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,
1747                         BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
1748 }
1749
1750 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1751                                 u64 num_bytes)
1752 {
1753         int ret;
1754         u64 map_length = num_bytes;
1755         struct btrfs_multi_bio *multi = NULL;
1756
1757         if (!btrfs_test_opt(root, DISCARD))
1758                 return 0;
1759
1760         /* Tell the block device(s) that the sectors can be discarded */
1761         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1762                               bytenr, &map_length, &multi, 0);
1763         if (!ret) {
1764                 struct btrfs_bio_stripe *stripe = multi->stripes;
1765                 int i;
1766
1767                 if (map_length > num_bytes)
1768                         map_length = num_bytes;
1769
1770                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1771                         btrfs_issue_discard(stripe->dev->bdev,
1772                                             stripe->physical,
1773                                             map_length);
1774                 }
1775                 kfree(multi);
1776         }
1777
1778         return ret;
1779 }
1780
1781 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1782                          struct btrfs_root *root,
1783                          u64 bytenr, u64 num_bytes, u64 parent,
1784                          u64 root_objectid, u64 owner, u64 offset)
1785 {
1786         int ret;
1787         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1788                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1789
1790         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1791                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1792                                         parent, root_objectid, (int)owner,
1793                                         BTRFS_ADD_DELAYED_REF, NULL);
1794         } else {
1795                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1796                                         parent, root_objectid, owner, offset,
1797                                         BTRFS_ADD_DELAYED_REF, NULL);
1798         }
1799         return ret;
1800 }
1801
1802 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1803                                   struct btrfs_root *root,
1804                                   u64 bytenr, u64 num_bytes,
1805                                   u64 parent, u64 root_objectid,
1806                                   u64 owner, u64 offset, int refs_to_add,
1807                                   struct btrfs_delayed_extent_op *extent_op)
1808 {
1809         struct btrfs_path *path;
1810         struct extent_buffer *leaf;
1811         struct btrfs_extent_item *item;
1812         u64 refs;
1813         int ret;
1814         int err = 0;
1815
1816         path = btrfs_alloc_path();
1817         if (!path)
1818                 return -ENOMEM;
1819
1820         path->reada = 1;
1821         path->leave_spinning = 1;
1822         /* this will setup the path even if it fails to insert the back ref */
1823         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1824                                            path, bytenr, num_bytes, parent,
1825                                            root_objectid, owner, offset,
1826                                            refs_to_add, extent_op);
1827         if (ret == 0)
1828                 goto out;
1829
1830         if (ret != -EAGAIN) {
1831                 err = ret;
1832                 goto out;
1833         }
1834
1835         leaf = path->nodes[0];
1836         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1837         refs = btrfs_extent_refs(leaf, item);
1838         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1839         if (extent_op)
1840                 __run_delayed_extent_op(extent_op, leaf, item);
1841
1842         btrfs_mark_buffer_dirty(leaf);
1843         btrfs_release_path(root->fs_info->extent_root, path);
1844
1845         path->reada = 1;
1846         path->leave_spinning = 1;
1847
1848         /* now insert the actual backref */
1849         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1850                                     path, bytenr, parent, root_objectid,
1851                                     owner, offset, refs_to_add);
1852         BUG_ON(ret);
1853 out:
1854         btrfs_free_path(path);
1855         return err;
1856 }
1857
1858 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1859                                 struct btrfs_root *root,
1860                                 struct btrfs_delayed_ref_node *node,
1861                                 struct btrfs_delayed_extent_op *extent_op,
1862                                 int insert_reserved)
1863 {
1864         int ret = 0;
1865         struct btrfs_delayed_data_ref *ref;
1866         struct btrfs_key ins;
1867         u64 parent = 0;
1868         u64 ref_root = 0;
1869         u64 flags = 0;
1870
1871         ins.objectid = node->bytenr;
1872         ins.offset = node->num_bytes;
1873         ins.type = BTRFS_EXTENT_ITEM_KEY;
1874
1875         ref = btrfs_delayed_node_to_data_ref(node);
1876         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1877                 parent = ref->parent;
1878         else
1879                 ref_root = ref->root;
1880
1881         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1882                 if (extent_op) {
1883                         BUG_ON(extent_op->update_key);
1884                         flags |= extent_op->flags_to_set;
1885                 }
1886                 ret = alloc_reserved_file_extent(trans, root,
