3112  */
3113 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3114 {
3115         struct btrfs_space_info *data_sinfo;
3116         struct btrfs_root *root = BTRFS_I(inode)->root;
3117         u64 used;
3118         int ret = 0, committed = 0, alloc_chunk = 1;
3119
3120         /* make sure bytes are sectorsize aligned */
3121         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3122
3123         if (root == root->fs_info->tree_root) {
3124                 alloc_chunk = 0;
3125                 committed = 1;
3126         }
3127
3128         data_sinfo = BTRFS_I(inode)->space_info;
3129         if (!data_sinfo)
3130                 goto alloc;
3131
3132 again:
3133         /* make sure we have enough space to handle the data first */
3134         spin_lock(&data_sinfo->lock);
3135         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3136                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3137                 data_sinfo->bytes_may_use;
3138
3139         if (used + bytes > data_sinfo->total_bytes) {
3140                 struct btrfs_trans_handle *trans;
3141
3142                 /*
3143                  * if we don't have enough free bytes in this space then we need
3144                  * to alloc a new chunk.
3145                  */
3146                 if (!data_sinfo->full && alloc_chunk) {
3147                         u64 alloc_target;
3148
3149                         data_sinfo->force_alloc = 1;
3150                         spin_unlock(&data_sinfo->lock);
3151 alloc:
3152                         alloc_target = btrfs_get_alloc_profile(root, 1);
3153                         trans = btrfs_join_transaction(root, 1);
3154                         if (IS_ERR(trans))
3155                                 return PTR_ERR(trans);
3156
3157                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3158                                              bytes + 2 * 1024 * 1024,
3159                                              alloc_target, 0);
3160                         btrfs_end_transaction(trans, root);
3161                         if (ret < 0) {
3162                                 if (ret != -ENOSPC)
3163                                         return ret;
3164                                 else
3165                                         goto commit_trans;
3166                         }
3167
3168                         if (!data_sinfo) {
3169                                 btrfs_set_inode_space_info(root, inode);
3170                                 data_sinfo = BTRFS_I(inode)->space_info;
3171                         }
3172                         goto again;
3173                 }
3174                 spin_unlock(&data_sinfo->lock);
3175
3176                 /* commit the current transaction and try again */
3177 commit_trans:
3178                 if (!committed && !root->fs_info->open_ioctl_trans) {
3179                         committed = 1;
3180                         trans = btrfs_join_transaction(root, 1);
3181                         if (IS_ERR(trans))
3182                                 return PTR_ERR(trans);
3183                         ret = btrfs_commit_transaction(trans, root);
3184                         if (ret)
3185                                 return ret;
3186                         goto again;
3187                 }
3188
3189 #if 0 /* I hope we never need this code again, just in case */
3190                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3191                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3192                        "%llu bytes_readonly, %llu may use %llu total\n",
3193                        (unsigned long long)bytes,
3194                        (unsigned long long)data_sinfo->bytes_used,
3195                        (unsigned long long)data_sinfo->bytes_reserved,
3196                        (unsigned long long)data_sinfo->bytes_pinned,
3197                        (unsigned long long)data_sinfo->bytes_readonly,
3198                        (unsigned long long)data_sinfo->bytes_may_use,
3199                        (unsigned long long)data_sinfo->total_bytes);
3200 #endif
3201                 return -ENOSPC;
3202         }
3203         data_sinfo->bytes_may_use += bytes;
3204         BTRFS_I(inode)->reserved_bytes += bytes;
3205         spin_unlock(&data_sinfo->lock);
3206
3207         return 0;
3208 }
3209
3210 /*
3211  * called when we are clearing an delalloc extent from the
3212  * inode's io_tree or there was an error for whatever reason
3213  * after calling btrfs_check_data_free_space
3214  */
3215 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3216 {
3217         struct btrfs_root *root = BTRFS_I(inode)->root;
3218         struct btrfs_space_info *data_sinfo;
3219
3220         /* make sure bytes are sectorsize aligned */
3221         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3222
3223         data_sinfo = BTRFS_I(inode)->space_info;
3224         spin_lock(&data_sinfo->lock);
3225         data_sinfo->bytes_may_use -= bytes;
3226         BTRFS_I(inode)->reserved_bytes -= bytes;
3227         spin_unlock(&data_sinfo->lock);
3228 }
3229
3230 static void force_metadata_allocation(struct btrfs_fs_info *info)
3231 {
3232         struct list_head *head = &info->space_info;
3233         struct btrfs_space_info *found;
3234
3235         rcu_read_lock();
3236         list_for_each_entry_rcu(found, head, list) {
3237                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3238                         found->force_alloc = 1;
3239         }
3240         rcu_read_unlock();
3241 }
3242
3243 static int should_alloc_chunk(struct btrfs_root *root,
3244                               struct btrfs_space_info *sinfo, u64 alloc_bytes)
3245 {
3246         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3247         u64 thresh;
3248
3249         if (sinfo->bytes_used + sinfo->bytes_reserved +
3250             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3251                 return 0;
3252
3253         if (sinfo->bytes_used + sinfo->bytes_reserved +
3254             alloc_bytes < div_factor(num_bytes, 8))
3255                 return 0;
3256
3257         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3258         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3259
3260         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3261                 return 0;
3262
3263         return 1;
3264 }
3265
3266 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3267                           struct btrfs_root *extent_root, u64 alloc_bytes,
3268                           u64 flags, int force)
3269 {
3270         struct btrfs_space_info *space_info;
3271         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3272         int ret = 0;
3273
3274         mutex_lock(&fs_info->chunk_mutex);
3275
3276         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3277
3278         space_info = __find_space_info(extent_root->fs_info, flags);
3279         if (!space_info) {
3280                 ret = update_space_info(extent_root->fs_info, flags,
3281                                         0, 0, &space_info);
3282                 BUG_ON(ret);
3283         }
3284         BUG_ON(!space_info);
3285
3286         spin_lock(&space_info->lock);
3287         if (space_info->force_alloc)
3288                 force = 1;
3289         if (space_info->full) {
3290                 spin_unlock(&space_info->lock);
3291                 goto out;
3292         }
3293
3294         if (!force && !should_alloc_chunk(extent_root, space_info,
3295                                           alloc_bytes)) {
3296                 spin_unlock(&space_info->lock);
3297                 goto out;
3298         }
3299         spin_unlock(&space_info->lock);
3300
3301         /*
3302          * If we have mixed data/metadata chunks we want to make sure we keep
3303          * allocating mixed chunks instead of individual chunks.
3304          */
3305         if (btrfs_mixed_space_info(space_info))
3306                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3307
3308         /*
3309          * if we're doing a data chunk, go ahead and make sure that
3310          * we keep a reasonable number of metadata chunks allocated in the
3311          * FS as well.
3312          */
3313         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3314                 fs_info->data_chunk_allocations++;
3315                 if (!(fs_info->data_chunk_allocations %
3316                       fs_info->metadata_ratio))
3317                         force_metadata_allocation(fs_info);
3318         }
3319
3320         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3321         spin_lock(&space_info->lock);
3322         if (ret)
3323                 space_info->full = 1;
3324         else
3325                 ret = 1;
3326         space_info->force_alloc = 0;
3327         spin_unlock(&space_info->lock);
3328 out:
3329         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3330         return ret;
3331 }
3332
3333 /*
3334  * shrink metadata reservation for delalloc
3335  */
3336 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3337                            struct btrfs_root *root, u64 to_reclaim, int sync)
3338 {
3339         struct btrfs_block_rsv *block_rsv;
3340         struct btrfs_space_info *space_info;
3341         u64 reserved;
3342         u64 max_reclaim;
3343         u64 reclaimed = 0;
3344         long time_left;
3345         int pause = 1;
3346         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3347         int loops = 0;
3348
3349         block_rsv = &root->fs_info->delalloc_block_rsv;
3350         space_info = block_rsv->space_info;
3351
3352         smp_mb();
3353         reserved = space_info->bytes_reserved;
3354
3355         if (reserved == 0)
3356                 return 0;
3357
3358         max_reclaim = min(reserved, to_reclaim);
3359
3360         while (loops < 1024) {
3361                 /* have the flusher threads jump in and do some IO */
3362                 smp_mb();
3363                 nr_pages = min_t(unsigned long, nr_pages,
3364                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3365                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3366
3367                 spin_lock(&space_info->lock);
3368                 if (reserved > space_info->bytes_reserved) {
3369                         loops = 0;
3370                         reclaimed += reserved - space_info->bytes_reserved;
3371                 } else {
3372                         loops++;
3373                 }
3374                 reserved = space_info->bytes_reserved;
3375                 spin_unlock(&space_info->lock);
3376
3377                 if (reserved == 0 || reclaimed >= max_reclaim)
3378                         break;
3379
3380                 if (trans && trans->transaction->blocked)
3381                         return -EAGAIN;
3382
3383                 __set_current_state(TASK_INTERRUPTIBLE);
3384                 time_left = schedule_timeout(pause);
3385
3386                 /* We were interrupted, exit */
3387                 if (time_left)
3388                         break;
3389
3390                 pause <<= 1;
3391                 if (pause > HZ / 10)
3392                         pause = HZ / 10;
3393
3394         }
3395         return reclaimed >= to_reclaim;
3396 }
3397
3398 /*
3399  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3400  * idea is if this is the first call (retries == 0) then we will add to our
3401  * reserved count if we can't make the allocation in order to hold our place
3402  * while we go and try and free up space.  That way for retries > 1 we don't try
3403  * and add space, we just check to see if the amount of unused space is >= the
3404  * total space, meaning that our reservation is valid.
3405  *
3406  * However if we don't intend to retry this reservation, pass -1 as retries so
3407  * that it short circuits this logic.
3408  */
3409 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3410                                   struct btrfs_root *root,
3411                                   struct btrfs_block_rsv *block_rsv,
3412                                   u64 orig_bytes, int flush)
3413 {
3414         struct btrfs_space_info *space_info = block_rsv->space_info;
3415         u64 unused;
3416         u64 num_bytes = orig_bytes;
3417         int retries = 0;
3418         int ret = 0;
3419         bool reserved = false;
3420         bool committed = false;
3421
3422 again:
3423         ret = -ENOSPC;
3424         if (reserved)
3425                 num_bytes = 0;
3426
3427         spin_lock(&space_info->lock);
3428         unused = space_info->bytes_used + space_info->bytes_reserved +
3429                  space_info->bytes_pinned + space_info->bytes_readonly +
3430                  space_info->bytes_may_use;
3431
3432         /*
3433          * The idea here is that we've not already over-reserved the block group
3434          * then we can go ahead and save our reservation first and then start
3435          * flushing if we need to.  Otherwise if we've already overcommitted
3436          * lets start flushing stuff first and then come back and try to make
3437          * our reservation.
3438          */
3439         if (unused <= space_info->total_bytes) {
3440                 unused = space_info->total_bytes - unused;
3441                 if (unused >= num_bytes) {
3442                         if (!reserved)
3443                                 space_info->bytes_reserved += orig_bytes;
3444                         ret = 0;
3445                 } else {
3446                         /*
3447                          * Ok set num_bytes to orig_bytes since we aren't
3448                          * overocmmitted, this way we only try and reclaim what
3449                          * we need.
3450                          */
3451                         num_bytes = orig_bytes;
3452                 }
3453         } else {
3454                 /*
3455                  * Ok we're over committed, set num_bytes to the overcommitted
3456                  * amount plus the amount of bytes that we need for this
3457                  * reservation.
3458                  */
3459                 num_bytes = unused - space_info->total_bytes +
3460                         (orig_bytes * (retries + 1));
3461         }
3462
3463         /*
3464          * Couldn't make our reservation, save our place so while we're trying
3465          * to reclaim space we can actually use it instead of somebody else
3466          * stealing it from us.
3467          */
3468         if (ret && !reserved) {
3469                 space_info->bytes_reserved += orig_bytes;
3470                 reserved = true;
3471         }
3472
3473         spin_unlock(&space_info->lock);
3474
3475         if (!ret)
3476                 return 0;
3477
3478         if (!flush)
3479                 goto out;
3480
3481         /*
3482          * We do synchronous shrinking since we don't actually unreserve
3483          * metadata until after the IO is completed.
3484          */
3485         ret = shrink_delalloc(trans, root, num_bytes, 1);
3486         if (ret > 0)
3487                 return 0;
3488         else if (ret < 0)
3489                 goto out;
3490
3491         /*
3492          * So if we were overcommitted it's possible that somebody else flushed
3493          * out enough space and we simply didn't have enough space to reclaim,
3494          * so go back around and try again.
3495          */
3496         if (retries < 2) {
3497                 retries++;
3498                 goto again;
3499         }
3500
3501         spin_lock(&space_info->lock);
3502         /*
3503          * Not enough space to be reclaimed, don't bother committing the
3504          * transaction.
3505          */
3506         if (space_info->bytes_pinned < orig_bytes)
3507                 ret = -ENOSPC;
3508         spin_unlock(&space_info->lock);
3509         if (ret)
3510                 goto out;
3511
3512         ret = -EAGAIN;
3513         if (trans || committed)
3514                 goto out;
3515
3516         ret = -ENOSPC;
3517         trans = btrfs_join_transaction(root, 1);
3518         if (IS_ERR(trans))
3519                 goto out;
3520         ret = btrfs_commit_transaction(trans, root);
3521         if (!ret) {
3522                 trans = NULL;
3523                 committed = true;
3524                 goto again;
3525         }
3526
3527 out:
3528         if (reserved) {
3529                 spin_lock(&space_info->lock);
3530                 space_info->bytes_reserved -= orig_bytes;
3531                 spin_unlock(&space_info->lock);
3532         }
3533
3534         return ret;
3535 }
3536
3537 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3538                                              struct btrfs_root *root)
3539 {
3540         struct btrfs_block_rsv *block_rsv;
3541         if (root->ref_cows)
3542                 block_rsv = trans->block_rsv;
3543         else
3544                 block_rsv = root->block_rsv;
3545
3546         if (!block_rsv)
3547                 block_rsv = &root->fs_info->empty_block_rsv;
3548
3549         return block_rsv;
3550 }
3551
3552 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3553                                u64 num_bytes)
3554 {
3555         int ret = -ENOSPC;
3556         spin_lock(&block_rsv->lock);
3557         if (block_rsv->reserved >= num_bytes) {
3558                 block_rsv->reserved -= num_bytes;
3559                 if (block_rsv->reserved < block_rsv->size)
3560                         block_rsv->full = 0;
3561                 ret = 0;
3562         }
3563         spin_unlock(&block_rsv->lock);
3564         return ret;
3565 }
3566
3567 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3568                                 u64 num_bytes, int update_size)
3569 {
3570         spin_lock(&block_rsv->lock);
3571         block_rsv->reserved += num_bytes;
3572         if (update_size)
3573                 block_rsv->size += num_bytes;
3574         else if (block_rsv->reserved >= block_rsv->size)
3575                 block_rsv->full = 1;
3576         spin_unlock(&block_rsv->lock);
3577 }
3578
3579 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3580                              struct btrfs_block_rsv *dest, u64 num_bytes)
3581 {
3582         struct btrfs_space_info *space_info = block_rsv->space_info;
3583
3584         spin_lock(&block_rsv->lock);
3585         if (num_bytes == (u64)-1)
3586                 num_bytes = block_rsv->size;
3587         block_rsv->size -= num_bytes;
3588         if (block_rsv->reserved >= block_rsv->size) {
3589                 num_bytes = block_rsv->reserved - block_rsv->size;
3590                 block_rsv->reserved = block_rsv->size;
3591                 block_rsv->full = 1;
3592         } else {
3593                 num_bytes = 0;
3594         }
3595         spin_unlock(&block_rsv->lock);
3596
3597         if (num_bytes > 0) {
3598                 if (dest) {
3599                         spin_lock(&dest->lock);
3600                         if (!dest->full) {
3601                                 u64 bytes_to_add;
3602
3603                                 bytes_to_add = dest->size - dest->reserved;
3604                                 bytes_to_add = min(num_bytes, bytes_to_add);
3605                                 dest->reserved += bytes_to_add;
3606                                 if (dest->reserved >= dest->size)
3607                                         dest->full = 1;
3608                                 num_bytes -= bytes_to_add;
3609                         }
3610                         spin_unlock(&dest->lock);
3611                 }
3612                 if (num_bytes) {
3613                         spin_lock(&space_info->lock);
3614                         space_info->bytes_reserved -= num_bytes;
3615                         spin_unlock(&space_info->lock);
3616                 }
3617         }
3618 }
3619
3620 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3621                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3622 {
3623         int ret;
3624
3625         ret = block_rsv_use_bytes(src, num_bytes);
3626         if (ret)
3627                 return ret;
3628
3629         block_rsv_add_bytes(dst, num_bytes, 1);
3630         return 0;
3631 }
3632
3633 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3634 {
3635         memset(rsv, 0, sizeof(*rsv));
3636         spin_lock_init(&rsv->lock);
3637         atomic_set(&rsv->usage, 1);
3638         rsv->priority = 6;
3639         INIT_LIST_HEAD(&rsv->list);
3640 }
3641
3642 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3643 {
3644         struct btrfs_block_rsv *block_rsv;
3645         struct btrfs_fs_info *fs_info = root->fs_info;
3646
3647         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3648         if (!block_rsv)
3649                 return NULL;
3650
3651         btrfs_init_block_rsv(block_rsv);
3652         block_rsv->space_info = __find_space_info(fs_info,
3653                                                   BTRFS_BLOCK_GROUP_METADATA);
3654         return block_rsv;
3655 }
3656
3657 void btrfs_free_block_rsv(struct btrfs_root *root,
3658                           struct btrfs_block_rsv *rsv)
3659 {
3660         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3661                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3662                 if (!rsv->durable)
3663                         kfree(rsv);
3664         }
3665 }
3666
3667 /*
3668  * make the block_rsv struct be able to capture freed space.
3669  * the captured space will re-add to the the block_rsv struct
3670  * after transaction commit
3671  */
3672 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3673                                  struct btrfs_block_rsv *block_rsv)
3674 {
3675         block_rsv->durable = 1;
3676         mutex_lock(&fs_info->durable_block_rsv_mutex);
3677         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3678         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3679 }
3680
3681 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3682                         struct btrfs_root *root,
3683                         struct btrfs_block_rsv *block_rsv,
3684                         u64 num_bytes)
3685 {
3686         int ret;
3687
3688         if (num_bytes == 0)
3689                 return 0;
3690
3691         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3692         if (!ret) {
3693                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3694                 return 0;
3695         }
3696
3697         return ret;
3698 }
3699
3700 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3701                           struct btrfs_root *root,
3702                           struct btrfs_block_rsv *block_rsv,
3703                           u64 min_reserved, int min_factor)
3704 {
3705         u64 num_bytes = 0;
3706         int commit_trans = 0;
3707         int ret = -ENOSPC;
3708
3709         if (!block_rsv)
3710                 return 0;
3711
3712         spin_lock(&block_rsv->lock);
3713         if (min_factor > 0)
3714                 num_bytes = div_factor(block_rsv->size, min_factor);
3715         if (min_reserved > num_bytes)
3716                 num_bytes = min_reserved;
3717
3718         if (block_rsv->reserved >= num_bytes) {
3719                 ret = 0;
3720         } else {
3721                 num_bytes -= block_rsv->reserved;
3722                 if (block_rsv->durable &&
3723                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3724                         commit_trans = 1;
3725         }
3726         spin_unlock(&block_rsv->lock);
3727         if (!ret)
3728                 return 0;
3729
3730         if (block_rsv->refill_used) {
3731                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3732                                              num_bytes, 0);
3733                 if (!ret) {
3734                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3735                         return 0;
3736                 }
3737         }
3738
3739         if (commit_trans) {
3740                 if (trans)
3741                         return -EAGAIN;
3742
3743                 trans = btrfs_join_transaction(root, 1);
3744                 BUG_ON(IS_ERR(trans));
3745                 ret = btrfs_commit_transaction(trans, root);
3746                 return 0;
3747         }
3748
3749         return -ENOSPC;
3750 }
3751
3752 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3753                             struct btrfs_block_rsv *dst_rsv,
3754                             u64 num_bytes)
3755 {
3756         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3757 }
3758
3759 void btrfs_block_rsv_release(struct btrfs_root *root,
3760                              struct btrfs_block_rsv *block_rsv,
3761                              u64 num_bytes)
3762 {
3763         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3764         if (global_rsv->full || global_rsv == block_rsv ||
3765             block_rsv->space_info != global_rsv->space_info)
3766                 global_rsv = NULL;
3767         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3768 }
3769
3770 /*
3771  * helper to calculate size of global block reservation.
3772  * the desired value is sum of space used by extent tree,
3773  * checksum tree and root tree
3774  */
3775 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3776 {
3777         struct btrfs_space_info *sinfo;
3778         u64 num_bytes;
3779         u64 meta_used;
3780         u64 data_used;
3781         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3782 #if 0
3783         /*
3784          * per tree used space accounting can be inaccuracy, so we
3785          * can't rely on it.