1887                                                  parent, ref_root, flags,
1888                                                  ref->objectid, ref->offset,
1889                                                  &ins, node->ref_mod);
1890         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1891                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1892                                              node->num_bytes, parent,
1893                                              ref_root, ref->objectid,
1894                                              ref->offset, node->ref_mod,
1895                                              extent_op);
1896         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1897                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1898                                           node->num_bytes, parent,
1899                                           ref_root, ref->objectid,
1900                                           ref->offset, node->ref_mod,
1901                                           extent_op);
1902         } else {
1903                 BUG();
1904         }
1905         return ret;
1906 }
1907
1908 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1909                                     struct extent_buffer *leaf,
1910                                     struct btrfs_extent_item *ei)
1911 {
1912         u64 flags = btrfs_extent_flags(leaf, ei);
1913         if (extent_op->update_flags) {
1914                 flags |= extent_op->flags_to_set;
1915                 btrfs_set_extent_flags(leaf, ei, flags);
1916         }
1917
1918         if (extent_op->update_key) {
1919                 struct btrfs_tree_block_info *bi;
1920                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1921                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1922                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1923         }
1924 }
1925
1926 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1927                                  struct btrfs_root *root,
1928                                  struct btrfs_delayed_ref_node *node,
1929                                  struct btrfs_delayed_extent_op *extent_op)
1930 {
1931         struct btrfs_key key;
1932         struct btrfs_path *path;
1933         struct btrfs_extent_item *ei;
1934         struct extent_buffer *leaf;
1935         u32 item_size;
1936         int ret;
1937         int err = 0;
1938
1939         path = btrfs_alloc_path();
1940         if (!path)
1941                 return -ENOMEM;
1942
1943         key.objectid = node->bytenr;
1944         key.type = BTRFS_EXTENT_ITEM_KEY;
1945         key.offset = node->num_bytes;
1946
1947         path->reada = 1;
1948         path->leave_spinning = 1;
1949         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1950                                 path, 0, 1);
1951         if (ret < 0) {
1952                 err = ret;
1953                 goto out;
1954         }
1955         if (ret > 0) {
1956                 err = -EIO;
1957                 goto out;
1958         }
1959
1960         leaf = path->nodes[0];
1961         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1962 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1963         if (item_size < sizeof(*ei)) {
1964                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1965                                              path, (u64)-1, 0);
1966                 if (ret < 0) {
1967                         err = ret;
1968                         goto out;
1969                 }
1970                 leaf = path->nodes[0];
1971                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1972         }
1973 #endif
1974         BUG_ON(item_size < sizeof(*ei));
1975         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1976         __run_delayed_extent_op(extent_op, leaf, ei);
1977
1978         btrfs_mark_buffer_dirty(leaf);
1979 out:
1980         btrfs_free_path(path);
1981         return err;
1982 }
1983
1984 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1985                                 struct btrfs_root *root,
1986                                 struct btrfs_delayed_ref_node *node,
1987                                 struct btrfs_delayed_extent_op *extent_op,
1988                                 int insert_reserved)
1989 {
1990         int ret = 0;
1991         struct btrfs_delayed_tree_ref *ref;
1992         struct btrfs_key ins;
1993         u64 parent = 0;
1994         u64 ref_root = 0;
1995
1996         ins.objectid = node->bytenr;
1997         ins.offset = node->num_bytes;
1998         ins.type = BTRFS_EXTENT_ITEM_KEY;
1999
2000         ref = btrfs_delayed_node_to_tree_ref(node);
2001         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2002                 parent = ref->parent;
2003         else
2004                 ref_root = ref->root;
2005
2006         BUG_ON(node->ref_mod != 1);
2007         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2008                 BUG_ON(!extent_op || !extent_op->update_flags ||
2009                        !extent_op->update_key);
2010                 ret = alloc_reserved_tree_block(trans, root,
2011                                                 parent, ref_root,
2012                                                 extent_op->flags_to_set,
2013                                                 &extent_op->key,
2014                                                 ref->level, &ins);
2015         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2016                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2017                                              node->num_bytes, parent, ref_root,
2018                                              ref->level, 0, 1, extent_op);
2019         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2020                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2021                                           node->num_bytes, parent, ref_root,
2022                                           ref->level, 0, 1, extent_op);
2023         } else {
2024                 BUG();
2025         }
2026         return ret;
2027 }
2028
2029 /* helper function to actually process a single delayed ref entry */
2030 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2031                                struct btrfs_root *root,
2032                                struct btrfs_delayed_ref_node *node,
2033                                struct btrfs_delayed_extent_op *extent_op,
2034                                int insert_reserved)
2035 {
2036         int ret;
2037         if (btrfs_delayed_ref_is_head(node)) {
2038                 struct btrfs_delayed_ref_head *head;
2039                 /*
2040                  * we've hit the end of the chain and we were supposed
2041                  * to insert this extent into the tree.  But, it got
2042                  * deleted before we ever needed to insert it, so all
2043                  * we have to do is clean up the accounting
2044                  */
2045                 BUG_ON(extent_op);
2046                 head = btrfs_delayed_node_to_head(node);
2047                 if (insert_reserved) {
2048                         btrfs_pin_extent(root, node->bytenr,
2049                                          node->num_bytes, 1);
2050                         if (head->is_data) {
2051                                 ret = btrfs_del_csums(trans, root,
2052                                                       node->bytenr,
2053                                                       node->num_bytes);
2054                                 BUG_ON(ret);
2055                         }
2056                 }
2057                 mutex_unlock(&head->mutex);
2058                 return 0;
2059         }
2060
2061         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2062             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2063                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2064                                            insert_reserved);
2065         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2066                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2067                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2068                                            insert_reserved);
2069         else
2070                 BUG();
2071         return ret;
2072 }
2073
2074 static noinline struct btrfs_delayed_ref_node *
2075 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2076 {
2077         struct rb_node *node;
2078         struct btrfs_delayed_ref_node *ref;
2079         int action = BTRFS_ADD_DELAYED_REF;
2080 again:
2081         /*
2082          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2083          * this prevents ref count from going down to zero when
2084          * there still are pending delayed ref.
2085          */
2086         node = rb_prev(&head->node.rb_node);
2087         while (1) {
2088                 if (!node)
2089                         break;
2090                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2091                                 rb_node);
2092                 if (ref->bytenr != head->node.bytenr)
2093                         break;
2094                 if (ref->action == action)
2095                         return ref;
2096                 node = rb_prev(node);
2097         }
2098         if (action == BTRFS_ADD_DELAYED_REF) {
2099                 action = BTRFS_DROP_DELAYED_REF;
2100                 goto again;
2101         }
2102         return NULL;
2103 }
2104
2105 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2106                                        struct btrfs_root *root,
2107                                        struct list_head *cluster)
2108 {
2109         struct btrfs_delayed_ref_root *delayed_refs;
2110         struct btrfs_delayed_ref_node *ref;
2111         struct btrfs_delayed_ref_head *locked_ref = NULL;
2112         struct btrfs_delayed_extent_op *extent_op;
2113         int ret;
2114         int count = 0;
2115         int must_insert_reserved = 0;
2116
2117         delayed_refs = &trans->transaction->delayed_refs;
2118         while (1) {
2119                 if (!locked_ref) {
2120                         /* pick a new head ref from the cluster list */
2121                         if (list_empty(cluster))
2122                                 break;
2123
2124                         locked_ref = list_entry(cluster->next,
2125                                      struct btrfs_delayed_ref_head, cluster);
2126
2127                         /* grab the lock that says we are going to process
2128                          * all the refs for this head */
2129                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2130
2131                         /*
2132                          * we may have dropped the spin lock to get the head
2133                          * mutex lock, and that might have given someone else
2134                          * time to free the head.  If that's true, it has been
2135                          * removed from our list and we can move on.
2136                          */
2137                         if (ret == -EAGAIN) {
2138                                 locked_ref = NULL;
2139                                 count++;
2140                                 continue;
2141                         }
2142                 }
2143
2144                 /*
2145                  * record the must insert reserved flag before we
2146                  * drop the spin lock.