3786          */
3787         spin_lock(&fs_info->extent_root->accounting_lock);
3788         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3789         spin_unlock(&fs_info->extent_root->accounting_lock);
3790
3791         spin_lock(&fs_info->csum_root->accounting_lock);
3792         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3793         spin_unlock(&fs_info->csum_root->accounting_lock);
3794
3795         spin_lock(&fs_info->tree_root->accounting_lock);
3796         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3797         spin_unlock(&fs_info->tree_root->accounting_lock);
3798 #endif
3799         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3800         spin_lock(&sinfo->lock);
3801         data_used = sinfo->bytes_used;
3802         spin_unlock(&sinfo->lock);
3803
3804         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3805         spin_lock(&sinfo->lock);
3806         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3807                 data_used = 0;
3808         meta_used = sinfo->bytes_used;
3809         spin_unlock(&sinfo->lock);
3810
3811         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3812                     csum_size * 2;
3813         num_bytes += div64_u64(data_used + meta_used, 50);
3814
3815         if (num_bytes * 3 > meta_used)
3816                 num_bytes = div64_u64(meta_used, 3);
3817
3818         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3819 }
3820
3821 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3822 {
3823         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3824         struct btrfs_space_info *sinfo = block_rsv->space_info;
3825         u64 num_bytes;
3826
3827         num_bytes = calc_global_metadata_size(fs_info);
3828
3829         spin_lock(&block_rsv->lock);
3830         spin_lock(&sinfo->lock);
3831
3832         block_rsv->size = num_bytes;
3833
3834         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3835                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3836                     sinfo->bytes_may_use;
3837
3838         if (sinfo->total_bytes > num_bytes) {
3839                 num_bytes = sinfo->total_bytes - num_bytes;
3840                 block_rsv->reserved += num_bytes;
3841                 sinfo->bytes_reserved += num_bytes;
3842         }
3843
3844         if (block_rsv->reserved >= block_rsv->size) {
3845                 num_bytes = block_rsv->reserved - block_rsv->size;
3846                 sinfo->bytes_reserved -= num_bytes;
3847                 block_rsv->reserved = block_rsv->size;
3848                 block_rsv->full = 1;
3849         }
3850 #if 0
3851         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3852                 block_rsv->size, block_rsv->reserved);
3853 #endif
3854         spin_unlock(&sinfo->lock);
3855         spin_unlock(&block_rsv->lock);
3856 }
3857
3858 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3859 {
3860         struct btrfs_space_info *space_info;
3861
3862         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3863         fs_info->chunk_block_rsv.space_info = space_info;
3864         fs_info->chunk_block_rsv.priority = 10;
3865
3866         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3867         fs_info->global_block_rsv.space_info = space_info;
3868         fs_info->global_block_rsv.priority = 10;
3869         fs_info->global_block_rsv.refill_used = 1;
3870         fs_info->delalloc_block_rsv.space_info = space_info;
3871         fs_info->trans_block_rsv.space_info = space_info;
3872         fs_info->empty_block_rsv.space_info = space_info;
3873         fs_info->empty_block_rsv.priority = 10;
3874
3875         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3876         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3877         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3878         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3879         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3880
3881         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3882
3883         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3884
3885         update_global_block_rsv(fs_info);
3886 }
3887
3888 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3889 {
3890         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3891         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3892         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3893         WARN_ON(fs_info->trans_block_rsv.size > 0);
3894         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3895         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3896         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3897 }
3898
3899 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3900 {
3901         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3902                 3 * num_items;
3903 }
3904
3905 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3906                                  struct btrfs_root *root,
3907                                  int num_items)
3908 {
3909         u64 num_bytes;
3910         int ret;
3911
3912         if (num_items == 0 || root->fs_info->chunk_root == root)
3913                 return 0;
3914
3915         num_bytes = calc_trans_metadata_size(root, num_items);
3916         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3917                                   num_bytes);
3918         if (!ret) {
3919                 trans->bytes_reserved += num_bytes;
3920                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3921         }
3922         return ret;
3923 }
3924
3925 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3926                                   struct btrfs_root *root)
3927 {
3928         if (!trans->bytes_reserved)
3929                 return;
3930
3931         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3932         btrfs_block_rsv_release(root, trans->block_rsv,
3933                                 trans->bytes_reserved);
3934         trans->bytes_reserved = 0;
3935 }
3936
3937 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3938                                   struct inode *inode)
3939 {
3940         struct btrfs_root *root = BTRFS_I(inode)->root;
3941         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3942         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3943
3944         /*
3945          * one for deleting orphan item, one for updating inode and
3946          * two for calling btrfs_truncate_inode_items.
3947          *
3948          * btrfs_truncate_inode_items is a delete operation, it frees
3949          * more space than it uses in most cases. So two units of
3950          * metadata space should be enough for calling it many times.
3951          * If all of the metadata space is used, we can commit
3952          * transaction and use space it freed.
3953          */
3954         u64 num_bytes = calc_trans_metadata_size(root, 4);
3955         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3956 }
3957
3958 void btrfs_orphan_release_metadata(struct inode *inode)
3959 {
3960         struct btrfs_root *root = BTRFS_I(inode)->root;
3961         u64 num_bytes = calc_trans_metadata_size(root, 4);
3962         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3963 }
3964
3965 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3966                                 struct btrfs_pending_snapshot *pending)
3967 {
3968         struct btrfs_root *root = pending->root;
3969         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3970         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3971         /*
3972          * two for root back/forward refs, two for directory entries
3973          * and one for root of the snapshot.
3974          */
3975         u64 num_bytes = calc_trans_metadata_size(root, 5);
3976         dst_rsv->space_info = src_rsv->space_info;
3977         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3978 }
3979
3980 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3981 {
3982         return num_bytes >>= 3;
3983 }
3984
3985 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3986 {
3987         struct btrfs_root *root = BTRFS_I(inode)->root;
3988         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3989         u64 to_reserve;
3990         int nr_extents;
3991         int ret;
3992
3993         if (btrfs_transaction_in_commit(root->fs_info))
3994                 schedule_timeout(1);
3995
3996         num_bytes = ALIGN(num_bytes, root->sectorsize);
3997
3998         spin_lock(&BTRFS_I(inode)->accounting_lock);
3999         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
4000         if (nr_extents > BTRFS_I(inode)->reserved_extents) {
4001                 nr_extents -= BTRFS_I(inode)->reserved_extents;
4002                 to_reserve = calc_trans_metadata_size(root, nr_extents);
4003         } else {
4004                 nr_extents = 0;
4005                 to_reserve = 0;
4006         }
4007         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4008
4009         to_reserve += calc_csum_metadata_size(inode, num_bytes);
4010         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4011         if (ret)
4012                 return ret;
4013
4014         spin_lock(&BTRFS_I(inode)->accounting_lock);
4015         BTRFS_I(inode)->reserved_extents += nr_extents;
4016         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4017         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4018
4019         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4020
4021         if (block_rsv->size > 512 * 1024 * 1024)
4022                 shrink_delalloc(NULL, root, to_reserve, 0);
4023
4024         return 0;
4025 }
4026
4027 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4028 {
4029         struct btrfs_root *root = BTRFS_I(inode)->root;
4030         u64 to_free;
4031         int nr_extents;
4032
4033         num_bytes = ALIGN(num_bytes, root->sectorsize);
4034         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4035         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
4036
4037         spin_lock(&BTRFS_I(inode)->accounting_lock);
4038         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4039         if (nr_extents < BTRFS_I(inode)->reserved_extents) {
4040                 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
4041                 BTRFS_I(inode)->reserved_extents -= nr_extents;
4042         } else {
4043                 nr_extents = 0;
4044         }
4045         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4046
4047         to_free = calc_csum_metadata_size(inode, num_bytes);
4048         if (nr_extents > 0)
4049                 to_free += calc_trans_metadata_size(root, nr_extents);
4050
4051         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4052                                 to_free);
4053 }
4054
4055 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4056 {
4057         int ret;
4058
4059         ret = btrfs_check_data_free_space(inode, num_bytes);
4060         if (ret)
4061                 return ret;
4062
4063         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4064         if (ret) {
4065                 btrfs_free_reserved_data_space(inode, num_bytes);
4066                 return ret;
4067         }
4068
4069         return 0;
4070 }
4071
4072 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4073 {
4074         btrfs_delalloc_release_metadata(inode, num_bytes);
4075         btrfs_free_reserved_data_space(inode, num_bytes);
4076 }
4077
4078 static int update_block_group(struct btrfs_trans_handle *trans,
4079                               struct btrfs_root *root,
4080                               u64 bytenr, u64 num_bytes, int alloc)
4081 {
4082         struct btrfs_block_group_cache *cache = NULL;
4083         struct btrfs_fs_info *info = root->fs_info;
4084         u64 total = num_bytes;
4085         u64 old_val;
4086         u64 byte_in_group;
4087         int factor;
4088
4089         /* block accounting for super block */
4090         spin_lock(&info->delalloc_lock);
4091         old_val = btrfs_super_bytes_used(&info->super_copy);
4092         if (alloc)
4093                 old_val += num_bytes;
4094         else
4095                 old_val -= num_bytes;
4096         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4097         spin_unlock(&info->delalloc_lock);
4098
4099         while (total) {
4100                 cache = btrfs_lookup_block_group(info, bytenr);
4101                 if (!cache)
4102                         return -1;
4103                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4104                                     BTRFS_BLOCK_GROUP_RAID1 |
4105                                     BTRFS_BLOCK_GROUP_RAID10))
4106                         factor = 2;
4107                 else
4108                         factor = 1;
4109                 /*
4110                  * If this block group has free space cache written out, we
4111                  * need to make sure to load it if we are removing space.  This
4112                  * is because we need the unpinning stage to actually add the
4113                  * space back to the block group, otherwise we will leak space.
4114                  */
4115                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4116                         cache_block_group(cache, trans, NULL, 1);
4117
4118                 byte_in_group = bytenr - cache->key.objectid;
4119                 WARN_ON(byte_in_group > cache->key.offset);
4120
4121                 spin_lock(&cache->space_info->lock);
4122                 spin_lock(&cache->lock);
4123
4124                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4125                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4126                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4127
4128                 cache->dirty = 1;
4129                 old_val = btrfs_block_group_used(&cache->item);
4130                 num_bytes = min(total, cache->key.offset - byte_in_group);
4131                 if (alloc) {
4132                         old_val += num_bytes;
4133                         btrfs_set_block_group_used(&cache->item, old_val);
4134                         cache->reserved -= num_bytes;
4135                         cache->space_info->bytes_reserved -= num_bytes;
4136                         cache->space_info->bytes_used += num_bytes;
4137                         cache->space_info->disk_used += num_bytes * factor;
4138                         spin_unlock(&cache->lock);
4139                         spin_unlock(&cache->space_info->lock);
4140                 } else {
4141                         old_val -= num_bytes;
4142                         btrfs_set_block_group_used(&cache->item, old_val);
4143                         cache->pinned += num_bytes;
4144                         cache->space_info->bytes_pinned += num_bytes;
4145                         cache->space_info->bytes_used -= num_bytes;
4146                         cache->space_info->disk_used -= num_bytes * factor;
4147                         spin_unlock(&cache->lock);
4148                         spin_unlock(&cache->space_info->lock);
4149
4150                         set_extent_dirty(info->pinned_extents,
4151                                          bytenr, bytenr + num_bytes - 1,
4152                                          GFP_NOFS | __GFP_NOFAIL);
4153                 }
4154                 btrfs_put_block_group(cache);
4155                 total -= num_bytes;
4156                 bytenr += num_bytes;
4157         }
4158         return 0;
4159 }
4160
4161 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4162 {
4163         struct btrfs_block_group_cache *cache;
4164         u64 bytenr;
4165
4166         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4167         if (!cache)
4168                 return 0;
4169
4170         bytenr = cache->key.objectid;
4171         btrfs_put_block_group(cache);
4172
4173         return bytenr;
4174 }
4175
4176 static int pin_down_extent(struct btrfs_root *root,
4177                            struct btrfs_block_group_cache *cache,
4178                            u64 bytenr, u64 num_bytes, int reserved)
4179 {
4180         spin_lock(&cache->space_info->lock);
4181         spin_lock(&cache->lock);
4182         cache->pinned += num_bytes;
4183         cache->space_info->bytes_pinned += num_bytes;
4184         if (reserved) {
4185                 cache->reserved -= num_bytes;
4186                 cache->space_info->bytes_reserved -= num_bytes;
4187         }
4188         spin_unlock(&cache->lock);
4189         spin_unlock(&cache->space_info->lock);
4190
4191         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4192                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4193         return 0;
4194 }
4195
4196 /*
4197  * this function must be called within transaction
4198  */
4199 int btrfs_pin_extent(struct btrfs_root *root,
4200                      u64 bytenr, u64 num_bytes, int reserved)
4201 {
4202         struct btrfs_block_group_cache *cache;
4203
4204         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4205         BUG_ON(!cache);
4206
4207         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4208
4209         btrfs_put_block_group(cache);
4210         return 0;
4211 }
4212
4213 /*
4214  * update size of reserved extents. this function may return -EAGAIN
4215  * if 'reserve' is true or 'sinfo' is false.
4216  */
4217 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4218                                  u64 num_bytes, int reserve, int sinfo)
4219 {
4220         int ret = 0;
4221         if (sinfo) {
4222                 struct btrfs_space_info *space_info = cache->space_info;
4223                 spin_lock(&space_info->lock);
4224                 spin_lock(&cache->lock);
4225                 if (reserve) {
4226                         if (cache->ro) {
4227                                 ret = -EAGAIN;
4228                         } else {
4229                                 cache->reserved += num_bytes;
4230                                 space_info->bytes_reserved += num_bytes;
4231                         }
4232                 } else {
4233                         if (cache->ro)
4234                                 space_info->bytes_readonly += num_bytes;
4235                         cache->reserved -= num_bytes;
4236                         space_info->bytes_reserved -= num_bytes;
4237                 }
4238                 spin_unlock(&cache->lock);
4239                 spin_unlock(&space_info->lock);
4240         } else {
4241                 spin_lock(&cache->lock);
4242                 if (cache->ro) {
4243                         ret = -EAGAIN;
4244                 } else {
4245                         if (reserve)
4246                                 cache->reserved += num_bytes;
4247                         else
4248                                 cache->reserved -= num_bytes;
4249                 }
4250                 spin_unlock(&cache->lock);
4251         }
4252         return ret;
4253 }
4254
4255 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4256                                 struct btrfs_root *root)
4257 {
4258         struct btrfs_fs_info *fs_info = root->fs_info;
4259         struct btrfs_caching_control *next;
4260         struct btrfs_caching_control *caching_ctl;
4261         struct btrfs_block_group_cache *cache;
4262
4263         down_write(&fs_info->extent_commit_sem);
4264
4265         list_for_each_entry_safe(caching_ctl, next,
4266                                  &fs_info->caching_block_groups, list) {
4267                 cache = caching_ctl->block_group;
4268                 if (block_group_cache_done(cache)) {
4269                         cache->last_byte_to_unpin = (u64)-1;
4270                         list_del_init(&caching_ctl->list);
4271                         put_caching_control(caching_ctl);
4272                 } else {
4273                         cache->last_byte_to_unpin = caching_ctl->progress;
4274                 }
4275         }
4276
4277         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4278                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4279         else
4280                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4281
4282         up_write(&fs_info->extent_commit_sem);
4283
4284         update_global_block_rsv(fs_info);
4285         return 0;
4286 }
4287
4288 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4289 {
4290         struct btrfs_fs_info *fs_info = root->fs_info;
4291         struct btrfs_block_group_cache *cache = NULL;
4292         u64 len;
4293
4294         while (start <= end) {
4295                 if (!cache ||
4296                     start >= cache->key.objectid + cache->key.offset) {
4297                         if (cache)
4298                                 btrfs_put_block_group(cache);
4299                         cache = btrfs_lookup_block_group(fs_info, start);
4300                         BUG_ON(!cache);
4301                 }
4302
4303                 len = cache->key.objectid + cache->key.offset - start;
4304                 len = min(len, end + 1 - start);
4305
4306                 if (start < cache->last_byte_to_unpin) {
4307                         len = min(len, cache->last_byte_to_unpin - start);
4308                         btrfs_add_free_space(cache, start, len);
4309                 }
4310
4311                 start += len;
4312
4313                 spin_lock(&cache->space_info->lock);
4314                 spin_lock(&cache->lock);
4315                 cache->pinned -= len;
4316                 cache->space_info->bytes_pinned -= len;
4317                 if (cache->ro) {
4318                         cache->space_info->bytes_readonly += len;
4319                 } else if (cache->reserved_pinned > 0) {
4320                         len = min(len, cache->reserved_pinned);
4321                         cache->reserved_pinned -= len;
4322                         cache->space_info->bytes_reserved += len;
4323                 }
4324                 spin_unlock(&cache->lock);
4325                 spin_unlock(&cache->space_info->lock);
4326         }
4327
4328         if (cache)
4329                 btrfs_put_block_group(cache);
4330         return 0;
4331 }
4332
4333 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4334                                struct btrfs_root *root)
4335 {
4336         struct btrfs_fs_info *fs_info = root->fs_info;
4337         struct extent_io_tree *unpin;
4338         struct btrfs_block_rsv *block_rsv;
4339         struct btrfs_block_rsv *next_rsv;
4340         u64 start;
4341         u64 end;
4342         int idx;
4343         int ret;
4344
4345         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4346                 unpin = &fs_info->freed_extents[1];
4347         else
4348                 unpin = &fs_info->freed_extents[0];
4349
4350         while (1) {
4351                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4352                                             EXTENT_DIRTY);
4353                 if (ret)
4354                         break;
4355
4356                 ret = btrfs_discard_extent(root, start, end + 1 - start);
4357
4358                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4359                 unpin_extent_range(root, start, end);
4360                 cond_resched();
4361         }
4362
4363         mutex_lock(&fs_info->durable_block_rsv_mutex);
4364         list_for_each_entry_safe(block_rsv, next_rsv,
4365                                  &fs_info->durable_block_rsv_list, list) {
4366
4367                 idx = trans->transid & 0x1;
4368                 if (block_rsv->freed[idx] > 0) {
4369                         block_rsv_add_bytes(block_rsv,
4370                                             block_rsv->freed[idx], 0);
4371                         block_rsv->freed[idx] = 0;
4372                 }
4373                 if (atomic_read(&block_rsv->usage) == 0) {
4374                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4375
4376                         if (block_rsv->freed[0] == 0 &&
4377                             block_rsv->freed[1] == 0) {
4378                                 list_del_init(&block_rsv->list);
4379                                 kfree(block_rsv);
4380                         }
4381                 } else {
4382                         btrfs_block_rsv_release(root, block_rsv, 0);
4383                 }
4384         }
4385         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4386
4387         return 0;
4388 }
4389
4390 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4391                                 struct btrfs_root *root,
4392                                 u64 bytenr, u64 num_bytes, u64 parent,
4393                                 u64 root_objectid, u64 owner_objectid,
4394                                 u64 owner_offset, int refs_to_drop,
4395                                 struct btrfs_delayed_extent_op *extent_op)
4396 {
4397         struct btrfs_key key;
4398         struct btrfs_path *path;
4399         struct btrfs_fs_info *info = root->fs_info;
4400         struct btrfs_root *extent_root = info->extent_root;
4401         struct extent_buffer *leaf;
4402         struct btrfs_extent_item *ei;
4403         struct btrfs_extent_inline_ref *iref;
4404         int ret;
4405         int is_data;
4406         int extent_slot = 0;
4407         int found_extent = 0;
4408         int num_to_del = 1;
4409         u32 item_size;
4410         u64 refs;
4411
4412         path = btrfs_alloc_path();
4413         if (!path)
4414                 return -ENOMEM;
4415
4416         path->reada = 1;
4417         path->leave_spinning = 1;
4418
4419         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4420         BUG_ON(!is_data && refs_to_drop != 1);
4421
4422         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4423                                     bytenr, num_bytes, parent,
4424                                     root_objectid, owner_objectid,
4425                                     owner_offset);
4426         if (ret == 0) {
4427                 extent_slot = path->slots[0];
4428                 while (extent_slot >= 0) {
4429                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4430                                               extent_slot);
4431                         if (key.objectid != bytenr)
4432                                 break;
4433                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4434                             key.offset == num_bytes) {
4435                                 found_extent = 1;
4436                                 break;
4437                         }
4438                         if (path->slots[0] - extent_slot > 5)
4439                                 break;
4440                         extent_slot--;
4441                 }
4442 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4443                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4444                 if (found_extent && item_size < sizeof(*ei))
4445                         found_extent = 0;
4446 #endif
4447                 if (!found_extent) {
4448                         BUG_ON(iref);
4449                         ret = remove_extent_backref(trans, extent_root, path,
4450                                                     NULL, refs_to_drop,
4451                                                     is_data);
4452                         BUG_ON(ret);
4453                         btrfs_release_path(extent_root, path);
4454                         path->leave_spinning = 1;
4455
4456                         key.objectid = bytenr;
4457                         key.type = BTRFS_EXTENT_ITEM_KEY;
4458                         key.offset = num_bytes;
4459
4460                         ret = btrfs_search_slot(trans, extent_root,
4461                                                 &key, path, -1, 1);
4462                         if (ret) {
4463                                 printk(KERN_ERR "umm, got %d back from search"
4464                                        ", was looking for %llu\n", ret,
4465                                        (unsigned long long)bytenr);
4466                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4467                         }
4468                         BUG_ON(ret);
4469                         extent_slot = path->slots[0];
4470                 }
4471         } else {
4472                 btrfs_print_leaf(extent_root, path->nodes[0]);
4473                 WARN_ON(1);
4474                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4475                        "parent %llu root %llu  owner %llu offset %llu\n",
4476                        (unsigned long long)bytenr,
4477                        (unsigned long long)parent,
4478                        (unsigned long long)root_objectid,
4479                        (unsigned long long)owner_objectid,
4480                        (unsigned long long)owner_offset);
4481         }
4482
4483         leaf = path->nodes[0];
4484         item_size = btrfs_item_size_nr(leaf, extent_slot);
4485 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4486         if (item_size < sizeof(*ei)) {
4487                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4488                 ret = convert_extent_item_v0(trans, extent_root, path,
4489                                              owner_objectid, 0);
4490                 BUG_ON(ret < 0);
4491
4492                 btrfs_release_path(extent_root, path);
4493                 path->leave_spinning = 1;
4494
4495                 key.objectid = bytenr;
4496                 key.type = BTRFS_EXTENT_ITEM_KEY;
4497                 key.offset = num_bytes;
4498
4499                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4500                                         -1, 1);
4501                 if (ret) {
4502                         printk(KERN_ERR "umm, got %d back from search"
4503                                ", was looking for %llu\n", ret,
4504                                (unsigned long long)bytenr);
4505                         btrfs_print_leaf(extent_root, path->nodes[0]);
4506                 }
4507                 BUG_ON(ret);
4508                 extent_slot = path->slots[0];
4509                 leaf = path->nodes[0];
4510                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4511         }
4512 #endif
4513         BUG_ON(item_size < sizeof(*ei));
4514         ei = btrfs_item_ptr(leaf, extent_slot,
4515                             struct btrfs_extent_item);
4516         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4517                 struct btrfs_tree_block_info *bi;
4518                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4519                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4520                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4521         }
4522
4523         refs = btrfs_extent_refs(leaf, ei);
4524         BUG_ON(refs < refs_to_drop);
4525         refs -= refs_to_drop;
4526
4527         if (refs > 0) {
4528                 if (extent_op)
4529                         __run_delayed_extent_op(extent_op, leaf, ei);
4530                 /*
4531                  * In the case of inline back ref, reference count will
4532                  * be updated by remove_extent_backref
4533                  */
4534                 if (iref) {
4535                         BUG_ON(!found_extent);
4536                 } else {
4537                         btrfs_set_extent_refs(leaf, ei, refs);
4538                         btrfs_mark_buffer_dirty(leaf);
4539                 }
4540                 if (found_extent) {
4541                         ret = remove_extent_backref(trans, extent_root, path,
4542                                                     iref, refs_to_drop,
4543                                                     is_data);
4544                         BUG_ON(ret);
4545                 }
4546         } else {
4547                 if (found_extent) {
4548                         BUG_ON(is_data && refs_to_drop !=
4549                                extent_data_ref_count(root, path, iref));
4550                         if (iref) {
4551                                 BUG_ON(path->slots[0] != extent_slot);
4552                         } else {
4553                                 BUG_ON(path->slots[0] != extent_slot + 1);
4554                                 path->slots[0] = extent_slot;
4555                                 num_to_del = 2;
4556                         }
4557                 }
4558
4559                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4560                                       num_to_del);
4561                 BUG_ON(ret);
4562                 btrfs_release_path(extent_root, path);
4563
4564                 if (is_data) {
4565                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4566                         BUG_ON(ret);
4567                 } else {
4568                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4569                              bytenr >> PAGE_CACHE_SHIFT,
4570                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4571                 }
4572
4573                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4574                 BUG_ON(ret);
4575         }
4576         btrfs_free_path(path);
4577         return ret;
4578 }
4579
4580 /*
4581  * when we free an block, it is possible (and likely) that we free the last
4582  * delayed ref for that extent as well.  This searches the delayed ref tree for
4583  * a given extent, and if there are no other delayed refs to be processed, it
4584  * removes it from the tree.