2147                  */
2148                 must_insert_reserved = locked_ref->must_insert_reserved;
2149                 locked_ref->must_insert_reserved = 0;
2150
2151                 extent_op = locked_ref->extent_op;
2152                 locked_ref->extent_op = NULL;
2153
2154                 /*
2155                  * locked_ref is the head node, so we have to go one
2156                  * node back for any delayed ref updates
2157                  */
2158                 ref = select_delayed_ref(locked_ref);
2159                 if (!ref) {
2160                         /* All delayed refs have been processed, Go ahead
2161                          * and send the head node to run_one_delayed_ref,
2162                          * so that any accounting fixes can happen
2163                          */
2164                         ref = &locked_ref->node;
2165
2166                         if (extent_op && must_insert_reserved) {
2167                                 kfree(extent_op);
2168                                 extent_op = NULL;
2169                         }
2170
2171                         if (extent_op) {
2172                                 spin_unlock(&delayed_refs->lock);
2173
2174                                 ret = run_delayed_extent_op(trans, root,
2175                                                             ref, extent_op);
2176                                 BUG_ON(ret);
2177                                 kfree(extent_op);
2178
2179                                 cond_resched();
2180                                 spin_lock(&delayed_refs->lock);
2181                                 continue;
2182                         }
2183
2184                         list_del_init(&locked_ref->cluster);
2185                         locked_ref = NULL;
2186                 }
2187
2188                 ref->in_tree = 0;
2189                 rb_erase(&ref->rb_node, &delayed_refs->root);
2190                 delayed_refs->num_entries--;
2191
2192                 spin_unlock(&delayed_refs->lock);
2193
2194                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2195                                           must_insert_reserved);
2196                 BUG_ON(ret);
2197
2198                 btrfs_put_delayed_ref(ref);
2199                 kfree(extent_op);
2200                 count++;
2201
2202                 cond_resched();
2203                 spin_lock(&delayed_refs->lock);
2204         }
2205         return count;
2206 }
2207
2208 /*
2209  * this starts processing the delayed reference count updates and
2210  * extent insertions we have queued up so far.  count can be
2211  * 0, which means to process everything in the tree at the start
2212  * of the run (but not newly added entries), or it can be some target
2213  * number you'd like to process.
2214  */
2215 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2216                            struct btrfs_root *root, unsigned long count)
2217 {
2218         struct rb_node *node;
2219         struct btrfs_delayed_ref_root *delayed_refs;
2220         struct btrfs_delayed_ref_node *ref;
2221         struct list_head cluster;
2222         int ret;
2223         int run_all = count == (unsigned long)-1;
2224         int run_most = 0;
2225
2226         if (root == root->fs_info->extent_root)
2227                 root = root->fs_info->tree_root;
2228
2229         delayed_refs = &trans->transaction->delayed_refs;
2230         INIT_LIST_HEAD(&cluster);
2231 again:
2232         spin_lock(&delayed_refs->lock);
2233         if (count == 0) {
2234                 count = delayed_refs->num_entries * 2;
2235                 run_most = 1;
2236         }
2237         while (1) {
2238                 if (!(run_all || run_most) &&
2239                     delayed_refs->num_heads_ready < 64)
2240                         break;
2241
2242                 /*
2243                  * go find something we can process in the rbtree.  We start at
2244                  * the beginning of the tree, and then build a cluster
2245                  * of refs to process starting at the first one we are able to
2246                  * lock
2247                  */
2248                 ret = btrfs_find_ref_cluster(trans, &cluster,
2249                                              delayed_refs->run_delayed_start);
2250                 if (ret)
2251                         break;
2252
2253                 ret = run_clustered_refs(trans, root, &cluster);
2254                 BUG_ON(ret < 0);
2255
2256                 count -= min_t(unsigned long, ret, count);
2257
2258                 if (count == 0)
2259                         break;
2260         }
2261
2262         if (run_all) {
2263                 node = rb_first(&delayed_refs->root);
2264                 if (!node)
2265                         goto out;
2266                 count = (unsigned long)-1;
2267
2268                 while (node) {
2269                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2270                                        rb_node);
2271                         if (btrfs_delayed_ref_is_head(ref)) {
2272                                 struct btrfs_delayed_ref_head *head;
2273
2274                                 head = btrfs_delayed_node_to_head(ref);
2275                                 atomic_inc(&ref->refs);
2276
2277                                 spin_unlock(&delayed_refs->lock);
2278                                 mutex_lock(&head->mutex);
2279                                 mutex_unlock(&head->mutex);
2280
2281                                 btrfs_put_delayed_ref(ref);
2282                                 cond_resched();
2283                                 goto again;