4585  */
4586 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4587                                       struct btrfs_root *root, u64 bytenr)
4588 {
4589         struct btrfs_delayed_ref_head *head;
4590         struct btrfs_delayed_ref_root *delayed_refs;
4591         struct btrfs_delayed_ref_node *ref;
4592         struct rb_node *node;
4593         int ret = 0;
4594
4595         delayed_refs = &trans->transaction->delayed_refs;
4596         spin_lock(&delayed_refs->lock);
4597         head = btrfs_find_delayed_ref_head(trans, bytenr);
4598         if (!head)
4599                 goto out;
4600
4601         node = rb_prev(&head->node.rb_node);
4602         if (!node)
4603                 goto out;
4604
4605         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4606
4607         /* there are still entries for this ref, we can't drop it */
4608         if (ref->bytenr == bytenr)
4609                 goto out;
4610
4611         if (head->extent_op) {
4612                 if (!head->must_insert_reserved)
4613                         goto out;
4614                 kfree(head->extent_op);
4615                 head->extent_op = NULL;
4616         }
4617
4618         /*
4619          * waiting for the lock here would deadlock.  If someone else has it
4620          * locked they are already in the process of dropping it anyway
4621          */
4622         if (!mutex_trylock(&head->mutex))
4623                 goto out;
4624
4625         /*
4626          * at this point we have a head with no other entries.  Go
4627          * ahead and process it.
4628          */
4629         head->node.in_tree = 0;
4630         rb_erase(&head->node.rb_node, &delayed_refs->root);
4631
4632         delayed_refs->num_entries--;
4633
4634         /*
4635          * we don't take a ref on the node because we're removing it from the
4636          * tree, so we just steal the ref the tree was holding.
4637          */
4638         delayed_refs->num_heads--;
4639         if (list_empty(&head->cluster))
4640                 delayed_refs->num_heads_ready--;
4641
4642         list_del_init(&head->cluster);
4643         spin_unlock(&delayed_refs->lock);
4644
4645         BUG_ON(head->extent_op);
4646         if (head->must_insert_reserved)
4647                 ret = 1;
4648
4649         mutex_unlock(&head->mutex);
4650         btrfs_put_delayed_ref(&head->node);
4651         return ret;
4652 out:
4653         spin_unlock(&delayed_refs->lock);
4654         return 0;
4655 }
4656
4657 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4658                            struct btrfs_root *root,
4659                            struct extent_buffer *buf,
4660                            u64 parent, int last_ref)
4661 {
4662         struct btrfs_block_rsv *block_rsv;
4663         struct btrfs_block_group_cache *cache = NULL;
4664         int ret;
4665
4666         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4667                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4668                                                 parent, root->root_key.objectid,
4669                                                 btrfs_header_level(buf),
4670                                                 BTRFS_DROP_DELAYED_REF, NULL);
4671                 BUG_ON(ret);
4672         }
4673
4674         if (!last_ref)
4675                 return;
4676
4677         block_rsv = get_block_rsv(trans, root);
4678         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4679         if (block_rsv->space_info != cache->space_info)
4680                 goto out;
4681
4682         if (btrfs_header_generation(buf) == trans->transid) {
4683                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4684                         ret = check_ref_cleanup(trans, root, buf->start);
4685                         if (!ret)
4686                                 goto pin;
4687                 }
4688
4689                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4690                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4691                         goto pin;
4692                 }
4693
4694                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4695
4696                 btrfs_add_free_space(cache, buf->start, buf->len);
4697                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4698                 if (ret == -EAGAIN) {
4699                         /* block group became read-only */
4700                         update_reserved_bytes(cache, buf->len, 0, 1);
4701                         goto out;
4702                 }
4703
4704                 ret = 1;
4705                 spin_lock(&block_rsv->lock);
4706                 if (block_rsv->reserved < block_rsv->size) {
4707                         block_rsv->reserved += buf->len;
4708                         ret = 0;
4709                 }
4710                 spin_unlock(&block_rsv->lock);
4711
4712                 if (ret) {
4713                         spin_lock(&cache->space_info->lock);
4714                         cache->space_info->bytes_reserved -= buf->len;
4715                         spin_unlock(&cache->space_info->lock);
4716                 }
4717                 goto out;
4718         }
4719 pin:
4720         if (block_rsv->durable && !cache->ro) {
4721                 ret = 0;
4722                 spin_lock(&cache->lock);
4723                 if (!cache->ro) {
4724                         cache->reserved_pinned += buf->len;
4725                         ret = 1;
4726                 }
4727                 spin_unlock(&cache->lock);
4728
4729                 if (ret) {
4730                         spin_lock(&block_rsv->lock);
4731                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4732                         spin_unlock(&block_rsv->lock);
4733                 }
4734         }
4735 out:
4736         btrfs_put_block_group(cache);
4737 }
4738
4739 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4740                       struct btrfs_root *root,
4741                       u64 bytenr, u64 num_bytes, u64 parent,
4742                       u64 root_objectid, u64 owner, u64 offset)
4743 {
4744         int ret;
4745
4746         /*
4747          * tree log blocks never actually go into the extent allocation
4748          * tree, just update pinning info and exit early.
4749          */
4750         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4751                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4752                 /* unlocks the pinned mutex */
4753                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4754                 ret = 0;
4755         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4756                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4757                                         parent, root_objectid, (int)owner,
4758                                         BTRFS_DROP_DELAYED_REF, NULL);
4759                 BUG_ON(ret);
4760         } else {
4761                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4762                                         parent, root_objectid, owner,
4763                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4764                 BUG_ON(ret);
4765         }
4766         return ret;
4767 }
4768
4769 static u64 stripe_align(struct btrfs_root *root, u64 val)
4770 {
4771         u64 mask = ((u64)root->stripesize - 1);
4772         u64 ret = (val + mask) & ~mask;
4773         return ret;
4774 }
4775
4776 /*
4777  * when we wait for progress in the block group caching, its because
4778  * our allocation attempt failed at least once.  So, we must sleep
4779  * and let some progress happen before we try again.
4780  *
4781  * This function will sleep at least once waiting for new free space to
4782  * show up, and then it will check the block group free space numbers
4783  * for our min num_bytes.  Another option is to have it go ahead
4784  * and look in the rbtree for a free extent of a given size, but this
4785  * is a good start.
4786  */
4787 static noinline int
4788 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4789                                 u64 num_bytes)
4790 {
4791         struct btrfs_caching_control *caching_ctl;
4792         DEFINE_WAIT(wait);
4793
4794         caching_ctl = get_caching_control(cache);
4795         if (!caching_ctl)
4796                 return 0;
4797
4798         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4799                    (cache->free_space >= num_bytes));
4800
4801         put_caching_control(caching_ctl);
4802         return 0;
4803 }
4804
4805 static noinline int
4806 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4807 {
4808         struct btrfs_caching_control *caching_ctl;
4809         DEFINE_WAIT(wait);
4810
4811         caching_ctl = get_caching_control(cache);
4812         if (!caching_ctl)
4813                 return 0;
4814
4815         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4816
4817         put_caching_control(caching_ctl);
4818         return 0;
4819 }
4820
4821 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4822 {
4823         int index;
4824         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4825                 index = 0;
4826         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4827                 index = 1;
4828         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4829                 index = 2;
4830         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4831                 index = 3;
4832         else
4833                 index = 4;
4834         return index;
4835 }
4836
4837 enum btrfs_loop_type {
4838         LOOP_FIND_IDEAL = 0,
4839         LOOP_CACHING_NOWAIT = 1,
4840         LOOP_CACHING_WAIT = 2,
4841         LOOP_ALLOC_CHUNK = 3,
4842         LOOP_NO_EMPTY_SIZE = 4,
4843 };
4844
4845 /*
4846  * walks the btree of allocated extents and find a hole of a given size.
4847  * The key ins is changed to record the hole:
4848  * ins->objectid == block start
4849  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4850  * ins->offset == number of blocks
4851  * Any available blocks before search_start are skipped.
4852  */
4853 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4854                                      struct btrfs_root *orig_root,
4855                                      u64 num_bytes, u64 empty_size,
4856                                      u64 search_start, u64 search_end,
4857                                      u64 hint_byte, struct btrfs_key *ins,
4858                                      int data)
4859 {
4860         int ret = 0;
4861         struct btrfs_root *root = orig_root->fs_info->extent_root;
4862         struct btrfs_free_cluster *last_ptr = NULL;
4863         struct btrfs_block_group_cache *block_group = NULL;
4864         int empty_cluster = 2 * 1024 * 1024;
4865         int allowed_chunk_alloc = 0;
4866         int done_chunk_alloc = 0;
4867         struct btrfs_space_info *space_info;
4868         int last_ptr_loop = 0;
4869         int loop = 0;
4870         int index = 0;
4871         bool found_uncached_bg = false;
4872         bool failed_cluster_refill = false;
4873         bool failed_alloc = false;
4874         bool use_cluster = true;
4875         u64 ideal_cache_percent = 0;
4876         u64 ideal_cache_offset = 0;
4877
4878         WARN_ON(num_bytes < root->sectorsize);
4879         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4880         ins->objectid = 0;
4881         ins->offset = 0;
4882
4883         space_info = __find_space_info(root->fs_info, data);
4884         if (!space_info) {
4885                 printk(KERN_ERR "No space info for %d\n", data);
4886                 return -ENOSPC;
4887         }
4888
4889         /*
4890          * If the space info is for both data and metadata it means we have a
4891          * small filesystem and we can't use the clustering stuff.
4892          */
4893         if (btrfs_mixed_space_info(space_info))
4894                 use_cluster = false;
4895
4896         if (orig_root->ref_cows || empty_size)
4897                 allowed_chunk_alloc = 1;
4898
4899         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4900                 last_ptr = &root->fs_info->meta_alloc_cluster;
4901                 if (!btrfs_test_opt(root, SSD))
4902                         empty_cluster = 64 * 1024;
4903         }
4904
4905         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4906             btrfs_test_opt(root, SSD)) {
4907                 last_ptr = &root->fs_info->data_alloc_cluster;
4908         }
4909
4910         if (last_ptr) {
4911                 spin_lock(&last_ptr->lock);
4912                 if (last_ptr->block_group)
4913                         hint_byte = last_ptr->window_start;
4914                 spin_unlock(&last_ptr->lock);
4915         }
4916
4917         search_start = max(search_start, first_logical_byte(root, 0));
4918         search_start = max(search_start, hint_byte);
4919
4920         if (!last_ptr)
4921                 empty_cluster = 0;
4922
4923         if (search_start == hint_byte) {
4924 ideal_cache:
4925                 block_group = btrfs_lookup_block_group(root->fs_info,
4926                                                        search_start);
4927                 /*
4928                  * we don't want to use the block group if it doesn't match our
4929                  * allocation bits, or if its not cached.
4930                  *
4931                  * However if we are re-searching with an ideal block group
4932                  * picked out then we don't care that the block group is cached.
4933                  */
4934                 if (block_group && block_group_bits(block_group, data) &&
4935                     (block_group->cached != BTRFS_CACHE_NO ||
4936                      search_start == ideal_cache_offset)) {
4937                         down_read(&space_info->groups_sem);
4938                         if (list_empty(&block_group->list) ||
4939                             block_group->ro) {
4940                                 /*
4941                                  * someone is removing this block group,
4942                                  * we can't jump into the have_block_group
4943                                  * target because our list pointers are not
4944                                  * valid
4945                                  */
4946                                 btrfs_put_block_group(block_group);
4947                                 up_read(&space_info->groups_sem);
4948                         } else {
4949                                 index = get_block_group_index(block_group);
4950                                 goto have_block_group;
4951                         }
4952                 } else if (block_group) {
4953                         btrfs_put_block_group(block_group);
4954                 }
4955         }
4956 search:
4957         down_read(&space_info->groups_sem);
4958         list_for_each_entry(block_group, &space_info->block_groups[index],
4959                             list) {
4960                 u64 offset;
4961                 int cached;
4962
4963                 btrfs_get_block_group(block_group);
4964                 search_start = block_group->key.objectid;
4965
4966                 /*
4967                  * this can happen if we end up cycling through all the
4968                  * raid types, but we want to make sure we only allocate
4969                  * for the proper type.
4970                  */
4971                 if (!block_group_bits(block_group, data)) {
4972                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
4973                                 BTRFS_BLOCK_GROUP_RAID1 |
4974                                 BTRFS_BLOCK_GROUP_RAID10;
4975
4976                         /*
4977                          * if they asked for extra copies and this block group
4978                          * doesn't provide them, bail.  This does allow us to
4979                          * fill raid0 from raid1.
4980                          */
4981                         if ((data & extra) && !(block_group->flags & extra))
4982                                 goto loop;
4983                 }
4984
4985 have_block_group:
4986                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4987                         u64 free_percent;
4988
4989                         ret = cache_block_group(block_group, trans,
4990                                                 orig_root, 1);
4991                         if (block_group->cached == BTRFS_CACHE_FINISHED)
4992                                 goto have_block_group;
4993
4994                         free_percent = btrfs_block_group_used(&block_group->item);
4995                         free_percent *= 100;
4996                         free_percent = div64_u64(free_percent,
4997                                                  block_group->key.offset);
4998                         free_percent = 100 - free_percent;
4999                         if (free_percent > ideal_cache_percent &&
5000                             likely(!block_group->ro)) {
5001                                 ideal_cache_offset = block_group->key.objectid;
5002                                 ideal_cache_percent = free_percent;
5003                         }
5004
5005                         /*
5006                          * We only want to start kthread caching if we are at
5007                          * the point where we will wait for caching to make
5008                          * progress, or if our ideal search is over and we've
5009                          * found somebody to start caching.
5010                          */
5011                         if (loop > LOOP_CACHING_NOWAIT ||
5012                             (loop > LOOP_FIND_IDEAL &&
5013                              atomic_read(&space_info->caching_threads) < 2)) {
5014                                 ret = cache_block_group(block_group, trans,
5015                                                         orig_root, 0);
5016                                 BUG_ON(ret);
5017                         }
5018                         found_uncached_bg = true;
5019
5020                         /*
5021                          * If loop is set for cached only, try the next block
5022                          * group.
5023                          */
5024                         if (loop == LOOP_FIND_IDEAL)
5025                                 goto loop;
5026                 }
5027
5028                 cached = block_group_cache_done(block_group);
5029                 if (unlikely(!cached))
5030                         found_uncached_bg = true;
5031
5032                 if (unlikely(block_group->ro))
5033                         goto loop;
5034
5035                 /*
5036                  * Ok we want to try and use the cluster allocator, so lets look
5037                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5038                  * have tried the cluster allocator plenty of times at this
5039                  * point and not have found anything, so we are likely way too
5040                  * fragmented for the clustering stuff to find anything, so lets
5041                  * just skip it and let the allocator find whatever block it can
5042                  * find
5043                  */
5044                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5045                         /*
5046                          * the refill lock keeps out other
5047                          * people trying to start a new cluster
5048                          */
5049                         spin_lock(&last_ptr->refill_lock);
5050                         if (last_ptr->block_group &&
5051                             (last_ptr->block_group->ro ||
5052                             !block_group_bits(last_ptr->block_group, data))) {
5053                                 offset = 0;
5054                                 goto refill_cluster;
5055                         }
5056
5057                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5058                                                  num_bytes, search_start);
5059                         if (offset) {
5060                                 /* we have a block, we're done */
5061                                 spin_unlock(&last_ptr->refill_lock);
5062                                 goto checks;
5063                         }
5064
5065                         spin_lock(&last_ptr->lock);
5066                         /*
5067                          * whoops, this cluster doesn't actually point to
5068                          * this block group.  Get a ref on the block
5069                          * group is does point to and try again
5070                          */
5071                         if (!last_ptr_loop && last_ptr->block_group &&
5072                             last_ptr->block_group != block_group) {
5073
5074                                 btrfs_put_block_group(block_group);
5075                                 block_group = last_ptr->block_group;
5076                                 btrfs_get_block_group(block_group);
5077                                 spin_unlock(&last_ptr->lock);
5078                                 spin_unlock(&last_ptr->refill_lock);
5079
5080                                 last_ptr_loop = 1;
5081                                 search_start = block_group->key.objectid;
5082                                 /*
5083                                  * we know this block group is properly
5084                                  * in the list because
5085                                  * btrfs_remove_block_group, drops the
5086                                  * cluster before it removes the block
5087                                  * group from the list
5088                                  */
5089                                 goto have_block_group;
5090                         }
5091                         spin_unlock(&last_ptr->lock);
5092 refill_cluster:
5093                         /*
5094                          * this cluster didn't work out, free it and
5095                          * start over
5096                          */
5097                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5098
5099                         last_ptr_loop = 0;
5100
5101                         /* allocate a cluster in this block group */
5102                         ret = btrfs_find_space_cluster(trans, root,
5103                                                block_group, last_ptr,
5104                                                offset, num_bytes,
5105                                                empty_cluster + empty_size);
5106                         if (ret == 0) {
5107                                 /*
5108                                  * now pull our allocation out of this
5109                                  * cluster
5110                                  */
5111                                 offset = btrfs_alloc_from_cluster(block_group,
5112                                                   last_ptr, num_bytes,
5113                                                   search_start);
5114                                 if (offset) {
5115                                         /* we found one, proceed */
5116                                         spin_unlock(&last_ptr->refill_lock);
5117                                         goto checks;
5118                                 }
5119                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5120                                    && !failed_cluster_refill) {
5121                                 spin_unlock(&last_ptr->refill_lock);
5122
5123                                 failed_cluster_refill = true;
5124                                 wait_block_group_cache_progress(block_group,
5125                                        num_bytes + empty_cluster + empty_size);
5126                                 goto have_block_group;
5127                         }
5128
5129                         /*
5130                          * at this point we either didn't find a cluster
5131                          * or we weren't able to allocate a block from our
5132                          * cluster.  Free the cluster we've been trying
5133                          * to use, and go to the next block group
5134                          */
5135                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5136                         spin_unlock(&last_ptr->refill_lock);
5137                         goto loop;
5138                 }
5139
5140                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5141                                                     num_bytes, empty_size);
5142                 /*
5143                  * If we didn't find a chunk, and we haven't failed on this
5144                  * block group before, and this block group is in the middle of
5145                  * caching and we are ok with waiting, then go ahead and wait
5146                  * for progress to be made, and set failed_alloc to true.
5147                  *
5148                  * If failed_alloc is true then we've already waited on this
5149                  * block group once and should move on to the next block group.
5150                  */
5151                 if (!offset && !failed_alloc && !cached &&
5152                     loop > LOOP_CACHING_NOWAIT) {
5153                         wait_block_group_cache_progress(block_group,
5154                                                 num_bytes + empty_size);
5155                         failed_alloc = true;
5156                         goto have_block_group;
5157                 } else if (!offset) {
5158                         goto loop;
5159                 }
5160 checks:
5161                 search_start = stripe_align(root, offset);
5162                 /* move on to the next group */
5163                 if (search_start + num_bytes >= search_end) {
5164                         btrfs_add_free_space(block_group, offset, num_bytes);
5165                         goto loop;
5166                 }
5167
5168                 /* move on to the next group */
5169                 if (search_start + num_bytes >
5170                     block_group->key.objectid + block_group->key.offset) {
5171                         btrfs_add_free_space(block_group, offset, num_bytes);
5172                         goto loop;
5173                 }
5174
5175                 ins->objectid = search_start;
5176                 ins->offset = num_bytes;
5177
5178                 if (offset < search_start)
5179                         btrfs_add_free_space(block_group, offset,
5180                                              search_start - offset);
5181                 BUG_ON(offset > search_start);
5182
5183                 ret = update_reserved_bytes(block_group, num_bytes, 1,
5184                                             (data & BTRFS_BLOCK_GROUP_DATA));
5185                 if (ret == -EAGAIN) {
5186                         btrfs_add_free_space(block_group, offset, num_bytes);
5187                         goto loop;
5188                 }
5189
5190                 /* we are all good, lets return */
5191                 ins->objectid = search_start;
5192                 ins->offset = num_bytes;
5193
5194                 if (offset < search_start)
5195                         btrfs_add_free_space(block_group, offset,
5196                                              search_start - offset);
5197                 BUG_ON(offset > search_start);
5198                 break;
5199 loop:
5200                 failed_cluster_refill = false;
5201                 failed_alloc = false;
5202                 BUG_ON(index != get_block_group_index(block_group));
5203                 btrfs_put_block_group(block_group);
5204         }
5205         up_read(&space_info->groups_sem);
5206
5207         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5208                 goto search;
5209
5210         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5211          *                      for them to make caching progress.  Also
5212          *                      determine the best possible bg to cache
5213          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5214          *                      caching kthreads as we move along
5215          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5216          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5217          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5218          *                      again
5219          */
5220         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5221             (found_uncached_bg || empty_size || empty_cluster ||
5222              allowed_chunk_alloc)) {
5223                 index = 0;
5224                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5225                         found_uncached_bg = false;
5226                         loop++;
5227                         if (!ideal_cache_percent &&
5228                             atomic_read(&space_info->caching_threads))
5229                                 goto search;
5230
5231                         /*
5232                          * 1 of the following 2 things have happened so far
5233                          *
5234                          * 1) We found an ideal block group for caching that
5235                          * is mostly full and will cache quickly, so we might
5236                          * as well wait for it.
5237                          *
5238                          * 2) We searched for cached only and we didn't find
5239                          * anything, and we didn't start any caching kthreads
5240                          * either, so chances are we will loop through and
5241                          * start a couple caching kthreads, and then come back
5242                          * around and just wait for them.  This will be slower
5243                          * because we will have 2 caching kthreads reading at
5244                          * the same time when we could have just started one
5245                          * and waited for it to get far enough to give us an
5246                          * allocation, so go ahead and go to the wait caching
5247                          * loop.
5248                          */
5249                         loop = LOOP_CACHING_WAIT;
5250                         search_start = ideal_cache_offset;
5251                         ideal_cache_percent = 0;
5252                         goto ideal_cache;
5253                 } else if (loop == LOOP_FIND_IDEAL) {
5254                         /*
5255                          * Didn't find a uncached bg, wait on anything we find
5256                          * next.
5257                          */
5258                         loop = LOOP_CACHING_WAIT;
5259                         goto search;
5260                 }
5261
5262                 if (loop < LOOP_CACHING_WAIT) {
5263                         loop++;
5264                         goto search;
5265                 }
5266
5267                 if (loop == LOOP_ALLOC_CHUNK) {
5268                         empty_size = 0;
5269                         empty_cluster = 0;
5270                 }
5271
5272                 if (allowed_chunk_alloc) {
5273                         ret = do_chunk_alloc(trans, root, num_bytes +
5274                                              2 * 1024 * 1024, data, 1);
5275                         allowed_chunk_alloc = 0;
5276                         done_chunk_alloc = 1;
5277                 } else if (!done_chunk_alloc) {
5278                         space_info->force_alloc = 1;
5279                 }
5280
5281                 if (loop < LOOP_NO_EMPTY_SIZE) {
5282                         loop++;
5283                         goto search;
5284                 }
5285                 ret = -ENOSPC;
5286         } else if (!ins->objectid) {
5287                 ret = -ENOSPC;
5288         }
5289
5290         /* we found what we needed */
5291         if (ins->objectid) {
5292                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5293                         trans->block_group = block_group->key.objectid;
5294
5295                 btrfs_put_block_group(block_group);
5296                 ret = 0;
5297         }
5298
5299         return ret;
5300 }
5301
5302 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5303                             int dump_block_groups)
5304 {
5305         struct btrfs_block_group_cache *cache;
5306         int index = 0;
5307
5308         spin_lock(&info->lock);
5309         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5310                (unsigned long long)(info->total_bytes - info->bytes_used -
5311                                     info->bytes_pinned - info->bytes_reserved -
5312                                     info->bytes_readonly),
5313                (info->full) ? "" : "not ");
5314         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5315                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5316                (unsigned long long)info->total_bytes,
5317                (unsigned long long)info->bytes_used,
5318                (unsigned long long)info->bytes_pinned,
5319                (unsigned long long)info->bytes_reserved,
5320                (unsigned long long)info->bytes_may_use,
5321                (unsigned long long)info->bytes_readonly);
5322         spin_unlock(&info->lock);
5323
5324         if (!dump_block_groups)
5325                 return;
5326
5327         down_read(&info->groups_sem);
5328 again:
5329         list_for_each_entry(cache, &info->block_groups[index], list) {
5330                 spin_lock(&cache->lock);
5331                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5332                        "%llu pinned %llu reserved\n",
5333                        (unsigned long long)cache->key.objectid,
5334                        (unsigned long long)cache->key.offset,
5335                        (unsigned long long)btrfs_block_group_used(&cache->item),
5336                        (unsigned long long)cache->pinned,
5337                        (unsigned long long)cache->reserved);
5338                 btrfs_dump_free_space(cache, bytes);
5339                 spin_unlock(&cache->lock);
5340         }
5341         if (++index < BTRFS_NR_RAID_TYPES)
5342                 goto again;
5343         up_read(&info->groups_sem);
5344 }
5345
5346 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5347                          struct btrfs_root *root,
5348                          u64 num_bytes, u64 min_alloc_size,
5349                          u64 empty_size, u64 hint_byte,
5350                          u64 search_end, struct btrfs_key *ins,
5351                          u64 data)
5352 {
5353         int ret;
5354         u64 search_start = 0;
5355
5356         data = btrfs_get_alloc_profile(root, data);
5357 again:
5358         /*
5359          * the only place that sets empty_size is btrfs_realloc_node, which
5360          * is not called recursively on allocations
5361          */
5362         if (empty_size || root->ref_cows)
5363                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5364                                      num_bytes + 2 * 1024 * 1024, data, 0);
5365
5366         WARN_ON(num_bytes < root->sectorsize);
5367         ret = find_free_extent(trans, root, num_bytes, empty_size,
5368                                search_start, search_end, hint_byte,
5369                                ins, data);
5370
5371         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5372                 num_bytes = num_bytes >> 1;
5373                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5374                 num_bytes = max(num_bytes, min_alloc_size);
5375                 do_chunk_alloc(trans, root->fs_info->extent_root,
5376                                num_bytes, data, 1);
5377                 goto again;
5378         }
5379         if (ret == -ENOSPC) {
5380                 struct btrfs_space_info *sinfo;
5381
5382                 sinfo = __find_space_info(root->fs_info, data);
5383                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5384                        "wanted %llu\n", (unsigned long long)data,
5385                        (unsigned long long)num_bytes);
5386                 dump_space_info(sinfo, num_bytes, 1);
5387         }
5388
5389         return ret;
5390 }
5391
5392 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5393 {
5394         struct btrfs_block_group_cache *cache;
5395         int ret = 0;
5396
5397         cache = btrfs_lookup_block_group(root->fs_info, start);
5398         if (!cache) {
5399                 printk(KERN_ERR "Unable to find block group for %llu\n",
5400                        (unsigned long long)start);
5401                 return -ENOSPC;
5402         }
5403
5404         ret = btrfs_discard_extent(root, start, len);
5405
5406         btrfs_add_free_space(cache, start, len);
5407         update_reserved_bytes(cache, len, 0, 1);
5408         btrfs_put_block_group(cache);
5409
5410         return ret;
5411 }
5412
5413 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5414                                       struct btrfs_root *root,
5415                                       u64 parent, u64 root_objectid,
5416                                       u64 flags, u64 owner, u64 offset,
5417                                       struct btrfs_key *ins, int ref_mod)
5418 {
5419         int ret;
5420         struct btrfs_fs_info *fs_info = root->fs_info;
5421         struct btrfs_extent_item *extent_item;
5422         struct btrfs_extent_inline_ref *iref;
5423         struct btrfs_path *path;
5424         struct extent_buffer *leaf;
5425         int type;
5426         u32 size;
5427
5428         if (parent > 0)
5429                 type = BTRFS_SHARED_DATA_REF_KEY;
5430         else
5431                 type = BTRFS_EXTENT_DATA_REF_KEY;
5432
5433         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5434
5435         path = btrfs_alloc_path();
5436         BUG_ON(!path);
5437
5438         path->leave_spinning = 1;
5439         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5440                                       ins, size);
5441         BUG_ON(ret);
5442
5443         leaf = path->nodes[0];
5444         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5445                                      struct btrfs_extent_item);
5446         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5447         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5448         btrfs_set_extent_flags(leaf, extent_item,
5449                                flags | BTRFS_EXTENT_FLAG_DATA);
5450
5451         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5452         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5453         if (parent > 0) {
5454                 struct btrfs_shared_data_ref *ref;
5455                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5456                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5457                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5458         } else {
5459                 struct btrfs_extent_data_ref *ref;
5460                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5461                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5462                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5463                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5464                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5465         }
5466
5467         btrfs_mark_buffer_dirty(path->nodes[0]);
5468         btrfs_free_path(path);
5469
5470         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5471         if (ret) {
5472                 printk(KERN_ERR "btrfs update block group failed for %llu "
5473                        "%llu\n", (unsigned long long)ins->objectid,
5474                        (unsigned long long)ins->offset);
5475                 BUG();
5476         }
5477         return ret;
5478 }
5479
5480 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5481                                      struct btrfs_root *root,
5482                                      u64 parent, u64 root_objectid,
5483                                      u64 flags, struct btrfs_disk_key *key,
5484                                      int level, struct btrfs_key *ins)
5485 {
5486         int ret;
5487         struct btrfs_fs_info *fs_info = root->fs_info;
5488         struct btrfs_extent_item *extent_item;
5489         struct btrfs_tree_block_info *block_info;
5490         struct btrfs_extent_inline_ref *iref;
5491         struct btrfs_path *path;
5492         struct extent_buffer *leaf;
5493         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5494
5495         path = btrfs_alloc_path();
5496         BUG_ON(!path);
5497
5498         path->leave_spinning = 1;
5499         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5500                                       ins, size);
5501         BUG_ON(ret);
5502
5503         leaf = path->nodes[0];
5504         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5505                                      struct btrfs_extent_item);
5506         btrfs_set_extent_refs(leaf, extent_item, 1);
5507         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5508         btrfs_set_extent_flags(leaf, extent_item,
5509                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5510         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5511
5512         btrfs_set_tree_block_key(leaf, block_info, key);
5513         btrfs_set_tree_block_level(leaf, block_info, level);
5514
5515         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5516         if (parent > 0) {
5517                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5518                 btrfs_set_extent_inline_ref_type(leaf, iref,
5519                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5520                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5521         } else {
5522                 btrfs_set_extent_inline_ref_type(leaf, iref,
5523                                                  BTRFS_TREE_BLOCK_REF_KEY);
5524                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5525         }
5526
5527         btrfs_mark_buffer_dirty(leaf);
5528         btrfs_free_path(path);
5529
5530         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5531         if (ret) {
5532                 printk(KERN_ERR "btrfs update block group failed for %llu "
5533                        "%llu\n", (unsigned long long)ins->objectid,
5534                        (unsigned long long)ins->offset);
5535                 BUG();
5536         }
5537         return ret;
5538 }
5539
5540 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5541                                      struct btrfs_root *root,
5542                                      u64 root_objectid, u64 owner,
5543                                      u64 offset, struct btrfs_key *ins)
5544 {
5545         int ret;
5546
5547         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5548
5549         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5550                                          0, root_objectid, owner, offset,
5551                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5552         return ret;
5553 }
5554
5555 /*
5556  * this is used by the tree logging recovery code.  It records that
5557  * an extent has been allocated and makes sure to clear the free
5558  * space cache bits as well
5559  */
5560 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5561                                    struct btrfs_root *root,
5562                                    u64 root_objectid, u64 owner, u64 offset,
5563                                    struct btrfs_key *ins)
5564 {
5565         int ret;
5566         struct btrfs_block_group_cache *block_group;
5567         struct btrfs_caching_control *caching_ctl;
5568         u64 start = ins->objectid;
5569         u64 num_bytes = ins->offset;
5570
5571         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5572         cache_block_group(block_group, trans, NULL, 0);
5573         caching_ctl = get_caching_control(block_group);
5574
5575         if (!caching_ctl) {
5576                 BUG_ON(!block_group_cache_done(block_group));
5577                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5578                 BUG_ON(ret);
5579         } else {
5580                 mutex_lock(&caching_ctl->mutex);
5581
5582                 if (start >= caching_ctl->progress) {
5583                         ret = add_excluded_extent(root, start, num_bytes);
5584                         BUG_ON(ret);
5585                 } else if (start + num_bytes <= caching_ctl->progress) {
5586                         ret = btrfs_remove_free_space(block_group,
5587                                                       start, num_bytes);
5588                         BUG_ON(ret);
5589                 } else {
5590                         num_bytes = caching_ctl->progress - start;
5591                         ret = btrfs_remove_free_space(block_group,
5592                                                       start, num_bytes);
5593                         BUG_ON(ret);
5594
5595                         start = caching_ctl->progress;
5596                         num_bytes = ins->objectid + ins->offset -
5597                                     caching_ctl->progress;
5598                         ret = add_excluded_extent(root, start, num_bytes);
5599                         BUG_ON(ret);
5600                 }
5601
5602                 mutex_unlock(&caching_ctl->mutex);
5603                 put_caching_control(caching_ctl);
5604         }
5605
5606         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5607         BUG_ON(ret);
5608         btrfs_put_block_group(block_group);
5609         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5610                                          0, owner, offset, ins, 1);
5611         return ret;
5612 }
5613
5614 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5615                                             struct btrfs_root *root,
5616                                             u64 bytenr, u32 blocksize,
5617                                             int level)
5618 {
5619         struct extent_buffer *buf;
5620
5621         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5622         if (!buf)
5623                 return ERR_PTR(-ENOMEM);
5624         btrfs_set_header_generation(buf, trans->transid);
5625         btrfs_set_buffer_lockdep_class(buf, level);
5626         btrfs_tree_lock(buf);
5627         clean_tree_block(trans, root, buf);
5628
5629         btrfs_set_lock_blocking(buf);
5630         btrfs_set_buffer_uptodate(buf);
5631
5632         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5633                 /*
5634                  * we allow two log transactions at a time, use different
5635                  * EXENT bit to differentiate dirty pages.
5636                  */
5637                 if (root->log_transid % 2 == 0)
5638                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5639                                         buf->start + buf->len - 1, GFP_NOFS);
5640                 else
5641                         set_extent_new(&root->dirty_log_pages, buf->start,
5642                                         buf->start + buf->len - 1, GFP_NOFS);
5643         } else {
5644                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5645                          buf->start + buf->len - 1, GFP_NOFS);
5646         }
5647         trans->blocks_used++;
5648         /* this returns a buffer locked for blocking */
5649         return buf;
5650 }
5651
5652 static struct btrfs_block_rsv *
5653 use_block_rsv(struct btrfs_trans_handle *trans,
5654               struct btrfs_root *root, u32 blocksize)
5655 {
5656         struct btrfs_block_rsv *block_rsv;
5657         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5658         int ret;
5659
5660         block_rsv = get_block_rsv(trans, root);
5661
5662         if (block_rsv->size == 0) {
5663                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5664                                              blocksize, 0);
5665                 /*
5666                  * If we couldn't reserve metadata bytes try and use some from
5667                  * the global reserve.
5668                  */
5669                 if (ret && block_rsv != global_rsv) {
5670                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5671                         if (!ret)
5672                                 return global_rsv;
5673                         return ERR_PTR(ret);
5674                 } else if (ret) {
5675                         return ERR_PTR(ret);
5676                 }
5677                 return block_rsv;
5678         }
5679
5680         ret = block_rsv_use_bytes(block_rsv, blocksize);
5681         if (!ret)
5682                 return block_rsv;
5683         if (ret) {
5684                 WARN_ON(1);
5685                 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5686                                              0);
5687                 if (!ret) {
5688                         spin_lock(&block_rsv->lock);
5689                         block_rsv->size += blocksize;
5690                         spin_unlock(&block_rsv->lock);
5691                         return block_rsv;
5692                 } else if (ret && block_rsv != global_rsv) {
5693                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5694                         if (!ret)
5695                                 return global_rsv;
5696                 }
5697         }
5698
5699         return ERR_PTR(-ENOSPC);
5700 }
5701
5702 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5703 {
5704         block_rsv_add_bytes(block_rsv, blocksize, 0);
5705         block_rsv_release_bytes(block_rsv, NULL, 0);
5706 }
5707
5708 /*
5709  * finds a free extent and does all the dirty work required for allocation
5710  * returns the key for the extent through ins, and a tree buffer for
5711  * the first block of the extent through buf.
5712  *
5713  * returns the tree buffer or NULL.
5714  */
5715 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5716                                         struct btrfs_root *root, u32 blocksize,
5717                                         u64 parent, u64 root_objectid,
5718                                         struct btrfs_disk_key *key, int level,
5719                                         u64 hint, u64 empty_size)
5720 {
5721         struct btrfs_key ins;
5722         struct btrfs_block_rsv *block_rsv;
5723         struct extent_buffer *buf;
5724         u64 flags = 0;
5725         int ret;
5726
5727
5728         block_rsv = use_block_rsv(trans, root, blocksize);
5729         if (IS_ERR(block_rsv))
5730                 return ERR_CAST(block_rsv);
5731
5732         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5733                                    empty_size, hint, (u64)-1, &ins, 0);
5734         if (ret) {
5735                 unuse_block_rsv(block_rsv, blocksize);
5736                 return ERR_PTR(ret);
5737         }
5738
5739         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5740                                     blocksize, level);
5741         BUG_ON(IS_ERR(buf));
5742
5743         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5744                 if (parent == 0)
5745                         parent = ins.objectid;
5746                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5747         } else
5748                 BUG_ON(parent > 0);
5749
5750         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5751                 struct btrfs_delayed_extent_op *extent_op;
5752                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5753                 BUG_ON(!extent_op);
5754                 if (key)
5755                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5756                 else
5757                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5758                 extent_op->flags_to_set = flags;
5759                 extent_op->update_key = 1;
5760                 extent_op->update_flags = 1;
5761                 extent_op->is_data = 0;
5762
5763                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5764                                         ins.offset, parent, root_objectid,
5765                                         level, BTRFS_ADD_DELAYED_EXTENT,
5766                                         extent_op);
5767                 BUG_ON(ret);
5768         }
5769         return buf;
5770 }
5771
5772 struct walk_control {
5773         u64 refs[BTRFS_MAX_LEVEL];
5774         u64 flags[BTRFS_MAX_LEVEL];
5775         struct btrfs_key update_progress;
5776         int stage;
5777         int level;
5778         int shared_level;
5779         int update_ref;
5780         int keep_locks;
5781         int reada_slot;
5782         int reada_count;
5783 };
5784
5785 #define DROP_REFERENCE  1
5786 #define UPDATE_BACKREF  2
5787
5788 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5789                                      struct btrfs_root *root,
5790                                      struct walk_control *wc,
5791                                      struct btrfs_path *path)
5792 {
5793         u64 bytenr;
5794         u64 generation;
5795         u64 refs;
5796         u64 flags;
5797         u32 nritems;
5798         u32 blocksize;
5799         struct btrfs_key key;
5800         struct extent_buffer *eb;
5801         int ret;
5802         int slot;
5803         int nread = 0;
5804
5805         if (path->slots[wc->level] < wc->reada_slot) {
5806                 wc->reada_count = wc->reada_count * 2 / 3;
5807                 wc->reada_count = max(wc->reada_count, 2);
5808         } else {
5809                 wc->reada_count = wc->reada_count * 3 / 2;
5810                 wc->reada_count = min_t(int, wc->reada_count,
5811                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5812         }
5813
5814         eb = path->nodes[wc->level];
5815         nritems = btrfs_header_nritems(eb);
5816         blocksize = btrfs_level_size(root, wc->level - 1);
5817
5818         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5819                 if (nread >= wc->reada_count)
5820                         break;
5821
5822                 cond_resched();
5823                 bytenr = btrfs_node_blockptr(eb, slot);
5824                 generation = btrfs_node_ptr_generation(eb, slot);
5825
5826                 if (slot == path->slots[wc->level])
5827                         goto reada;
5828
5829                 if (wc->stage == UPDATE_BACKREF &&
5830                     generation <= root->root_key.offset)
5831                         continue;
5832
5833                 /* We don't lock the tree block, it's OK to be racy here */
5834                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5835                                                &refs, &flags);
5836                 BUG_ON(ret);
5837                 BUG_ON(refs == 0);
5838
5839                 if (wc->stage == DROP_REFERENCE) {
5840                         if (refs == 1)
5841                                 goto reada;
5842
5843                         if (wc->level == 1 &&
5844                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5845                                 continue;
5846                         if (!wc->update_ref ||
5847                             generation <= root->root_key.offset)
5848                                 continue;
5849                         btrfs_node_key_to_cpu(eb, &key, slot);
5850                         ret = btrfs_comp_cpu_keys(&key,
5851                                                   &wc->update_progress);
5852                         if (ret < 0)
5853                                 continue;
5854                 } else {
5855                         if (wc->level == 1 &&
5856                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5857                                 continue;
5858                 }
5859 reada:
5860                 ret = readahead_tree_block(root, bytenr, blocksize,
5861                                            generation);
5862                 if (ret)
5863                         break;
5864                 nread++;
5865         }
5866         wc->reada_slot = slot;
5867 }
5868
5869 /*
5870  * hepler to process tree block while walking down the tree.
5871  *
5872  * when wc->stage == UPDATE_BACKREF, this function updates
5873  * back refs for pointers in the block.
5874  *
5875  * NOTE: return value 1 means we should stop walking down.
5876  */
5877 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5878                                    struct btrfs_root *root,
5879                                    struct btrfs_path *path,
5880                                    struct walk_control *wc, int lookup_info)
5881 {
5882         int level = wc->level;
5883         struct extent_buffer *eb = path->nodes[level];
5884         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5885         int ret;
5886
5887         if (wc->stage == UPDATE_BACKREF &&
5888             btrfs_header_owner(eb) != root->root_key.objectid)
5889                 return 1;
5890
5891         /*
5892          * when reference count of tree block is 1, it won't increase
5893          * again. once full backref flag is set, we never clear it.