2284                         }
2285                         node = rb_next(node);
2286                 }
2287                 spin_unlock(&delayed_refs->lock);
2288                 schedule_timeout(1);
2289                 goto again;
2290         }
2291 out:
2292         spin_unlock(&delayed_refs->lock);
2293         return 0;
2294 }
2295
2296 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2297                                 struct btrfs_root *root,
2298                                 u64 bytenr, u64 num_bytes, u64 flags,
2299                                 int is_data)
2300 {
2301         struct btrfs_delayed_extent_op *extent_op;
2302         int ret;
2303
2304         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2305         if (!extent_op)
2306                 return -ENOMEM;
2307
2308         extent_op->flags_to_set = flags;
2309         extent_op->update_flags = 1;
2310         extent_op->update_key = 0;
2311         extent_op->is_data = is_data ? 1 : 0;
2312
2313         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2314         if (ret)
2315                 kfree(extent_op);
2316         return ret;
2317 }
2318
2319 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2320                                       struct btrfs_root *root,
2321                                       struct btrfs_path *path,
2322                                       u64 objectid, u64 offset, u64 bytenr)
2323 {
2324         struct btrfs_delayed_ref_head *head;
2325         struct btrfs_delayed_ref_node *ref;
2326         struct btrfs_delayed_data_ref *data_ref;
2327         struct btrfs_delayed_ref_root *delayed_refs;
2328         struct rb_node *node;
2329         int ret = 0;
2330
2331         ret = -ENOENT;
2332         delayed_refs = &trans->transaction->delayed_refs;
2333         spin_lock(&delayed_refs->lock);
2334         head = btrfs_find_delayed_ref_head(trans, bytenr);
2335         if (!head)
2336                 goto out;
2337
2338         if (!mutex_trylock(&head->mutex)) {
2339                 atomic_inc(&head->node.refs);
2340                 spin_unlock(&delayed_refs->lock);
2341
2342                 btrfs_release_path(root->fs_info->extent_root, path);
2343
2344                 mutex_lock(&head->mutex);
2345                 mutex_unlock(&head->mutex);
2346                 btrfs_put_delayed_ref(&head->node);
2347                 return -EAGAIN;
2348         }
2349
2350         node = rb_prev(&head->node.rb_node);
2351         if (!node)
2352                 goto out_unlock;
2353
2354         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2355
2356         if (ref->bytenr != bytenr)
2357                 goto out_unlock;
2358
2359         ret = 1;
2360         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2361                 goto out_unlock;
2362
2363         data_ref = btrfs_delayed_node_to_data_ref(ref);
2364
2365         node = rb_prev(node);
2366         if (node) {
2367                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2368                 if (ref->bytenr == bytenr)
2369                         goto out_unlock;
2370         }
2371
2372         if (data_ref->root != root->root_key.objectid ||
2373             data_ref->objectid != objectid || data_ref->offset != offset)
2374                 goto out_unlock;
2375
2376         ret = 0;
2377 out_unlock:
2378         mutex_unlock(&head->mutex);
2379 out:
2380         spin_unlock(&delayed_refs->lock);
2381         return ret;
2382 }
2383
2384 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2385                                         struct btrfs_root *root,
2386                                         struct btrfs_path *path,
2387                                         u64 objectid, u64 offset, u64 bytenr)
2388 {
2389         struct btrfs_root *extent_root = root->fs_info->extent_root;
2390         struct extent_buffer *leaf;
2391         struct btrfs_extent_data_ref *ref;
2392         struct btrfs_extent_inline_ref *iref;
2393         struct btrfs_extent_item *ei;
2394         struct btrfs_key key;
2395         u32 item_size;
2396         int ret;
2397
2398         key.objectid = bytenr;
2399         key.offset = (u64)-1;
2400         key.type = BTRFS_EXTENT_ITEM_KEY;
2401
2402         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2403         if (ret < 0)
2404                 goto out;
2405         BUG_ON(ret == 0);
2406
2407         ret = -ENOENT;
2408         if (path->slots[0] == 0)
2409                 goto out;
2410
2411         path->slots[0]--;
2412         leaf = path->nodes[0];
2413         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2414
2415         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2416                 goto out;
2417
2418         ret = 1;
2419         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2420 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2421         if (item_size < sizeof(*ei)) {
2422                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2423                 goto out;
2424         }
2425 #endif
2426         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2427
2428         if (item_size != sizeof(*ei) +
2429             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2430                 goto out;
2431
2432         if (btrfs_extent_generation(leaf, ei) <=
2433             btrfs_root_last_snapshot(&root->root_item))
2434                 goto out;
2435
2436         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2437         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2438             BTRFS_EXTENT_DATA_REF_KEY)
2439                 goto out;
2440