5894          */
5895         if (lookup_info &&
5896             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5897              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5898                 BUG_ON(!path->locks[level]);
5899                 ret = btrfs_lookup_extent_info(trans, root,
5900                                                eb->start, eb->len,
5901                                                &wc->refs[level],
5902                                                &wc->flags[level]);
5903                 BUG_ON(ret);
5904                 BUG_ON(wc->refs[level] == 0);
5905         }
5906
5907         if (wc->stage == DROP_REFERENCE) {
5908                 if (wc->refs[level] > 1)
5909                         return 1;
5910
5911                 if (path->locks[level] && !wc->keep_locks) {
5912                         btrfs_tree_unlock(eb);
5913                         path->locks[level] = 0;
5914                 }
5915                 return 0;
5916         }
5917
5918         /* wc->stage == UPDATE_BACKREF */
5919         if (!(wc->flags[level] & flag)) {
5920                 BUG_ON(!path->locks[level]);
5921                 ret = btrfs_inc_ref(trans, root, eb, 1);
5922                 BUG_ON(ret);
5923                 ret = btrfs_dec_ref(trans, root, eb, 0);
5924                 BUG_ON(ret);
5925                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5926                                                   eb->len, flag, 0);
5927                 BUG_ON(ret);
5928                 wc->flags[level] |= flag;
5929         }
5930
5931         /*
5932          * the block is shared by multiple trees, so it's not good to
5933          * keep the tree lock
5934          */
5935         if (path->locks[level] && level > 0) {
5936                 btrfs_tree_unlock(eb);
5937                 path->locks[level] = 0;
5938         }
5939         return 0;
5940 }
5941
5942 /*
5943  * hepler to process tree block pointer.
5944  *
5945  * when wc->stage == DROP_REFERENCE, this function checks
5946  * reference count of the block pointed to. if the block
5947  * is shared and we need update back refs for the subtree
5948  * rooted at the block, this function changes wc->stage to
5949  * UPDATE_BACKREF. if the block is shared and there is no
5950  * need to update back, this function drops the reference
5951  * to the block.
5952  *
5953  * NOTE: return value 1 means we should stop walking down.
5954  */
5955 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5956                                  struct btrfs_root *root,
5957                                  struct btrfs_path *path,
5958                                  struct walk_control *wc, int *lookup_info)
5959 {
5960         u64 bytenr;
5961         u64 generation;
5962         u64 parent;
5963         u32 blocksize;
5964         struct btrfs_key key;
5965         struct extent_buffer *next;
5966         int level = wc->level;
5967         int reada = 0;
5968         int ret = 0;
5969
5970         generation = btrfs_node_ptr_generation(path->nodes[level],
5971                                                path->slots[level]);
5972         /*
5973          * if the lower level block was created before the snapshot
5974          * was created, we know there is no need to update back refs
5975          * for the subtree
5976          */
5977         if (wc->stage == UPDATE_BACKREF &&
5978             generation <= root->root_key.offset) {
5979                 *lookup_info = 1;
5980                 return 1;
5981         }
5982
5983         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5984         blocksize = btrfs_level_size(root, level - 1);
5985
5986         next = btrfs_find_tree_block(root, bytenr, blocksize);
5987         if (!next) {
5988                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5989                 if (!next)
5990                         return -ENOMEM;
5991                 reada = 1;
5992         }
5993         btrfs_tree_lock(next);
5994         btrfs_set_lock_blocking(next);
5995
5996         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5997                                        &wc->refs[level - 1],
5998                                        &wc->flags[level - 1]);
5999         BUG_ON(ret);
6000         BUG_ON(wc->refs[level - 1] == 0);
6001         *lookup_info = 0;
6002
6003         if (wc->stage == DROP_REFERENCE) {
6004                 if (wc->refs[level - 1] > 1) {
6005                         if (level == 1 &&
6006                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6007                                 goto skip;
6008
6009                         if (!wc->update_ref ||
6010                             generation <= root->root_key.offset)
6011                                 goto skip;
6012
6013                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6014                                               path->slots[level]);
6015                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6016                         if (ret < 0)
6017                                 goto skip;
6018
6019                         wc->stage = UPDATE_BACKREF;
6020                         wc->shared_level = level - 1;
6021                 }
6022         } else {
6023                 if (level == 1 &&
6024                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6025                         goto skip;
6026         }
6027
6028         if (!btrfs_buffer_uptodate(next, generation)) {
6029                 btrfs_tree_unlock(next);
6030                 free_extent_buffer(next);
6031                 next = NULL;
6032                 *lookup_info = 1;
6033         }
6034
6035         if (!next) {
6036                 if (reada && level == 1)
6037                         reada_walk_down(trans, root, wc, path);
6038                 next = read_tree_block(root, bytenr, blocksize, generation);
6039                 btrfs_tree_lock(next);
6040                 btrfs_set_lock_blocking(next);
6041         }
6042
6043         level--;
6044         BUG_ON(level != btrfs_header_level(next));
6045         path->nodes[level] = next;
6046         path->slots[level] = 0;
6047         path->locks[level] = 1;
6048         wc->level = level;
6049         if (wc->level == 1)
6050                 wc->reada_slot = 0;
6051         return 0;
6052 skip:
6053         wc->refs[level - 1] = 0;
6054         wc->flags[level - 1] = 0;
6055         if (wc->stage == DROP_REFERENCE) {
6056                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6057                         parent = path->nodes[level]->start;
6058                 } else {
6059                         BUG_ON(root->root_key.objectid !=
6060                                btrfs_header_owner(path->nodes[level]));
6061                         parent = 0;
6062                 }
6063
6064                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6065                                         root->root_key.objectid, level - 1, 0);
6066                 BUG_ON(ret);
6067         }
6068         btrfs_tree_unlock(next);
6069         free_extent_buffer(next);
6070         *lookup_info = 1;
6071         return 1;
6072 }
6073
6074 /*
6075  * hepler to process tree block while walking up the tree.
6076  *
6077  * when wc->stage == DROP_REFERENCE, this function drops
6078  * reference count on the block.
6079  *
6080  * when wc->stage == UPDATE_BACKREF, this function changes
6081  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6082  * to UPDATE_BACKREF previously while processing the block.
6083  *
6084  * NOTE: return value 1 means we should stop walking up.
6085  */
6086 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6087                                  struct btrfs_root *root,
6088                                  struct btrfs_path *path,
6089                                  struct walk_control *wc)
6090 {
6091         int ret;
6092         int level = wc->level;
6093         struct extent_buffer *eb = path->nodes[level];
6094         u64 parent = 0;
6095
6096         if (wc->stage == UPDATE_BACKREF) {
6097                 BUG_ON(wc->shared_level < level);
6098                 if (level < wc->shared_level)
6099                         goto out;
6100
6101                 ret = find_next_key(path, level + 1, &wc->update_progress);
6102                 if (ret > 0)
6103                         wc->update_ref = 0;
6104
6105                 wc->stage = DROP_REFERENCE;
6106                 wc->shared_level = -1;
6107                 path->slots[level] = 0;
6108
6109                 /*
6110                  * check reference count again if the block isn't locked.
6111                  * we should start walking down the tree again if reference
6112                  * count is one.
6113                  */
6114                 if (!path->locks[level]) {
6115                         BUG_ON(level == 0);
6116                         btrfs_tree_lock(eb);
6117                         btrfs_set_lock_blocking(eb);
6118                         path->locks[level] = 1;
6119
6120                         ret = btrfs_lookup_extent_info(trans, root,
6121                                                        eb->start, eb->len,
6122                                                        &wc->refs[level],
6123                                                        &wc->flags[level]);
6124                         BUG_ON(ret);
6125                         BUG_ON(wc->refs[level] == 0);
6126                         if (wc->refs[level] == 1) {
6127                                 btrfs_tree_unlock(eb);
6128                                 path->locks[level] = 0;
6129                                 return 1;
6130                         }
6131                 }
6132         }
6133
6134         /* wc->stage == DROP_REFERENCE */
6135         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6136
6137         if (wc->refs[level] == 1) {
6138                 if (level == 0) {
6139                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6140                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6141                         else
6142                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6143                         BUG_ON(ret);
6144                 }
6145                 /* make block locked assertion in clean_tree_block happy */
6146                 if (!path->locks[level] &&
6147                     btrfs_header_generation(eb) == trans->transid) {
6148                         btrfs_tree_lock(eb);
6149                         btrfs_set_lock_blocking(eb);
6150                         path->locks[level] = 1;
6151                 }
6152                 clean_tree_block(trans, root, eb);
6153         }
6154
6155         if (eb == root->node) {
6156                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6157                         parent = eb->start;
6158                 else
6159                         BUG_ON(root->root_key.objectid !=
6160                                btrfs_header_owner(eb));
6161         } else {
6162                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6163                         parent = path->nodes[level + 1]->start;
6164                 else
6165                         BUG_ON(root->root_key.objectid !=
6166                                btrfs_header_owner(path->nodes[level + 1]));
6167         }
6168
6169         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6170 out:
6171         wc->refs[level] = 0;
6172         wc->flags[level] = 0;
6173         return 0;
6174 }
6175
6176 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6177                                    struct btrfs_root *root,
6178                                    struct btrfs_path *path,
6179                                    struct walk_control *wc)
6180 {
6181         int level = wc->level;
6182         int lookup_info = 1;
6183         int ret;
6184
6185         while (level >= 0) {
6186                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6187                 if (ret > 0)
6188                         break;
6189
6190                 if (level == 0)
6191                         break;
6192
6193                 if (path->slots[level] >=
6194                     btrfs_header_nritems(path->nodes[level]))
6195                         break;
6196
6197                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6198                 if (ret > 0) {
6199                         path->slots[level]++;
6200                         continue;
6201                 } else if (ret < 0)
6202                         return ret;
6203                 level = wc->level;
6204         }
6205         return 0;
6206 }
6207
6208 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6209                                  struct btrfs_root *root,
6210                                  struct btrfs_path *path,
6211                                  struct walk_control *wc, int max_level)
6212 {
6213         int level = wc->level;
6214         int ret;
6215
6216         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6217         while (level < max_level && path->nodes[level]) {
6218                 wc->level = level;
6219                 if (path->slots[level] + 1 <
6220                     btrfs_header_nritems(path->nodes[level])) {
6221                         path->slots[level]++;
6222                         return 0;
6223                 } else {
6224                         ret = walk_up_proc(trans, root, path, wc);
6225                         if (ret > 0)
6226                                 return 0;
6227
6228                         if (path->locks[level]) {
6229                                 btrfs_tree_unlock(path->nodes[level]);
6230                                 path->locks[level] = 0;
6231                         }
6232                         free_extent_buffer(path->nodes[level]);
6233                         path->nodes[level] = NULL;
6234                         level++;
6235                 }
6236         }
6237         return 1;
6238 }
6239
6240 /*
6241  * drop a subvolume tree.
6242  *
6243  * this function traverses the tree freeing any blocks that only
6244  * referenced by the tree.
6245  *
6246  * when a shared tree block is found. this function decreases its
6247  * reference count by one. if update_ref is true, this function
6248  * also make sure backrefs for the shared block and all lower level
6249  * blocks are properly updated.
6250  */
6251 int btrfs_drop_snapshot(struct btrfs_root *root,
6252                         struct btrfs_block_rsv *block_rsv, int update_ref)
6253 {
6254         struct btrfs_path *path;
6255         struct btrfs_trans_handle *trans;
6256         struct btrfs_root *tree_root = root->fs_info->tree_root;
6257         struct btrfs_root_item *root_item = &root->root_item;
6258         struct walk_control *wc;
6259         struct btrfs_key key;
6260         int err = 0;
6261         int ret;
6262         int level;
6263
6264         path = btrfs_alloc_path();
6265         BUG_ON(!path);
6266
6267         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6268         BUG_ON(!wc);
6269
6270         trans = btrfs_start_transaction(tree_root, 0);
6271         BUG_ON(IS_ERR(trans));
6272
6273         if (block_rsv)
6274                 trans->block_rsv = block_rsv;
6275
6276         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6277                 level = btrfs_header_level(root->node);
6278                 path->nodes[level] = btrfs_lock_root_node(root);
6279                 btrfs_set_lock_blocking(path->nodes[level]);
6280                 path->slots[level] = 0;
6281                 path->locks[level] = 1;
6282                 memset(&wc->update_progress, 0,
6283                        sizeof(wc->update_progress));
6284         } else {
6285                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6286                 memcpy(&wc->update_progress, &key,
6287                        sizeof(wc->update_progress));
6288
6289                 level = root_item->drop_level;
6290                 BUG_ON(level == 0);
6291                 path->lowest_level = level;
6292                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6293                 path->lowest_level = 0;
6294                 if (ret < 0) {
6295                         err = ret;
6296                         goto out;
6297                 }
6298                 WARN_ON(ret > 0);
6299
6300                 /*
6301                  * unlock our path, this is safe because only this
6302                  * function is allowed to delete this snapshot
6303                  */
6304                 btrfs_unlock_up_safe(path, 0);
6305
6306                 level = btrfs_header_level(root->node);
6307                 while (1) {
6308                         btrfs_tree_lock(path->nodes[level]);
6309                         btrfs_set_lock_blocking(path->nodes[level]);
6310
6311                         ret = btrfs_lookup_extent_info(trans, root,
6312                                                 path->nodes[level]->start,
6313                                                 path->nodes[level]->len,
6314                                                 &wc->refs[level],
6315                                                 &wc->flags[level]);
6316                         BUG_ON(ret);
6317                         BUG_ON(wc->refs[level] == 0);
6318
6319                         if (level == root_item->drop_level)
6320                                 break;
6321
6322                         btrfs_tree_unlock(path->nodes[level]);
6323                         WARN_ON(wc->refs[level] != 1);
6324                         level--;
6325                 }
6326         }
6327
6328         wc->level = level;
6329         wc->shared_level = -1;
6330         wc->stage = DROP_REFERENCE;
6331         wc->update_ref = update_ref;
6332         wc->keep_locks = 0;
6333         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6334
6335         while (1) {
6336                 ret = walk_down_tree(trans, root, path, wc);
6337                 if (ret < 0) {
6338                         err = ret;
6339                         break;
6340                 }
6341
6342                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6343                 if (ret < 0) {
6344                         err = ret;
6345                         break;
6346                 }
6347
6348                 if (ret > 0) {
6349                         BUG_ON(wc->stage != DROP_REFERENCE);
6350                         break;
6351                 }
6352
6353                 if (wc->stage == DROP_REFERENCE) {
6354                         level = wc->level;
6355                         btrfs_node_key(path->nodes[level],
6356                                        &root_item->drop_progress,
6357                                        path->slots[level]);
6358                         root_item->drop_level = level;
6359                 }
6360
6361                 BUG_ON(wc->level == 0);
6362                 if (btrfs_should_end_transaction(trans, tree_root)) {
6363                         ret = btrfs_update_root(trans, tree_root,
6364                                                 &root->root_key,
6365                                                 root_item);
6366                         BUG_ON(ret);
6367
6368                         btrfs_end_transaction_throttle(trans, tree_root);
6369                         trans = btrfs_start_transaction(tree_root, 0);
6370                         BUG_ON(IS_ERR(trans));
6371                         if (block_rsv)
6372                                 trans->block_rsv = block_rsv;
6373                 }
6374         }
6375         btrfs_release_path(root, path);
6376         BUG_ON(err);
6377
6378         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6379         BUG_ON(ret);
6380
6381         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6382                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6383                                            NULL, NULL);
6384                 BUG_ON(ret < 0);
6385                 if (ret > 0) {
6386                         /* if we fail to delete the orphan item this time
6387                          * around, it'll get picked up the next time.
6388                          *
6389                          * The most common failure here is just -ENOENT.
6390                          */
6391                         btrfs_del_orphan_item(trans, tree_root,
6392                                               root->root_key.objectid);
6393                 }
6394         }
6395
6396         if (root->in_radix) {
6397                 btrfs_free_fs_root(tree_root->fs_info, root);
6398         } else {
6399                 free_extent_buffer(root->node);
6400                 free_extent_buffer(root->commit_root);
6401                 kfree(root);
6402         }
6403 out:
6404         btrfs_end_transaction_throttle(trans, tree_root);
6405         kfree(wc);
6406         btrfs_free_path(path);
6407         return err;
6408 }
6409
6410 /*
6411  * drop subtree rooted at tree block 'node'.
6412  *
6413  * NOTE: this function will unlock and release tree block 'node'
6414  */
6415 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6416                         struct btrfs_root *root,
6417                         struct extent_buffer *node,
6418                         struct extent_buffer *parent)
6419 {
6420         struct btrfs_path *path;
6421         struct walk_control *wc;
6422         int level;
6423         int parent_level;
6424         int ret = 0;
6425         int wret;
6426
6427         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6428
6429         path = btrfs_alloc_path();
6430         BUG_ON(!path);
6431
6432         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6433         BUG_ON(!wc);
6434
6435         btrfs_assert_tree_locked(parent);
6436         parent_level = btrfs_header_level(parent);
6437         extent_buffer_get(parent);
6438         path->nodes[parent_level] = parent;
6439         path->slots[parent_level] = btrfs_header_nritems(parent);
6440
6441         btrfs_assert_tree_locked(node);
6442         level = btrfs_header_level(node);
6443         path->nodes[level] = node;
6444         path->slots[level] = 0;
6445         path->locks[level] = 1;
6446
6447         wc->refs[parent_level] = 1;
6448         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6449         wc->level = level;
6450         wc->shared_level = -1;
6451         wc->stage = DROP_REFERENCE;
6452         wc->update_ref = 0;
6453         wc->keep_locks = 1;
6454         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6455
6456         while (1) {
6457                 wret = walk_down_tree(trans, root, path, wc);
6458                 if (wret < 0) {
6459                         ret = wret;
6460                         break;
6461                 }
6462
6463                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6464                 if (wret < 0)
6465                         ret = wret;
6466                 if (wret != 0)
6467                         break;
6468         }
6469
6470         kfree(wc);
6471         btrfs_free_path(path);
6472         return ret;
6473 }
6474
6475 #if 0
6476 static unsigned long calc_ra(unsigned long start, unsigned long last,
6477                              unsigned long nr)
6478 {
6479         return min(last, start + nr - 1);
6480 }
6481
6482 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6483                                          u64 len)
6484 {
6485         u64 page_start;
6486         u64 page_end;
6487         unsigned long first_index;
6488         unsigned long last_index;
6489         unsigned long i;
6490         struct page *page;
6491         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6492         struct file_ra_state *ra;
6493         struct btrfs_ordered_extent *ordered;
6494         unsigned int total_read = 0;
6495         unsigned int total_dirty = 0;
6496         int ret = 0;
6497
6498         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6499         if (!ra)
6500                 return -ENOMEM;
6501
6502         mutex_lock(&inode->i_mutex);
6503         first_index = start >> PAGE_CACHE_SHIFT;
6504         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6505
6506         /* make sure the dirty trick played by the caller work */
6507         ret = invalidate_inode_pages2_range(inode->i_mapping,
6508                                             first_index, last_index);
6509         if (ret)
6510                 goto out_unlock;
6511
6512         file_ra_state_init(ra, inode->i_mapping);
6513
6514         for (i = first_index ; i <= last_index; i++) {
6515                 if (total_read % ra->ra_pages == 0) {
6516                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6517                                        calc_ra(i, last_index, ra->ra_pages));
6518                 }
6519                 total_read++;
6520 again:
6521                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6522                         BUG_ON(1);
6523                 page = grab_cache_page(inode->i_mapping, i);
6524                 if (!page) {
6525                         ret = -ENOMEM;
6526                         goto out_unlock;
6527                 }
6528                 if (!PageUptodate(page)) {
6529                         btrfs_readpage(NULL, page);
6530                         lock_page(page);
6531                         if (!PageUptodate(page)) {
6532                                 unlock_page(page);
6533                                 page_cache_release(page);
6534                                 ret = -EIO;
6535                                 goto out_unlock;
6536                         }
6537                 }
6538                 wait_on_page_writeback(page);
6539
6540                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6541                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6542                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6543
6544                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6545                 if (ordered) {
6546                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6547                         unlock_page(page);
6548                         page_cache_release(page);
6549                         btrfs_start_ordered_extent(inode, ordered, 1);
6550                         btrfs_put_ordered_extent(ordered);
6551                         goto again;
6552                 }
6553                 set_page_extent_mapped(page);
6554
6555                 if (i == first_index)
6556                         set_extent_bits(io_tree, page_start, page_end,
6557                                         EXTENT_BOUNDARY, GFP_NOFS);
6558                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6559
6560                 set_page_dirty(page);
6561                 total_dirty++;
6562
6563                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6564                 unlock_page(page);
6565                 page_cache_release(page);
6566         }
6567
6568 out_unlock:
6569         kfree(ra);
6570         mutex_unlock(&inode->i_mutex);
6571         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6572         return ret;
6573 }
6574
6575 static noinline int relocate_data_extent(struct inode *reloc_inode,
6576                                          struct btrfs_key *extent_key,
6577                                          u64 offset)
6578 {
6579         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6580         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6581         struct extent_map *em;
6582         u64 start = extent_key->objectid - offset;
6583         u64 end = start + extent_key->offset - 1;
6584
6585         em = alloc_extent_map(GFP_NOFS);
6586         BUG_ON(!em || IS_ERR(em));
6587
6588         em->start = start;
6589         em->len = extent_key->offset;
6590         em->block_len = extent_key->offset;
6591         em->block_start = extent_key->objectid;
6592         em->bdev = root->fs_info->fs_devices->latest_bdev;
6593         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6594
6595         /* setup extent map to cheat btrfs_readpage */
6596         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6597         while (1) {
6598                 int ret;
6599                 write_lock(&em_tree->lock);
6600                 ret = add_extent_mapping(em_tree, em);
6601                 write_unlock(&em_tree->lock);
6602                 if (ret != -EEXIST) {
6603                         free_extent_map(em);
6604                         break;
6605                 }
6606                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6607         }
6608         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6609
6610         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6611 }
6612
6613 struct btrfs_ref_path {
6614         u64 extent_start;
6615         u64 nodes[BTRFS_MAX_LEVEL];
6616         u64 root_objectid;
6617         u64 root_generation;
6618         u64 owner_objectid;
6619         u32 num_refs;
6620         int lowest_level;
6621         int current_level;
6622         int shared_level;
6623
6624         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6625         u64 new_nodes[BTRFS_MAX_LEVEL];
6626 };
6627
6628 struct disk_extent {
6629         u64 ram_bytes;
6630         u64 disk_bytenr;
6631         u64 disk_num_bytes;
6632         u64 offset;
6633         u64 num_bytes;
6634         u8 compression;
6635         u8 encryption;
6636         u16 other_encoding;
6637 };
6638
6639 static int is_cowonly_root(u64 root_objectid)
6640 {
6641         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6642             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6643             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6644             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6645             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6646             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6647                 return 1;
6648         return 0;
6649 }
6650
6651 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6652                                     struct btrfs_root *extent_root,
6653                                     struct btrfs_ref_path *ref_path,
6654                                     int first_time)
6655 {
6656         struct extent_buffer *leaf;
6657         struct btrfs_path *path;
6658         struct btrfs_extent_ref *ref;
6659         struct btrfs_key key;
6660         struct btrfs_key found_key;
6661         u64 bytenr;
6662         u32 nritems;
6663         int level;
6664         int ret = 1;
6665
6666         path = btrfs_alloc_path();
6667         if (!path)
6668                 return -ENOMEM;
6669
6670         if (first_time) {
6671                 ref_path->lowest_level = -1;
6672                 ref_path->current_level = -1;
6673                 ref_path->shared_level = -1;
6674                 goto walk_up;
6675         }
6676 walk_down:
6677         level = ref_path->current_level - 1;
6678         while (level >= -1) {
6679                 u64 parent;
6680                 if (level < ref_path->lowest_level)
6681                         break;
6682
6683                 if (level >= 0)
6684                         bytenr = ref_path->nodes[level];
6685                 else
6686                         bytenr = ref_path->extent_start;
6687                 BUG_ON(bytenr == 0);
6688
6689                 parent = ref_path->nodes[level + 1];
6690                 ref_path->nodes[level + 1] = 0;
6691                 ref_path->current_level = level;
6692                 BUG_ON(parent == 0);
6693
6694                 key.objectid = bytenr;
6695                 key.offset = parent + 1;
6696                 key.type = BTRFS_EXTENT_REF_KEY;
6697
6698                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6699                 if (ret < 0)
6700                         goto out;
6701                 BUG_ON(ret == 0);
6702
6703                 leaf = path->nodes[0];
6704                 nritems = btrfs_header_nritems(leaf);
6705                 if (path->slots[0] >= nritems) {
6706                         ret = btrfs_next_leaf(extent_root, path);
6707                         if (ret < 0)
6708                                 goto out;
6709                         if (ret > 0)
6710                                 goto next;
6711                         leaf = path->nodes[0];
6712                 }
6713
6714                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6715                 if (found_key.objectid == bytenr &&
6716                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6717                         if (level < ref_path->shared_level)
6718                                 ref_path->shared_level = level;
6719                         goto found;
6720                 }
6721 next:
6722                 level--;
6723                 btrfs_release_path(extent_root, path);
6724                 cond_resched();
6725         }
6726         /* reached lowest level */
6727         ret = 1;
6728         goto out;
6729 walk_up:
6730         level = ref_path->current_level;
6731         while (level < BTRFS_MAX_LEVEL - 1) {
6732                 u64 ref_objectid;
6733
6734                 if (level >= 0)
6735                         bytenr = ref_path->nodes[level];
6736                 else
6737                         bytenr = ref_path->extent_start;
6738
6739                 BUG_ON(bytenr == 0);
6740
6741                 key.objectid = bytenr;
6742                 key.offset = 0;
6743                 key.type = BTRFS_EXTENT_REF_KEY;
6744
6745                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6746                 if (ret < 0)
6747                         goto out;
6748
6749                 leaf = path->nodes[0];
6750                 nritems = btrfs_header_nritems(leaf);
6751                 if (path->slots[0] >= nritems) {
6752                         ret = btrfs_next_leaf(extent_root, path);
6753                         if (ret < 0)
6754                                 goto out;
6755                         if (ret > 0) {
6756                                 /* the extent was freed by someone */
6757                                 if (ref_path->lowest_level == level)
6758                                         goto out;
6759                                 btrfs_release_path(extent_root, path);
6760                                 goto walk_down;
6761                         }
6762                         leaf = path->nodes[0];
6763                 }
6764
6765                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6766                 if (found_key.objectid != bytenr ||
6767                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6768                         /* the extent was freed by someone */
6769                         if (ref_path->lowest_level == level) {
6770                                 ret = 1;
6771                                 goto out;
6772                         }
6773                         btrfs_release_path(extent_root, path);
6774                         goto walk_down;
6775                 }
6776 found:
6777                 ref = btrfs_item_ptr(leaf, path->slots[0],
6778                                 struct btrfs_extent_ref);
6779                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6780                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6781                         if (first_time) {
6782                                 level = (int)ref_objectid;
6783                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6784                                 ref_path->lowest_level = level;
6785                                 ref_path->current_level = level;
6786                                 ref_path->nodes[level] = bytenr;
6787                         } else {
6788                                 WARN_ON(ref_objectid != level);
6789                         }
6790                 } else {
6791                         WARN_ON(level != -1);
6792                 }
6793                 first_time = 0;
6794
6795                 if (ref_path->lowest_level == level) {
6796                         ref_path->owner_objectid = ref_objectid;
6797                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6798                 }
6799
6800                 /*
6801                  * the block is tree root or the block isn't in reference
6802                  * counted tree.
6803                  */
6804                 if (found_key.objectid == found_key.offset ||
6805                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6806                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6807                         ref_path->root_generation =
6808                                 btrfs_ref_generation(leaf, ref);
6809                         if (level < 0) {
6810                                 /* special reference from the tree log */
6811                                 ref_path->nodes[0] = found_key.offset;
6812                                 ref_path->current_level = 0;
6813                         }
6814                         ret = 0;
6815                         goto out;
6816                 }
6817
6818                 level++;
6819                 BUG_ON(ref_path->nodes[level] != 0);
6820                 ref_path->nodes[level] = found_key.offset;
6821                 ref_path->current_level = level;
6822
6823                 /*
6824                  * the reference was created in the running transaction,
6825                  * no need to continue walking up.
6826                  */
6827                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6828                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6829                         ref_path->root_generation =
6830                                 btrfs_ref_generation(leaf, ref);
6831                         ret = 0;
6832                         goto out;
6833                 }
6834
6835                 btrfs_release_path(extent_root, path);
6836                 cond_resched();
6837         }
6838         /* reached max tree level, but no tree root found. */
6839         BUG();
6840 out:
6841         btrfs_free_path(path);
6842         return ret;
6843 }
6844
6845 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6846                                 struct btrfs_root *extent_root,
6847                                 struct btrfs_ref_path *ref_path,
6848                                 u64 extent_start)
6849 {
6850         memset(ref_path, 0, sizeof(*ref_path));
6851         ref_path->extent_start = extent_start;
6852
6853         return __next_ref_path(trans, extent_root, ref_path, 1);
6854 }
6855
6856 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6857                                struct btrfs_root *extent_root,
6858                                struct btrfs_ref_path *ref_path)
6859 {
6860         return __next_ref_path(trans, extent_root, ref_path, 0);
6861 }
6862
6863 static noinline int get_new_locations(struct inode *reloc_inode,
6864                                       struct btrfs_key *extent_key,
6865                                       u64 offset, int no_fragment,
6866                                       struct disk_extent **extents,
6867                                       int *nr_extents)
6868 {
6869         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6870         struct btrfs_path *path;
6871         struct btrfs_file_extent_item *fi;
6872         struct extent_buffer *leaf;
6873         struct disk_extent *exts = *extents;
6874         struct btrfs_key found_key;
6875         u64 cur_pos;
6876         u64 last_byte;
6877         u32 nritems;
6878         int nr = 0;
6879         int max = *nr_extents;
6880         int ret;
6881
6882         WARN_ON(!no_fragment && *extents);
6883         if (!exts) {
6884                 max = 1;
6885                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6886                 if (!exts)
6887                         return -ENOMEM;
6888         }
6889
6890         path = btrfs_alloc_path();
6891         BUG_ON(!path);
6892
6893         cur_pos = extent_key->objectid - offset;
6894         last_byte = extent_key->objectid + extent_key->offset;
6895         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6896                                        cur_pos, 0);
6897         if (ret < 0)
6898                 goto out;
6899         if (ret > 0) {
6900                 ret = -ENOENT;
6901                 goto out;
6902         }
6903
6904         while (1) {
6905                 leaf = path->nodes[0];
6906                 nritems = btrfs_header_nritems(leaf);
6907                 if (path->slots[0] >= nritems) {
6908                         ret = btrfs_next_leaf(root, path);
6909                         if (ret < 0)
6910                                 goto out;
6911                         if (ret > 0)
6912                                 break;
6913                         leaf = path->nodes[0];
6914                 }
6915
6916                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6917                 if (found_key.offset != cur_pos ||
6918                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6919                     found_key.objectid != reloc_inode->i_ino)
6920                         break;
6921
6922                 fi = btrfs_item_ptr(leaf, path->slots[0],
6923                                     struct btrfs_file_extent_item);
6924                 if (btrfs_file_extent_type(leaf, fi) !=
6925                     BTRFS_FILE_EXTENT_REG ||
6926                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6927                         break;
6928
6929                 if (nr == max) {
6930                         struct disk_extent *old = exts;
6931                         max *= 2;
6932                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6933                         memcpy(exts, old, sizeof(*exts) * nr);
6934                         if (old != *extents)
6935                                 kfree(old);
6936                 }
6937
6938                 exts[nr].disk_bytenr =
6939                         btrfs_file_extent_disk_bytenr(leaf, fi);
6940                 exts[nr].disk_num_bytes =
6941                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6942                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6943                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6944                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6945                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6946                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6947                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6948                                                                            fi);
6949                 BUG_ON(exts[nr].offset > 0);
6950                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6951                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6952
6953                 cur_pos += exts[nr].num_bytes;
6954                 nr++;
6955
6956                 if (cur_pos + offset >= last_byte)
6957                         break;
6958
6959                 if (no_fragment) {
6960                         ret = 1;
6961                         goto out;
6962                 }
6963                 path->slots[0]++;
6964         }
6965
6966         BUG_ON(cur_pos + offset > last_byte);
6967         if (cur_pos + offset < last_byte) {
6968                 ret = -ENOENT;
6969                 goto out;
6970         }
6971         ret = 0;
6972 out:
6973         btrfs_free_path(path);
6974         if (ret) {
6975                 if (exts != *extents)
6976                         kfree(exts);
6977         } else {
6978                 *extents = exts;
6979                 *nr_extents = nr;
6980         }
6981         return ret;
6982 }
6983
6984 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6985                                         struct btrfs_root *root,
6986                                         struct btrfs_path *path,
6987                                         struct btrfs_key *extent_key,
6988                                         struct btrfs_key *leaf_key,
6989                                         struct btrfs_ref_path *ref_path,
6990                                         struct disk_extent *new_extents,
6991                                         int nr_extents)
6992 {
6993         struct extent_buffer *leaf;
6994         struct btrfs_file_extent_item *fi;
6995         struct inode *inode = NULL;
6996         struct btrfs_key key;
6997         u64 lock_start = 0;
6998         u64 lock_end = 0;
6999         u64 num_bytes;
7000         u64 ext_offset;
7001         u64 search_end = (u64)-1;
7002         u32 nritems;
7003         int nr_scaned = 0;
7004         int extent_locked = 0;
7005         int extent_type;
7006         int ret;
7007
7008         memcpy(&key, leaf_key, sizeof(key));
7009         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7010                 if (key.objectid < ref_path->owner_objectid ||
7011                     (key.objectid == ref_path->owner_objectid &&
7012                      key.type < BTRFS_EXTENT_DATA_KEY)) {
7013                         key.objectid = ref_path->owner_objectid;
7014                         key.type = BTRFS_EXTENT_DATA_KEY;
7015                         key.offset = 0;
7016                 }
7017         }
7018
7019         while (1) {
7020                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7021                 if (ret < 0)
7022                         goto out;
7023
7024                 leaf = path->nodes[0];
7025                 nritems = btrfs_header_nritems(leaf);
7026 next:
7027                 if (extent_locked && ret > 0) {
7028                         /*
7029                          * the file extent item was modified by someone
7030                          * before the extent got locked.
7031                          */
7032                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7033                                       lock_end, GFP_NOFS);
7034                         extent_locked = 0;
7035                 }
7036
7037                 if (path->slots[0] >= nritems) {
7038                         if (++nr_scaned > 2)
7039                                 break;
7040
7041                         BUG_ON(extent_locked);
7042                         ret = btrfs_next_leaf(root, path);
7043                         if (ret < 0)
7044                                 goto out;
7045                         if (ret > 0)
7046                                 break;
7047                         leaf = path->nodes[0];
7048                         nritems = btrfs_header_nritems(leaf);
7049                 }
7050
7051                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7052
7053                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7054                         if ((key.objectid > ref_path->owner_objectid) ||
7055                             (key.objectid == ref_path->owner_objectid &&
7056                              key.type > BTRFS_EXTENT_DATA_KEY) ||
7057                             key.offset >= search_end)
7058                                 break;
7059                 }
7060
7061                 if (inode && key.objectid != inode->i_ino) {
7062                         BUG_ON(extent_locked);
7063                         btrfs_release_path(root, path);
7064                         mutex_unlock(&inode->i_mutex);
7065                         iput(inode);
7066                         inode = NULL;
7067                         continue;
7068                 }
7069
7070                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7071                         path->slots[0]++;
7072                         ret = 1;
7073                         goto next;
7074                 }
7075                 fi = btrfs_item_ptr(leaf, path->slots[0],
7076                                     struct btrfs_file_extent_item);
7077                 extent_type = btrfs_file_extent_type(leaf, fi);
7078                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7079                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7080                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7081                      extent_key->objectid)) {
7082                         path->slots[0]++;
7083                         ret = 1;
7084                         goto next;
7085                 }
7086
7087                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7088                 ext_offset = btrfs_file_extent_offset(leaf, fi);
7089
7090                 if (search_end == (u64)-1) {
7091                         search_end = key.offset - ext_offset +
7092                                 btrfs_file_extent_ram_bytes(leaf, fi);
7093                 }
7094
7095                 if (!extent_locked) {
7096                         lock_start = key.offset;
7097                         lock_end = lock_start + num_bytes - 1;
7098                 } else {
7099                         if (lock_start > key.offset ||
7100                             lock_end + 1 < key.offset + num_bytes) {
7101                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7102                                               lock_start, lock_end, GFP_NOFS);
7103                                 extent_locked = 0;
7104                         }
7105                 }
7106
7107                 if (!inode) {
7108                         btrfs_release_path(root, path);
7109
7110                         inode = btrfs_iget_locked(root->fs_info->sb,
7111                                                   key.objectid, root);
7112                         if (inode->i_state & I_NEW) {
7113                                 BTRFS_I(inode)->root = root;
7114                                 BTRFS_I(inode)->location.objectid =
7115                                         key.objectid;
7116                                 BTRFS_I(inode)->location.type =
7117                                         BTRFS_INODE_ITEM_KEY;
7118                                 BTRFS_I(inode)->location.offset = 0;
7119                                 btrfs_read_locked_inode(inode);
7120                                 unlock_new_inode(inode);
7121                         }
7122                         /*
7123                          * some code call btrfs_commit_transaction while
7124                          * holding the i_mutex, so we can't use mutex_lock
7125                          * here.