2441         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2442         if (btrfs_extent_refs(leaf, ei) !=
2443             btrfs_extent_data_ref_count(leaf, ref) ||
2444             btrfs_extent_data_ref_root(leaf, ref) !=
2445             root->root_key.objectid ||
2446             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2447             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2448                 goto out;
2449
2450         ret = 0;
2451 out:
2452         return ret;
2453 }
2454
2455 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2456                           struct btrfs_root *root,
2457                           u64 objectid, u64 offset, u64 bytenr)
2458 {
2459         struct btrfs_path *path;
2460         int ret;
2461         int ret2;
2462
2463         path = btrfs_alloc_path();
2464         if (!path)
2465                 return -ENOENT;
2466
2467         do {
2468                 ret = check_committed_ref(trans, root, path, objectid,
2469                                           offset, bytenr);
2470                 if (ret && ret != -ENOENT)
2471                         goto out;
2472
2473                 ret2 = check_delayed_ref(trans, root, path, objectid,
2474                                          offset, bytenr);
2475         } while (ret2 == -EAGAIN);
2476
2477         if (ret2 && ret2 != -ENOENT) {
2478                 ret = ret2;
2479                 goto out;
2480         }
2481
2482         if (ret != -ENOENT || ret2 != -ENOENT)
2483                 ret = 0;
2484 out:
2485         btrfs_free_path(path);
2486         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2487                 WARN_ON(ret > 0);
2488         return ret;
2489 }
2490
2491 #if 0
2492 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2493                     struct extent_buffer *buf, u32 nr_extents)
2494 {
2495         struct btrfs_key key;
2496         struct btrfs_file_extent_item *fi;
2497         u64 root_gen;
2498         u32 nritems;
2499         int i;
2500         int level;
2501         int ret = 0;
2502         int shared = 0;
2503
2504         if (!root->ref_cows)
2505                 return 0;
2506
2507         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2508                 shared = 0;
2509                 root_gen = root->root_key.offset;
2510         } else {
2511                 shared = 1;
2512                 root_gen = trans->transid - 1;
2513         }
2514
2515         level = btrfs_header_level(buf);
2516         nritems = btrfs_header_nritems(buf);
2517
2518         if (level == 0) {
2519                 struct btrfs_leaf_ref *ref;
2520                 struct btrfs_extent_info *info;
2521
2522                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2523                 if (!ref) {
2524                         ret = -ENOMEM;
2525                         goto out;
2526                 }
2527
2528                 ref->root_gen = root_gen;
2529                 ref->bytenr = buf->start;
2530                 ref->owner = btrfs_header_owner(buf);
2531                 ref->generation = btrfs_header_generation(buf);
2532                 ref->nritems = nr_extents;
2533                 info = ref->extents;
2534
2535                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2536                         u64 disk_bytenr;
2537                         btrfs_item_key_to_cpu(buf, &key, i);
2538                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2539                                 continue;
2540                         fi = btrfs_item_ptr(buf, i,
2541                                             struct btrfs_file_extent_item);
2542                         if (btrfs_file_extent_type(buf, fi) ==
2543                             BTRFS_FILE_EXTENT_INLINE)
2544                                 continue;
2545                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2546                         if (disk_bytenr == 0)
2547                                 continue;
2548
2549                         info->bytenr = disk_bytenr;
2550                         info->num_bytes =
2551                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2552                         info->objectid = key.objectid;
2553                         info->offset = key.offset;
2554                         info++;
2555                 }
2556
2557                 ret = btrfs_add_leaf_ref(root, ref, shared);
2558                 if (ret == -EEXIST && shared) {
2559                         struct btrfs_leaf_ref *old;
2560                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2561                         BUG_ON(!old);
2562                         btrfs_remove_leaf_ref(root, old);
2563                         btrfs_free_leaf_ref(root, old);
2564                         ret = btrfs_add_leaf_ref(root, ref, shared);
2565                 }
2566                 WARN_ON(ret);
2567                 btrfs_free_leaf_ref(root, ref);
2568         }
2569 out:
2570         return ret;
2571 }
2572
2573 /* when a block goes through cow, we update the reference counts of
2574  * everything that block points to.  The internal pointers of the block
2575  * can be in just about any order, and it is likely to have clusters of
2576  * things that are close together and clusters of things that are not.
2577  *
2578  * To help reduce the seeks that come with updating all of these reference
2579  * counts, sort them by byte number before actual updates are done.
2580  *
2581  * struct refsort is used to match byte number to slot in the btree block.
2582  * we sort based on the byte number and then use the slot to actually
2583  * find the item.