7126                          */
7127                         if (is_bad_inode(inode) ||
7128                             !mutex_trylock(&inode->i_mutex)) {
7129                                 iput(inode);
7130                                 inode = NULL;
7131                                 key.offset = (u64)-1;
7132                                 goto skip;
7133                         }
7134                 }
7135
7136                 if (!extent_locked) {
7137                         struct btrfs_ordered_extent *ordered;
7138
7139                         btrfs_release_path(root, path);
7140
7141                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7142                                     lock_end, GFP_NOFS);
7143                         ordered = btrfs_lookup_first_ordered_extent(inode,
7144                                                                     lock_end);
7145                         if (ordered &&
7146                             ordered->file_offset <= lock_end &&
7147                             ordered->file_offset + ordered->len > lock_start) {
7148                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7149                                               lock_start, lock_end, GFP_NOFS);
7150                                 btrfs_start_ordered_extent(inode, ordered, 1);
7151                                 btrfs_put_ordered_extent(ordered);
7152                                 key.offset += num_bytes;
7153                                 goto skip;
7154                         }
7155                         if (ordered)
7156                                 btrfs_put_ordered_extent(ordered);
7157
7158                         extent_locked = 1;
7159                         continue;
7160                 }
7161
7162                 if (nr_extents == 1) {
7163                         /* update extent pointer in place */
7164                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7165                                                 new_extents[0].disk_bytenr);
7166                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7167                                                 new_extents[0].disk_num_bytes);
7168                         btrfs_mark_buffer_dirty(leaf);
7169
7170                         btrfs_drop_extent_cache(inode, key.offset,
7171                                                 key.offset + num_bytes - 1, 0);
7172
7173                         ret = btrfs_inc_extent_ref(trans, root,
7174                                                 new_extents[0].disk_bytenr,
7175                                                 new_extents[0].disk_num_bytes,
7176                                                 leaf->start,
7177                                                 root->root_key.objectid,
7178                                                 trans->transid,
7179                                                 key.objectid);
7180                         BUG_ON(ret);
7181
7182                         ret = btrfs_free_extent(trans, root,
7183                                                 extent_key->objectid,
7184                                                 extent_key->offset,
7185                                                 leaf->start,
7186                                                 btrfs_header_owner(leaf),
7187                                                 btrfs_header_generation(leaf),
7188                                                 key.objectid, 0);
7189                         BUG_ON(ret);
7190
7191                         btrfs_release_path(root, path);
7192                         key.offset += num_bytes;
7193                 } else {
7194                         BUG_ON(1);
7195 #if 0
7196                         u64 alloc_hint;
7197                         u64 extent_len;
7198                         int i;
7199                         /*
7200                          * drop old extent pointer at first, then insert the
7201                          * new pointers one bye one
7202                          */
7203                         btrfs_release_path(root, path);
7204                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7205                                                  key.offset + num_bytes,
7206                                                  key.offset, &alloc_hint);
7207                         BUG_ON(ret);
7208
7209                         for (i = 0; i < nr_extents; i++) {
7210                                 if (ext_offset >= new_extents[i].num_bytes) {
7211                                         ext_offset -= new_extents[i].num_bytes;
7212                                         continue;
7213                                 }
7214                                 extent_len = min(new_extents[i].num_bytes -
7215                                                  ext_offset, num_bytes);
7216
7217                                 ret = btrfs_insert_empty_item(trans, root,
7218                                                               path, &key,
7219                                                               sizeof(*fi));
7220                                 BUG_ON(ret);
7221
7222                                 leaf = path->nodes[0];
7223                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7224                                                 struct btrfs_file_extent_item);
7225                                 btrfs_set_file_extent_generation(leaf, fi,
7226                                                         trans->transid);
7227                                 btrfs_set_file_extent_type(leaf, fi,
7228                                                         BTRFS_FILE_EXTENT_REG);
7229                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7230                                                 new_extents[i].disk_bytenr);
7231                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7232                                                 new_extents[i].disk_num_bytes);
7233                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7234                                                 new_extents[i].ram_bytes);
7235
7236                                 btrfs_set_file_extent_compression(leaf, fi,
7237                                                 new_extents[i].compression);
7238                                 btrfs_set_file_extent_encryption(leaf, fi,
7239                                                 new_extents[i].encryption);
7240                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7241                                                 new_extents[i].other_encoding);
7242
7243                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7244                                                         extent_len);
7245                                 ext_offset += new_extents[i].offset;
7246                                 btrfs_set_file_extent_offset(leaf, fi,
7247                                                         ext_offset);
7248                                 btrfs_mark_buffer_dirty(leaf);
7249
7250                                 btrfs_drop_extent_cache(inode, key.offset,
7251                                                 key.offset + extent_len - 1, 0);
7252
7253                                 ret = btrfs_inc_extent_ref(trans, root,
7254                                                 new_extents[i].disk_bytenr,
7255                                                 new_extents[i].disk_num_bytes,
7256                                                 leaf->start,
7257                                                 root->root_key.objectid,
7258                                                 trans->transid, key.objectid);
7259                                 BUG_ON(ret);
7260                                 btrfs_release_path(root, path);
7261
7262                                 inode_add_bytes(inode, extent_len);
7263
7264                                 ext_offset = 0;
7265                                 num_bytes -= extent_len;
7266                                 key.offset += extent_len;
7267
7268                                 if (num_bytes == 0)
7269                                         break;
7270                         }
7271                         BUG_ON(i >= nr_extents);
7272 #endif
7273                 }
7274
7275                 if (extent_locked) {
7276                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7277                                       lock_end, GFP_NOFS);
7278                         extent_locked = 0;
7279                 }
7280 skip:
7281                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7282                     key.offset >= search_end)
7283                         break;
7284
7285                 cond_resched();
7286         }
7287         ret = 0;
7288 out:
7289         btrfs_release_path(root, path);
7290         if (inode) {
7291                 mutex_unlock(&inode->i_mutex);
7292                 if (extent_locked) {
7293                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7294                                       lock_end, GFP_NOFS);
7295                 }
7296                 iput(inode);
7297         }
7298         return ret;
7299 }
7300
7301 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7302                                struct btrfs_root *root,
7303                                struct extent_buffer *buf, u64 orig_start)
7304 {
7305         int level;
7306         int ret;
7307
7308         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7309         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7310
7311         level = btrfs_header_level(buf);
7312         if (level == 0) {
7313                 struct btrfs_leaf_ref *ref;
7314                 struct btrfs_leaf_ref *orig_ref;
7315
7316                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7317                 if (!orig_ref)
7318                         return -ENOENT;
7319
7320                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7321                 if (!ref) {
7322                         btrfs_free_leaf_ref(root, orig_ref);
7323                         return -ENOMEM;
7324                 }
7325
7326                 ref->nritems = orig_ref->nritems;
7327                 memcpy(ref->extents, orig_ref->extents,
7328                         sizeof(ref->extents[0]) * ref->nritems);
7329
7330                 btrfs_free_leaf_ref(root, orig_ref);
7331
7332                 ref->root_gen = trans->transid;
7333                 ref->bytenr = buf->start;
7334                 ref->owner = btrfs_header_owner(buf);
7335                 ref->generation = btrfs_header_generation(buf);
7336
7337                 ret = btrfs_add_leaf_ref(root, ref, 0);
7338                 WARN_ON(ret);
7339                 btrfs_free_leaf_ref(root, ref);
7340         }
7341         return 0;
7342 }
7343
7344 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7345                                         struct extent_buffer *leaf,
7346                                         struct btrfs_block_group_cache *group,
7347                                         struct btrfs_root *target_root)
7348 {
7349         struct btrfs_key key;
7350         struct inode *inode = NULL;
7351         struct btrfs_file_extent_item *fi;
7352         struct extent_state *cached_state = NULL;
7353         u64 num_bytes;
7354         u64 skip_objectid = 0;
7355         u32 nritems;
7356         u32 i;
7357
7358         nritems = btrfs_header_nritems(leaf);
7359         for (i = 0; i < nritems; i++) {
7360                 btrfs_item_key_to_cpu(leaf, &key, i);
7361                 if (key.objectid == skip_objectid ||
7362                     key.type != BTRFS_EXTENT_DATA_KEY)
7363                         continue;
7364                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7365                 if (btrfs_file_extent_type(leaf, fi) ==
7366                     BTRFS_FILE_EXTENT_INLINE)
7367                         continue;
7368                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7369                         continue;
7370                 if (!inode || inode->i_ino != key.objectid) {
7371                         iput(inode);
7372                         inode = btrfs_ilookup(target_root->fs_info->sb,
7373                                               key.objectid, target_root, 1);
7374                 }
7375                 if (!inode) {
7376                         skip_objectid = key.objectid;
7377                         continue;
7378                 }
7379                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7380
7381                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7382                                  key.offset + num_bytes - 1, 0, &cached_state,
7383                                  GFP_NOFS);
7384                 btrfs_drop_extent_cache(inode, key.offset,
7385                                         key.offset + num_bytes - 1, 1);
7386                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7387                                      key.offset + num_bytes - 1, &cached_state,
7388                                      GFP_NOFS);
7389                 cond_resched();
7390         }
7391         iput(inode);
7392         return 0;
7393 }
7394
7395 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7396                                         struct btrfs_root *root,
7397                                         struct extent_buffer *leaf,
7398                                         struct btrfs_block_group_cache *group,
7399                                         struct inode *reloc_inode)
7400 {
7401         struct btrfs_key key;
7402         struct btrfs_key extent_key;
7403         struct btrfs_file_extent_item *fi;
7404         struct btrfs_leaf_ref *ref;
7405         struct disk_extent *new_extent;
7406         u64 bytenr;
7407         u64 num_bytes;
7408         u32 nritems;
7409         u32 i;
7410         int ext_index;
7411         int nr_extent;
7412         int ret;
7413
7414         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7415         BUG_ON(!new_extent);
7416
7417         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7418         BUG_ON(!ref);
7419
7420         ext_index = -1;
7421         nritems = btrfs_header_nritems(leaf);
7422         for (i = 0; i < nritems; i++) {
7423                 btrfs_item_key_to_cpu(leaf, &key, i);
7424                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7425                         continue;
7426                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7427                 if (btrfs_file_extent_type(leaf, fi) ==
7428                     BTRFS_FILE_EXTENT_INLINE)
7429                         continue;
7430                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7431                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7432                 if (bytenr == 0)
7433                         continue;
7434
7435                 ext_index++;
7436                 if (bytenr >= group->key.objectid + group->key.offset ||
7437                     bytenr + num_bytes <= group->key.objectid)
7438                         continue;
7439
7440                 extent_key.objectid = bytenr;
7441                 extent_key.offset = num_bytes;
7442                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7443                 nr_extent = 1;
7444                 ret = get_new_locations(reloc_inode, &extent_key,
7445                                         group->key.objectid, 1,
7446                                         &new_extent, &nr_extent);
7447                 if (ret > 0)
7448                         continue;
7449                 BUG_ON(ret < 0);
7450
7451                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7452                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7453                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7454                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7455
7456                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7457                                                 new_extent->disk_bytenr);
7458                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7459                                                 new_extent->disk_num_bytes);
7460                 btrfs_mark_buffer_dirty(leaf);
7461
7462                 ret = btrfs_inc_extent_ref(trans, root,
7463                                         new_extent->disk_bytenr,
7464                                         new_extent->disk_num_bytes,
7465                                         leaf->start,
7466                                         root->root_key.objectid,
7467                                         trans->transid, key.objectid);
7468                 BUG_ON(ret);
7469
7470                 ret = btrfs_free_extent(trans, root,
7471                                         bytenr, num_bytes, leaf->start,
7472                                         btrfs_header_owner(leaf),
7473                                         btrfs_header_generation(leaf),
7474                                         key.objectid, 0);
7475                 BUG_ON(ret);
7476                 cond_resched();
7477         }
7478         kfree(new_extent);
7479         BUG_ON(ext_index + 1 != ref->nritems);
7480         btrfs_free_leaf_ref(root, ref);
7481         return 0;
7482 }
7483
7484 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7485                           struct btrfs_root *root)
7486 {
7487         struct btrfs_root *reloc_root;
7488         int ret;
7489
7490         if (root->reloc_root) {
7491                 reloc_root = root->reloc_root;
7492                 root->reloc_root = NULL;
7493                 list_add(&reloc_root->dead_list,
7494                          &root->fs_info->dead_reloc_roots);
7495
7496                 btrfs_set_root_bytenr(&reloc_root->root_item,
7497                                       reloc_root->node->start);
7498                 btrfs_set_root_level(&root->root_item,
7499                                      btrfs_header_level(reloc_root->node));
7500                 memset(&reloc_root->root_item.drop_progress, 0,
7501                         sizeof(struct btrfs_disk_key));
7502                 reloc_root->root_item.drop_level = 0;
7503
7504                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7505                                         &reloc_root->root_key,
7506                                         &reloc_root->root_item);
7507                 BUG_ON(ret);
7508         }
7509         return 0;
7510 }
7511
7512 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7513 {
7514         struct btrfs_trans_handle *trans;
7515         struct btrfs_root *reloc_root;
7516         struct btrfs_root *prev_root = NULL;
7517         struct list_head dead_roots;
7518         int ret;
7519         unsigned long nr;
7520
7521         INIT_LIST_HEAD(&dead_roots);
7522         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7523
7524         while (!list_empty(&dead_roots)) {
7525                 reloc_root = list_entry(dead_roots.prev,
7526                                         struct btrfs_root, dead_list);
7527                 list_del_init(&reloc_root->dead_list);
7528
7529                 BUG_ON(reloc_root->commit_root != NULL);
7530                 while (1) {
7531                         trans = btrfs_join_transaction(root, 1);
7532                         BUG_ON(IS_ERR(trans));
7533
7534                         mutex_lock(&root->fs_info->drop_mutex);
7535                         ret = btrfs_drop_snapshot(trans, reloc_root);
7536                         if (ret != -EAGAIN)
7537                                 break;
7538                         mutex_unlock(&root->fs_info->drop_mutex);
7539
7540                         nr = trans->blocks_used;
7541                         ret = btrfs_end_transaction(trans, root);
7542                         BUG_ON(ret);
7543                         btrfs_btree_balance_dirty(root, nr);
7544                 }
7545
7546                 free_extent_buffer(reloc_root->node);
7547
7548                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7549                                      &reloc_root->root_key);
7550                 BUG_ON(ret);
7551                 mutex_unlock(&root->fs_info->drop_mutex);
7552
7553                 nr = trans->blocks_used;
7554                 ret = btrfs_end_transaction(trans, root);
7555                 BUG_ON(ret);
7556                 btrfs_btree_balance_dirty(root, nr);
7557
7558                 kfree(prev_root);
7559                 prev_root = reloc_root;
7560         }
7561         if (prev_root) {
7562                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7563                 kfree(prev_root);
7564         }
7565         return 0;
7566 }
7567
7568 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7569 {
7570         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7571         return 0;
7572 }
7573
7574 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7575 {
7576         struct btrfs_root *reloc_root;
7577         struct btrfs_trans_handle *trans;
7578         struct btrfs_key location;
7579         int found;
7580         int ret;
7581
7582         mutex_lock(&root->fs_info->tree_reloc_mutex);
7583         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7584         BUG_ON(ret);
7585         found = !list_empty(&root->fs_info->dead_reloc_roots);
7586         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7587
7588         if (found) {
7589                 trans = btrfs_start_transaction(root, 1);
7590                 BUG_ON(IS_ERR(trans));
7591                 ret = btrfs_commit_transaction(trans, root);
7592                 BUG_ON(ret);
7593         }
7594
7595         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7596         location.offset = (u64)-1;
7597         location.type = BTRFS_ROOT_ITEM_KEY;
7598
7599         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7600         BUG_ON(!reloc_root);
7601         btrfs_orphan_cleanup(reloc_root);
7602         return 0;
7603 }
7604
7605 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7606                                     struct btrfs_root *root)
7607 {
7608         struct btrfs_root *reloc_root;
7609         struct extent_buffer *eb;
7610         struct btrfs_root_item *root_item;
7611         struct btrfs_key root_key;
7612         int ret;
7613
7614         BUG_ON(!root->ref_cows);
7615         if (root->reloc_root)
7616                 return 0;
7617
7618         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7619         BUG_ON(!root_item);
7620
7621         ret = btrfs_copy_root(trans, root, root->commit_root,
7622                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7623         BUG_ON(ret);
7624
7625         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7626         root_key.offset = root->root_key.objectid;
7627         root_key.type = BTRFS_ROOT_ITEM_KEY;
7628
7629         memcpy(root_item, &root->root_item, sizeof(root_item));
7630         btrfs_set_root_refs(root_item, 0);
7631         btrfs_set_root_bytenr(root_item, eb->start);
7632         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7633         btrfs_set_root_generation(root_item, trans->transid);
7634
7635         btrfs_tree_unlock(eb);
7636         free_extent_buffer(eb);
7637
7638         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7639                                 &root_key, root_item);
7640         BUG_ON(ret);
7641         kfree(root_item);
7642
7643         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7644                                                  &root_key);
7645         BUG_ON(!reloc_root);
7646         reloc_root->last_trans = trans->transid;
7647         reloc_root->commit_root = NULL;
7648         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7649
7650         root->reloc_root = reloc_root;
7651         return 0;
7652 }
7653
7654 /*
7655  * Core function of space balance.
7656  *
7657  * The idea is using reloc trees to relocate tree blocks in reference
7658  * counted roots. There is one reloc tree for each subvol, and all
7659  * reloc trees share same root key objectid. Reloc trees are snapshots
7660  * of the latest committed roots of subvols (root->commit_root).
7661  *
7662  * To relocate a tree block referenced by a subvol, there are two steps.
7663  * COW the block through subvol's reloc tree, then update block pointer
7664  * in the subvol to point to the new block. Since all reloc trees share
7665  * same root key objectid, doing special handing for tree blocks owned
7666  * by them is easy. Once a tree block has been COWed in one reloc tree,
7667  * we can use the resulting new block directly when the same block is
7668  * required to COW again through other reloc trees. By this way, relocated
7669  * tree blocks are shared between reloc trees, so they are also shared
7670  * between subvols.
7671  */
7672 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7673                                       struct btrfs_root *root,
7674                                       struct btrfs_path *path,
7675                                       struct btrfs_key *first_key,
7676                                       struct btrfs_ref_path *ref_path,
7677                                       struct btrfs_block_group_cache *group,
7678                                       struct inode *reloc_inode)
7679 {
7680         struct btrfs_root *reloc_root;
7681         struct extent_buffer *eb = NULL;
7682         struct btrfs_key *keys;
7683         u64 *nodes;
7684         int level;
7685         int shared_level;
7686         int lowest_level = 0;
7687         int ret;
7688
7689         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7690                 lowest_level = ref_path->owner_objectid;
7691
7692         if (!root->ref_cows) {
7693                 path->lowest_level = lowest_level;
7694                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7695                 BUG_ON(ret < 0);
7696                 path->lowest_level = 0;
7697                 btrfs_release_path(root, path);
7698                 return 0;
7699         }
7700
7701         mutex_lock(&root->fs_info->tree_reloc_mutex);
7702         ret = init_reloc_tree(trans, root);
7703         BUG_ON(ret);
7704         reloc_root = root->reloc_root;
7705
7706         shared_level = ref_path->shared_level;
7707         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7708
7709         keys = ref_path->node_keys;
7710         nodes = ref_path->new_nodes;
7711         memset(&keys[shared_level + 1], 0,
7712                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7713         memset(&nodes[shared_level + 1], 0,
7714                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7715
7716         if (nodes[lowest_level] == 0) {
7717                 path->lowest_level = lowest_level;
7718                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7719                                         0, 1);
7720                 BUG_ON(ret);
7721                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7722                         eb = path->nodes[level];
7723                         if (!eb || eb == reloc_root->node)
7724                                 break;
7725                         nodes[level] = eb->start;
7726                         if (level == 0)
7727                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7728                         else
7729                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7730                 }
7731                 if (nodes[0] &&
7732                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7733                         eb = path->nodes[0];
7734                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7735                                                       group, reloc_inode);
7736                         BUG_ON(ret);
7737                 }
7738                 btrfs_release_path(reloc_root, path);
7739         } else {
7740                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7741                                        lowest_level);
7742                 BUG_ON(ret);
7743         }
7744
7745         /*
7746          * replace tree blocks in the fs tree with tree blocks in
7747          * the reloc tree.
7748          */
7749         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7750         BUG_ON(ret < 0);
7751
7752         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7753                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7754                                         0, 0);
7755                 BUG_ON(ret);
7756                 extent_buffer_get(path->nodes[0]);
7757                 eb = path->nodes[0];
7758                 btrfs_release_path(reloc_root, path);
7759                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7760                 BUG_ON(ret);
7761                 free_extent_buffer(eb);
7762         }
7763
7764         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7765         path->lowest_level = 0;
7766         return 0;
7767 }
7768
7769 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7770                                         struct btrfs_root *root,
7771                                         struct btrfs_path *path,
7772                                         struct btrfs_key *first_key,
7773                                         struct btrfs_ref_path *ref_path)
7774 {
7775         int ret;
7776
7777         ret = relocate_one_path(trans, root, path, first_key,
7778                                 ref_path, NULL, NULL);
7779         BUG_ON(ret);
7780
7781         return 0;
7782 }
7783
7784 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7785                                     struct btrfs_root *extent_root,
7786                                     struct btrfs_path *path,
7787                                     struct btrfs_key *extent_key)
7788 {
7789         int ret;
7790
7791         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7792         if (ret)
7793                 goto out;
7794         ret = btrfs_del_item(trans, extent_root, path);
7795 out:
7796         btrfs_release_path(extent_root, path);
7797         return ret;
7798 }
7799
7800 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7801                                                 struct btrfs_ref_path *ref_path)
7802 {
7803         struct btrfs_key root_key;
7804
7805         root_key.objectid = ref_path->root_objectid;
7806         root_key.type = BTRFS_ROOT_ITEM_KEY;
7807         if (is_cowonly_root(ref_path->root_objectid))
7808                 root_key.offset = 0;
7809         else
7810                 root_key.offset = (u64)-1;
7811
7812         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7813 }
7814
7815 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7816                                         struct btrfs_path *path,
7817                                         struct btrfs_key *extent_key,
7818                                         struct btrfs_block_group_cache *group,
7819                                         struct inode *reloc_inode, int pass)
7820 {
7821         struct btrfs_trans_handle *trans;
7822         struct btrfs_root *found_root;
7823         struct btrfs_ref_path *ref_path = NULL;
7824         struct disk_extent *new_extents = NULL;
7825         int nr_extents = 0;
7826         int loops;
7827         int ret;
7828         int level;
7829         struct btrfs_key first_key;
7830         u64 prev_block = 0;
7831
7832
7833         trans = btrfs_start_transaction(extent_root, 1);
7834         BUG_ON(IS_ERR(trans));
7835
7836         if (extent_key->objectid == 0) {
7837                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7838                 goto out;
7839         }
7840
7841         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7842         if (!ref_path) {
7843                 ret = -ENOMEM;
7844                 goto out;
7845         }
7846
7847         for (loops = 0; ; loops++) {
7848                 if (loops == 0) {
7849                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7850                                                    extent_key->objectid);
7851                 } else {
7852                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7853                 }
7854                 if (ret < 0)
7855                         goto out;
7856                 if (ret > 0)
7857                         break;
7858
7859                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7860                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7861                         continue;
7862
7863                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7864                 BUG_ON(!found_root);
7865                 /*
7866                  * for reference counted tree, only process reference paths
7867                  * rooted at the latest committed root.
7868                  */
7869                 if (found_root->ref_cows &&
7870                     ref_path->root_generation != found_root->root_key.offset)
7871                         continue;
7872
7873                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7874                         if (pass == 0) {
7875                                 /*
7876                                  * copy data extents to new locations
7877                                  */
7878                                 u64 group_start = group->key.objectid;
7879                                 ret = relocate_data_extent(reloc_inode,
7880                                                            extent_key,
7881                                                            group_start);
7882                                 if (ret < 0)
7883                                         goto out;
7884                                 break;
7885                         }
7886                         level = 0;
7887                 } else {
7888                         level = ref_path->owner_objectid;
7889                 }
7890
7891                 if (prev_block != ref_path->nodes[level]) {
7892                         struct extent_buffer *eb;
7893                         u64 block_start = ref_path->nodes[level];
7894                         u64 block_size = btrfs_level_size(found_root, level);
7895
7896                         eb = read_tree_block(found_root, block_start,
7897                                              block_size, 0);
7898                         btrfs_tree_lock(eb);
7899                         BUG_ON(level != btrfs_header_level(eb));
7900
7901                         if (level == 0)
7902                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7903                         else
7904                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7905
7906                         btrfs_tree_unlock(eb);
7907                         free_extent_buffer(eb);
7908                         prev_block = block_start;
7909                 }
7910
7911                 mutex_lock(&extent_root->fs_info->trans_mutex);
7912                 btrfs_record_root_in_trans(found_root);
7913                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7914                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7915                         /*
7916                          * try to update data extent references while
7917                          * keeping metadata shared between snapshots.
7918                          */
7919                         if (pass == 1) {
7920                                 ret = relocate_one_path(trans, found_root,
7921                                                 path, &first_key, ref_path,
7922                                                 group, reloc_inode);
7923                                 if (ret < 0)
7924                                         goto out;
7925                                 continue;
7926                         }
7927                         /*
7928                          * use fallback method to process the remaining
7929                          * references.
7930                          */
7931                         if (!new_extents) {
7932                                 u64 group_start = group->key.objectid;
7933                                 new_extents = kmalloc(sizeof(*new_extents),
7934                                                       GFP_NOFS);
7935                                 nr_extents = 1;
7936                                 ret = get_new_locations(reloc_inode,
7937                                                         extent_key,
7938                                                         group_start, 1,
7939                                                         &new_extents,
7940                                                         &nr_extents);
7941                                 if (ret)
7942                                         goto out;
7943                         }
7944                         ret = replace_one_extent(trans, found_root,
7945                                                 path, extent_key,
7946                                                 &first_key, ref_path,
7947                                                 new_extents, nr_extents);
7948                 } else {
7949                         ret = relocate_tree_block(trans, found_root, path,
7950                                                   &first_key, ref_path);
7951                 }
7952                 if (ret < 0)
7953                         goto out;
7954         }
7955         ret = 0;
7956 out:
7957         btrfs_end_transaction(trans, extent_root);
7958         kfree(new_extents);
7959         kfree(ref_path);
7960         return ret;
7961 }
7962 #endif
7963
7964 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7965 {
7966         u64 num_devices;
7967         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7968                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7969
7970         /*
7971          * we add in the count of missing devices because we want
7972          * to make sure that any RAID levels on a degraded FS
7973          * continue to be honored.
7974          */
7975         num_devices = root->fs_info->fs_devices->rw_devices +
7976                 root->fs_info->fs_devices->missing_devices;
7977
7978         if (num_devices == 1) {
7979                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7980                 stripped = flags & ~stripped;
7981
7982                 /* turn raid0 into single device chunks */
7983                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7984                         return stripped;
7985
7986                 /* turn mirroring into duplication */
7987                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7988                              BTRFS_BLOCK_GROUP_RAID10))
7989                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7990                 return flags;
7991         } else {
7992                 /* they already had raid on here, just return */
7993                 if (flags & stripped)
7994                         return flags;
7995
7996                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7997                 stripped = flags & ~stripped;
7998
7999                 /* switch duplicated blocks with raid1 */
8000                 if (flags & BTRFS_BLOCK_GROUP_DUP)
8001                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
8002
8003                 /* turn single device chunks into raid0 */
8004                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8005         }
8006         return flags;
8007 }
8008
8009 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
8010 {
8011         struct btrfs_space_info *sinfo = cache->space_info;
8012         u64 num_bytes;
8013         int ret = -ENOSPC;
8014
8015         if (cache->ro)
8016                 return 0;
8017
8018         spin_lock(&sinfo->lock);
8019         spin_lock(&cache->lock);
8020         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8021                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8022
8023         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8024             sinfo->bytes_may_use + sinfo->bytes_readonly +
8025             cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
8026                 sinfo->bytes_readonly += num_bytes;
8027                 sinfo->bytes_reserved += cache->reserved_pinned;
8028                 cache->reserved_pinned = 0;
8029                 cache->ro = 1;
8030                 ret = 0;
8031         }
8032
8033         spin_unlock(&cache->lock);
8034         spin_unlock(&sinfo->lock);
8035         return ret;
8036 }
8037
8038 int btrfs_set_block_group_ro(struct btrfs_root *root,
8039                              struct btrfs_block_group_cache *cache)
8040
8041 {
8042         struct btrfs_trans_handle *trans;
8043         u64 alloc_flags;
8044         int ret;
8045
8046         BUG_ON(cache->ro);
8047
8048         trans = btrfs_join_transaction(root, 1);
8049         BUG_ON(IS_ERR(trans));
8050
8051         alloc_flags = update_block_group_flags(root, cache->flags);
8052         if (alloc_flags != cache->flags)
8053                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8054
8055         ret = set_block_group_ro(cache);
8056         if (!ret)
8057                 goto out;
8058         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8059         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8060         if (ret < 0)
8061                 goto out;
8062         ret = set_block_group_ro(cache);
8063 out:
8064         btrfs_end_transaction(trans, root);
8065         return ret;
8066 }
8067
8068 /*
8069  * helper to account the unused space of all the readonly block group in the
8070  * list. takes mirrors into account.