2584  *
2585  * struct refsort is smaller than strcut btrfs_item and smaller than
2586  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2587  * for a btree block, there's no way for a kmalloc of refsorts for a
2588  * single node to be bigger than a page.
2589  */
2590 struct refsort {
2591         u64 bytenr;
2592         u32 slot;
2593 };
2594
2595 /*
2596  * for passing into sort()
2597  */
2598 static int refsort_cmp(const void *a_void, const void *b_void)
2599 {
2600         const struct refsort *a = a_void;
2601         const struct refsort *b = b_void;
2602
2603         if (a->bytenr < b->bytenr)
2604                 return -1;
2605         if (a->bytenr > b->bytenr)
2606                 return 1;
2607         return 0;
2608 }
2609 #endif
2610
2611 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2612                            struct btrfs_root *root,
2613                            struct extent_buffer *buf,
2614                            int full_backref, int inc)
2615 {
2616         u64 bytenr;
2617         u64 num_bytes;
2618         u64 parent;
2619         u64 ref_root;
2620         u32 nritems;
2621         struct btrfs_key key;
2622         struct btrfs_file_extent_item *fi;
2623         int i;
2624         int level;
2625         int ret = 0;
2626         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2627                             u64, u64, u64, u64, u64, u64);
2628
2629         ref_root = btrfs_header_owner(buf);
2630         nritems = btrfs_header_nritems(buf);
2631         level = btrfs_header_level(buf);
2632
2633         if (!root->ref_cows && level == 0)
2634                 return 0;
2635
2636         if (inc)
2637                 process_func = btrfs_inc_extent_ref;
2638         else
2639                 process_func = btrfs_free_extent;
2640
2641         if (full_backref)
2642                 parent = buf->start;
2643         else
2644                 parent = 0;
2645
2646         for (i = 0; i < nritems; i++) {
2647                 if (level == 0) {
2648                         btrfs_item_key_to_cpu(buf, &key, i);
2649                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2650                                 continue;
2651                         fi = btrfs_item_ptr(buf, i,
2652                                             struct btrfs_file_extent_item);
2653                         if (btrfs_file_extent_type(buf, fi) ==
2654                             BTRFS_FILE_EXTENT_INLINE)
2655                                 continue;
2656                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2657                         if (bytenr == 0)
2658                                 continue;
2659
2660                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2661                         key.offset -= btrfs_file_extent_offset(buf, fi);
2662                         ret = process_func(trans, root, bytenr, num_bytes,
2663                                            parent, ref_root, key.objectid,
2664                                            key.offset);
2665                         if (ret)
2666                                 goto fail;
2667                 } else {
2668                         bytenr = btrfs_node_blockptr(buf, i);
2669                         num_bytes = btrfs_level_size(root, level - 1);
2670                         ret = process_func(trans, root, bytenr, num_bytes,
2671                                            parent, ref_root, level - 1, 0);
2672                         if (ret)
2673                                 goto fail;
2674                 }
2675         }
2676         return 0;
2677 fail:
2678         BUG();
2679         return ret;
2680 }
2681
2682 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2683                   struct extent_buffer *buf, int full_backref)
2684 {
2685         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2686 }
2687
2688 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2689                   struct extent_buffer *buf, int full_backref)
2690 {
2691         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2692 }
2693
2694 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2695                                  struct btrfs_root *root,
2696                                  struct btrfs_path *path,
2697                                  struct btrfs_block_group_cache *cache)
2698 {
2699         int ret;
2700         struct btrfs_root *extent_root = root->fs_info->extent_root;
2701         unsigned long bi;
2702         struct extent_buffer *leaf;
2703
2704         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2705         if (ret < 0)
2706                 goto fail;
2707         BUG_ON(ret);
2708
2709         leaf = path->nodes[0];
2710         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2711         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2712         btrfs_mark_buffer_dirty(leaf);
2713         btrfs_release_path(extent_root, path);
2714 fail:
2715         if (ret)
2716                 return ret;
2717         return 0;
2718
2719 }
2720
2721 static struct btrfs_block_group_cache *
2722 next_block_group(struct btrfs_root *root,
2723                  struct btrfs_block_group_cache *cache)
2724 {
2725         struct rb_node *node;
2726         spin_lock(&root->fs_info->block_group_cache_lock);
2727         node = rb_next(&cache->cache_node);
2728         btrfs_put_block_group(cache);
2729         if (node) {
2730                 cache = rb_entry(node, struct btrfs_block_group_cache,
2731                                  cache_node);
2732                 btrfs_get_block_group(cache);
2733         } else
2734                 cache = NULL;
2735         spin_unlock(&root->fs_info->block_group_cache_lock);
2736         return cache;
2737 }
2738
2739 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2740                             struct btrfs_trans_handle *trans,
2741                             struct btrfs_path *path)
2742 {
2743         struct btrfs_root *root = block_group->fs_info->tree_root;
2744         struct inode *inode = NULL;
2745         u64 alloc_hint = 0;
2746         int dcs = BTRFS_DC_ERROR;
2747         int num_pages = 0;
2748         int retries = 0;
2749         int ret = 0;
2750
2751         /*
2752          * If this block group is smaller than 100 megs don't bother caching the
2753          * block group.