8071  */
8072 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8073 {
8074         struct btrfs_block_group_cache *block_group;
8075         u64 free_bytes = 0;
8076         int factor;
8077
8078         list_for_each_entry(block_group, groups_list, list) {
8079                 spin_lock(&block_group->lock);
8080
8081                 if (!block_group->ro) {
8082                         spin_unlock(&block_group->lock);
8083                         continue;
8084                 }
8085
8086                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8087                                           BTRFS_BLOCK_GROUP_RAID10 |
8088                                           BTRFS_BLOCK_GROUP_DUP))
8089                         factor = 2;
8090                 else
8091                         factor = 1;
8092
8093                 free_bytes += (block_group->key.offset -
8094                                btrfs_block_group_used(&block_group->item)) *
8095                                factor;
8096
8097                 spin_unlock(&block_group->lock);
8098         }
8099
8100         return free_bytes;
8101 }
8102
8103 /*
8104  * helper to account the unused space of all the readonly block group in the
8105  * space_info. takes mirrors into account.
8106  */
8107 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8108 {
8109         int i;
8110         u64 free_bytes = 0;
8111
8112         spin_lock(&sinfo->lock);
8113
8114         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8115                 if (!list_empty(&sinfo->block_groups[i]))
8116                         free_bytes += __btrfs_get_ro_block_group_free_space(
8117                                                 &sinfo->block_groups[i]);
8118
8119         spin_unlock(&sinfo->lock);
8120
8121         return free_bytes;
8122 }
8123
8124 int btrfs_set_block_group_rw(struct btrfs_root *root,
8125                               struct btrfs_block_group_cache *cache)
8126 {
8127         struct btrfs_space_info *sinfo = cache->space_info;
8128         u64 num_bytes;
8129
8130         BUG_ON(!cache->ro);
8131
8132         spin_lock(&sinfo->lock);
8133         spin_lock(&cache->lock);
8134         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8135                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8136         sinfo->bytes_readonly -= num_bytes;
8137         cache->ro = 0;
8138         spin_unlock(&cache->lock);
8139         spin_unlock(&sinfo->lock);
8140         return 0;
8141 }
8142
8143 /*
8144  * checks to see if its even possible to relocate this block group.
8145  *
8146  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8147  * ok to go ahead and try.
8148  */
8149 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8150 {
8151         struct btrfs_block_group_cache *block_group;
8152         struct btrfs_space_info *space_info;
8153         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8154         struct btrfs_device *device;
8155         int full = 0;
8156         int ret = 0;
8157
8158         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8159
8160         /* odd, couldn't find the block group, leave it alone */
8161         if (!block_group)
8162                 return -1;
8163
8164         /* no bytes used, we're good */
8165         if (!btrfs_block_group_used(&block_group->item))
8166                 goto out;
8167
8168         space_info = block_group->space_info;
8169         spin_lock(&space_info->lock);
8170
8171         full = space_info->full;
8172
8173         /*
8174          * if this is the last block group we have in this space, we can't
8175          * relocate it unless we're able to allocate a new chunk below.
8176          *
8177          * Otherwise, we need to make sure we have room in the space to handle
8178          * all of the extents from this block group.  If we can, we're good
8179          */
8180         if ((space_info->total_bytes != block_group->key.offset) &&
8181            (space_info->bytes_used + space_info->bytes_reserved +
8182             space_info->bytes_pinned + space_info->bytes_readonly +
8183             btrfs_block_group_used(&block_group->item) <
8184             space_info->total_bytes)) {
8185                 spin_unlock(&space_info->lock);
8186                 goto out;
8187         }
8188         spin_unlock(&space_info->lock);
8189
8190         /*
8191          * ok we don't have enough space, but maybe we have free space on our
8192          * devices to allocate new chunks for relocation, so loop through our
8193          * alloc devices and guess if we have enough space.  However, if we
8194          * were marked as full, then we know there aren't enough chunks, and we
8195          * can just return.
8196          */
8197         ret = -1;
8198         if (full)
8199                 goto out;
8200
8201         mutex_lock(&root->fs_info->chunk_mutex);
8202         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8203                 u64 min_free = btrfs_block_group_used(&block_group->item);
8204                 u64 dev_offset;
8205
8206                 /*
8207                  * check to make sure we can actually find a chunk with enough
8208                  * space to fit our block group in.
8209                  */
8210                 if (device->total_bytes > device->bytes_used + min_free) {
8211                         ret = find_free_dev_extent(NULL, device, min_free,
8212                                                    &dev_offset, NULL);
8213                         if (!ret)
8214                                 break;
8215                         ret = -1;
8216                 }
8217         }
8218         mutex_unlock(&root->fs_info->chunk_mutex);
8219 out:
8220         btrfs_put_block_group(block_group);
8221         return ret;
8222 }
8223
8224 static int find_first_block_group(struct btrfs_root *root,
8225                 struct btrfs_path *path, struct btrfs_key *key)
8226 {
8227         int ret = 0;
8228         struct btrfs_key found_key;
8229         struct extent_buffer *leaf;
8230         int slot;
8231
8232         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8233         if (ret < 0)
8234                 goto out;
8235
8236         while (1) {
8237                 slot = path->slots[0];
8238                 leaf = path->nodes[0];
8239                 if (slot >= btrfs_header_nritems(leaf)) {
8240                         ret = btrfs_next_leaf(root, path);
8241                         if (ret == 0)
8242                                 continue;
8243                         if (ret < 0)
8244                                 goto out;
8245                         break;
8246                 }
8247                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8248
8249                 if (found_key.objectid >= key->objectid &&
8250                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8251                         ret = 0;
8252                         goto out;
8253                 }
8254                 path->slots[0]++;
8255         }
8256 out:
8257         return ret;
8258 }
8259
8260 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8261 {
8262         struct btrfs_block_group_cache *block_group;
8263         u64 last = 0;
8264
8265         while (1) {
8266                 struct inode *inode;
8267
8268                 block_group = btrfs_lookup_first_block_group(info, last);
8269                 while (block_group) {
8270                         spin_lock(&block_group->lock);
8271                         if (block_group->iref)
8272                                 break;
8273                         spin_unlock(&block_group->lock);
8274                         block_group = next_block_group(info->tree_root,
8275                                                        block_group);
8276                 }
8277                 if (!block_group) {
8278                         if (last == 0)
8279                                 break;
8280                         last = 0;
8281                         continue;
8282                 }
8283
8284                 inode = block_group->inode;
8285                 block_group->iref = 0;
8286                 block_group->inode = NULL;
8287                 spin_unlock(&block_group->lock);
8288                 iput(inode);
8289                 last = block_group->key.objectid + block_group->key.offset;
8290                 btrfs_put_block_group(block_group);
8291         }
8292 }
8293
8294 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8295 {
8296         struct btrfs_block_group_cache *block_group;
8297         struct btrfs_space_info *space_info;
8298         struct btrfs_caching_control *caching_ctl;
8299         struct rb_node *n;
8300
8301         down_write(&info->extent_commit_sem);
8302         while (!list_empty(&info->caching_block_groups)) {
8303                 caching_ctl = list_entry(info->caching_block_groups.next,
8304                                          struct btrfs_caching_control, list);
8305                 list_del(&caching_ctl->list);
8306                 put_caching_control(caching_ctl);
8307         }
8308         up_write(&info->extent_commit_sem);
8309
8310         spin_lock(&info->block_group_cache_lock);
8311         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8312                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8313                                        cache_node);
8314                 rb_erase(&block_group->cache_node,
8315                          &info->block_group_cache_tree);
8316                 spin_unlock(&info->block_group_cache_lock);
8317
8318                 down_write(&block_group->space_info->groups_sem);
8319                 list_del(&block_group->list);
8320                 up_write(&block_group->space_info->groups_sem);
8321
8322                 if (block_group->cached == BTRFS_CACHE_STARTED)
8323                         wait_block_group_cache_done(block_group);
8324
8325                 /*
8326                  * We haven't cached this block group, which means we could
8327                  * possibly have excluded extents on this block group.
8328                  */
8329                 if (block_group->cached == BTRFS_CACHE_NO)
8330                         free_excluded_extents(info->extent_root, block_group);
8331
8332                 btrfs_remove_free_space_cache(block_group);
8333                 btrfs_put_block_group(block_group);
8334
8335                 spin_lock(&info->block_group_cache_lock);
8336         }
8337         spin_unlock(&info->block_group_cache_lock);
8338
8339         /* now that all the block groups are freed, go through and
8340          * free all the space_info structs.  This is only called during
8341          * the final stages of unmount, and so we know nobody is
8342          * using them.  We call synchronize_rcu() once before we start,
8343          * just to be on the safe side.
8344          */
8345         synchronize_rcu();
8346
8347         release_global_block_rsv(info);
8348
8349         while(!list_empty(&info->space_info)) {
8350                 space_info = list_entry(info->space_info.next,
8351                                         struct btrfs_space_info,
8352                                         list);
8353                 if (space_info->bytes_pinned > 0 ||
8354                     space_info->bytes_reserved > 0) {
8355                         WARN_ON(1);
8356                         dump_space_info(space_info, 0, 0);
8357                 }
8358                 list_del(&space_info->list);
8359                 kfree(space_info);
8360         }
8361         return 0;
8362 }
8363
8364 static void __link_block_group(struct btrfs_space_info *space_info,
8365                                struct btrfs_block_group_cache *cache)
8366 {
8367         int index = get_block_group_index(cache);
8368
8369         down_write(&space_info->groups_sem);
8370         list_add_tail(&cache->list, &space_info->block_groups[index]);
8371         up_write(&space_info->groups_sem);
8372 }
8373
8374 int btrfs_read_block_groups(struct btrfs_root *root)
8375 {
8376         struct btrfs_path *path;
8377         int ret;
8378         struct btrfs_block_group_cache *cache;
8379         struct btrfs_fs_info *info = root->fs_info;
8380         struct btrfs_space_info *space_info;
8381         struct btrfs_key key;
8382         struct btrfs_key found_key;
8383         struct extent_buffer *leaf;
8384         int need_clear = 0;
8385         u64 cache_gen;
8386
8387         root = info->extent_root;
8388         key.objectid = 0;
8389         key.offset = 0;
8390         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8391         path = btrfs_alloc_path();
8392         if (!path)
8393                 return -ENOMEM;
8394
8395         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8396         if (cache_gen != 0 &&
8397             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8398                 need_clear = 1;
8399         if (btrfs_test_opt(root, CLEAR_CACHE))
8400                 need_clear = 1;
8401         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8402                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8403
8404         while (1) {
8405                 ret = find_first_block_group(root, path, &key);
8406                 if (ret > 0)
8407                         break;
8408                 if (ret != 0)
8409                         goto error;
8410                 leaf = path->nodes[0];
8411                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8412                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8413                 if (!cache) {
8414                         ret = -ENOMEM;
8415                         goto error;
8416                 }
8417
8418                 atomic_set(&cache->count, 1);
8419                 spin_lock_init(&cache->lock);
8420                 spin_lock_init(&cache->tree_lock);
8421                 cache->fs_info = info;
8422                 INIT_LIST_HEAD(&cache->list);
8423                 INIT_LIST_HEAD(&cache->cluster_list);
8424
8425                 if (need_clear)
8426                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8427
8428                 /*
8429                  * we only want to have 32k of ram per block group for keeping
8430                  * track of free space, and if we pass 1/2 of that we want to
8431                  * start converting things over to using bitmaps
8432                  */
8433                 cache->extents_thresh = ((1024 * 32) / 2) /
8434                         sizeof(struct btrfs_free_space);
8435
8436                 read_extent_buffer(leaf, &cache->item,
8437                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8438                                    sizeof(cache->item));
8439                 memcpy(&cache->key, &found_key, sizeof(found_key));
8440
8441                 key.objectid = found_key.objectid + found_key.offset;
8442                 btrfs_release_path(root, path);
8443                 cache->flags = btrfs_block_group_flags(&cache->item);
8444                 cache->sectorsize = root->sectorsize;
8445
8446                 /*
8447                  * We need to exclude the super stripes now so that the space
8448                  * info has super bytes accounted for, otherwise we'll think
8449                  * we have more space than we actually do.
8450                  */
8451                 exclude_super_stripes(root, cache);
8452
8453                 /*
8454                  * check for two cases, either we are full, and therefore
8455                  * don't need to bother with the caching work since we won't
8456                  * find any space, or we are empty, and we can just add all
8457                  * the space in and be done with it.  This saves us _alot_ of
8458                  * time, particularly in the full case.
8459                  */
8460                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8461                         cache->last_byte_to_unpin = (u64)-1;
8462                         cache->cached = BTRFS_CACHE_FINISHED;
8463                         free_excluded_extents(root, cache);
8464                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8465                         cache->last_byte_to_unpin = (u64)-1;
8466                         cache->cached = BTRFS_CACHE_FINISHED;
8467                         add_new_free_space(cache, root->fs_info,
8468                                            found_key.objectid,
8469                                            found_key.objectid +
8470                                            found_key.offset);
8471                         free_excluded_extents(root, cache);
8472                 }
8473
8474                 ret = update_space_info(info, cache->flags, found_key.offset,
8475                                         btrfs_block_group_used(&cache->item),
8476                                         &space_info);
8477                 BUG_ON(ret);
8478                 cache->space_info = space_info;
8479                 spin_lock(&cache->space_info->lock);
8480                 cache->space_info->bytes_readonly += cache->bytes_super;
8481                 spin_unlock(&cache->space_info->lock);
8482
8483                 __link_block_group(space_info, cache);
8484
8485                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8486                 BUG_ON(ret);
8487
8488                 set_avail_alloc_bits(root->fs_info, cache->flags);
8489                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8490                         set_block_group_ro(cache);
8491         }
8492
8493         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8494                 if (!(get_alloc_profile(root, space_info->flags) &
8495                       (BTRFS_BLOCK_GROUP_RAID10 |
8496                        BTRFS_BLOCK_GROUP_RAID1 |
8497                        BTRFS_BLOCK_GROUP_DUP)))
8498                         continue;
8499                 /*
8500                  * avoid allocating from un-mirrored block group if there are
8501                  * mirrored block groups.
8502                  */
8503                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8504                         set_block_group_ro(cache);
8505                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8506                         set_block_group_ro(cache);
8507         }
8508
8509         init_global_block_rsv(info);
8510         ret = 0;
8511 error:
8512         btrfs_free_path(path);
8513         return ret;
8514 }
8515
8516 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8517                            struct btrfs_root *root, u64 bytes_used,
8518                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8519                            u64 size)
8520 {
8521         int ret;
8522         struct btrfs_root *extent_root;
8523         struct btrfs_block_group_cache *cache;
8524
8525         extent_root = root->fs_info->extent_root;
8526
8527         root->fs_info->last_trans_log_full_commit = trans->transid;
8528
8529         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8530         if (!cache)
8531                 return -ENOMEM;
8532
8533         cache->key.objectid = chunk_offset;
8534         cache->key.offset = size;
8535         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8536         cache->sectorsize = root->sectorsize;
8537         cache->fs_info = root->fs_info;
8538
8539         /*
8540          * we only want to have 32k of ram per block group for keeping track
8541          * of free space, and if we pass 1/2 of that we want to start
8542          * converting things over to using bitmaps
8543          */
8544         cache->extents_thresh = ((1024 * 32) / 2) /
8545                 sizeof(struct btrfs_free_space);
8546         atomic_set(&cache->count, 1);
8547         spin_lock_init(&cache->lock);
8548         spin_lock_init(&cache->tree_lock);
8549         INIT_LIST_HEAD(&cache->list);
8550         INIT_LIST_HEAD(&cache->cluster_list);
8551
8552         btrfs_set_block_group_used(&cache->item, bytes_used);
8553         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8554         cache->flags = type;
8555         btrfs_set_block_group_flags(&cache->item, type);
8556
8557         cache->last_byte_to_unpin = (u64)-1;
8558         cache->cached = BTRFS_CACHE_FINISHED;
8559         exclude_super_stripes(root, cache);
8560
8561         add_new_free_space(cache, root->fs_info, chunk_offset,
8562                            chunk_offset + size);
8563
8564         free_excluded_extents(root, cache);
8565
8566         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8567                                 &cache->space_info);
8568         BUG_ON(ret);
8569
8570         spin_lock(&cache->space_info->lock);
8571         cache->space_info->bytes_readonly += cache->bytes_super;
8572         spin_unlock(&cache->space_info->lock);
8573
8574         __link_block_group(cache->space_info, cache);
8575
8576         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8577         BUG_ON(ret);
8578
8579         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8580                                 sizeof(cache->item));
8581         BUG_ON(ret);
8582
8583         set_avail_alloc_bits(extent_root->fs_info, type);
8584
8585         return 0;
8586 }
8587
8588 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8589                              struct btrfs_root *root, u64 group_start)
8590 {
8591         struct btrfs_path *path;
8592         struct btrfs_block_group_cache *block_group;
8593         struct btrfs_free_cluster *cluster;
8594         struct btrfs_root *tree_root = root->fs_info->tree_root;
8595         struct btrfs_key key;
8596         struct inode *inode;
8597         int ret;
8598         int factor;
8599
8600         root = root->fs_info->extent_root;
8601
8602         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8603         BUG_ON(!block_group);
8604         BUG_ON(!block_group->ro);
8605
8606         memcpy(&key, &block_group->key, sizeof(key));
8607         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8608                                   BTRFS_BLOCK_GROUP_RAID1 |
8609                                   BTRFS_BLOCK_GROUP_RAID10))
8610                 factor = 2;
8611         else
8612                 factor = 1;
8613
8614         /* make sure this block group isn't part of an allocation cluster */
8615         cluster = &root->fs_info->data_alloc_cluster;
8616         spin_lock(&cluster->refill_lock);
8617         btrfs_return_cluster_to_free_space(block_group, cluster);
8618         spin_unlock(&cluster->refill_lock);
8619
8620         /*
8621          * make sure this block group isn't part of a metadata
8622          * allocation cluster
8623          */
8624         cluster = &root->fs_info->meta_alloc_cluster;
8625         spin_lock(&cluster->refill_lock);
8626         btrfs_return_cluster_to_free_space(block_group, cluster);
8627         spin_unlock(&cluster->refill_lock);
8628
8629         path = btrfs_alloc_path();
8630         BUG_ON(!path);
8631
8632         inode = lookup_free_space_inode(root, block_group, path);
8633         if (!IS_ERR(inode)) {
8634                 btrfs_orphan_add(trans, inode);
8635                 clear_nlink(inode);
8636                 /* One for the block groups ref */
8637                 spin_lock(&block_group->lock);
8638                 if (block_group->iref) {
8639                         block_group->iref = 0;
8640                         block_group->inode = NULL;
8641                         spin_unlock(&block_group->lock);
8642                         iput(inode);
8643                 } else {
8644                         spin_unlock(&block_group->lock);
8645                 }
8646                 /* One for our lookup ref */
8647                 iput(inode);
8648         }
8649
8650         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8651         key.offset = block_group->key.objectid;
8652         key.type = 0;
8653
8654         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8655         if (ret < 0)
8656                 goto out;
8657         if (ret > 0)
8658                 btrfs_release_path(tree_root, path);
8659         if (ret == 0) {
8660                 ret = btrfs_del_item(trans, tree_root, path);
8661                 if (ret)
8662                         goto out;
8663                 btrfs_release_path(tree_root, path);
8664         }
8665
8666         spin_lock(&root->fs_info->block_group_cache_lock);
8667         rb_erase(&block_group->cache_node,
8668                  &root->fs_info->block_group_cache_tree);
8669         spin_unlock(&root->fs_info->block_group_cache_lock);
8670
8671         down_write(&block_group->space_info->groups_sem);
8672         /*
8673          * we must use list_del_init so people can check to see if they
8674          * are still on the list after taking the semaphore
8675          */
8676         list_del_init(&block_group->list);
8677         up_write(&block_group->space_info->groups_sem);
8678
8679         if (block_group->cached == BTRFS_CACHE_STARTED)
8680                 wait_block_group_cache_done(block_group);
8681
8682         btrfs_remove_free_space_cache(block_group);
8683
8684         spin_lock(&block_group->space_info->lock);
8685         block_group->space_info->total_bytes -= block_group->key.offset;
8686         block_group->space_info->bytes_readonly -= block_group->key.offset;
8687         block_group->space_info->disk_total -= block_group->key.offset * factor;
8688         spin_unlock(&block_group->space_info->lock);
8689
8690         memcpy(&key, &block_group->key, sizeof(key));
8691
8692         btrfs_clear_space_info_full(root->fs_info);
8693
8694         btrfs_put_block_group(block_group);
8695         btrfs_put_block_group(block_group);
8696
8697         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8698         if (ret > 0)
8699                 ret = -EIO;
8700         if (ret < 0)
8701                 goto out;
8702
8703         ret = btrfs_del_item(trans, root, path);
8704 out:
8705         btrfs_free_path(path);
8706         return ret;
8707 }
8708
8709 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8710 {
8711         return unpin_extent_range(root, start, end);
8712 }
8713
8714 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8715                                u64 num_bytes)
8716 {
8717         return btrfs_discard_extent(root, bytenr, num_bytes);
8718 }