2754          */
2755         if (block_group->key.offset < (100 * 1024 * 1024)) {
2756                 spin_lock(&block_group->lock);
2757                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2758                 spin_unlock(&block_group->lock);
2759                 return 0;
2760         }
2761
2762 again:
2763         inode = lookup_free_space_inode(root, block_group, path);
2764         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2765                 ret = PTR_ERR(inode);
2766                 btrfs_release_path(root, path);
2767                 goto out;
2768         }
2769
2770         if (IS_ERR(inode)) {
2771                 BUG_ON(retries);
2772                 retries++;
2773
2774                 if (block_group->ro)
2775                         goto out_free;
2776
2777                 ret = create_free_space_inode(root, trans, block_group, path);
2778                 if (ret)
2779                         goto out_free;
2780                 goto again;
2781         }
2782
2783         /*
2784          * We want to set the generation to 0, that way if anything goes wrong
2785          * from here on out we know not to trust this cache when we load up next
2786          * time.
2787          */
2788         BTRFS_I(inode)->generation = 0;
2789         ret = btrfs_update_inode(trans, root, inode);
2790         WARN_ON(ret);
2791
2792         if (i_size_read(inode) > 0) {
2793                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2794                                                       inode);
2795                 if (ret)
2796                         goto out_put;
2797         }
2798
2799         spin_lock(&block_group->lock);
2800         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2801                 /* We're not cached, don't bother trying to write stuff out */
2802                 dcs = BTRFS_DC_WRITTEN;
2803                 spin_unlock(&block_group->lock);
2804                 goto out_put;
2805         }
2806         spin_unlock(&block_group->lock);
2807
2808         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2809         if (!num_pages)
2810                 num_pages = 1;
2811
2812         /*
2813          * Just to make absolutely sure we have enough space, we're going to
2814          * preallocate 12 pages worth of space for each block group.  In
2815          * practice we ought to use at most 8, but we need extra space so we can
2816          * add our header and have a terminator between the extents and the
2817          * bitmaps.
2818          */
2819         num_pages *= 16;
2820         num_pages *= PAGE_CACHE_SIZE;
2821
2822         ret = btrfs_check_data_free_space(inode, num_pages);
2823         if (ret)
2824                 goto out_put;
2825
2826         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2827                                               num_pages, num_pages,
2828                                               &alloc_hint);
2829         if (!ret)
2830                 dcs = BTRFS_DC_SETUP;
2831         btrfs_free_reserved_data_space(inode, num_pages);
2832 out_put:
2833         iput(inode);
2834 out_free:
2835         btrfs_release_path(root, path);
2836 out:
2837         spin_lock(&block_group->lock);
2838         block_group->disk_cache_state = dcs;
2839         spin_unlock(&block_group->lock);
2840
2841         return ret;
2842 }
2843
2844 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2845                                    struct btrfs_root *root)
2846 {
2847         struct btrfs_block_group_cache *cache;
2848         int err = 0;
2849         struct btrfs_path *path;
2850         u64 last = 0;
2851
2852         path = btrfs_alloc_path();
2853         if (!path)
2854                 return -ENOMEM;
2855
2856 again:
2857         while (1) {
2858                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2859                 while (cache) {
2860                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2861                                 break;
2862                         cache = next_block_group(root, cache);
2863                 }
2864                 if (!cache) {
2865                         if (last == 0)
2866                                 break;
2867                         last = 0;
2868                         continue;
2869                 }
2870                 err = cache_save_setup(cache, trans, path);
2871                 last = cache->key.objectid + cache->key.offset;
2872                 btrfs_put_block_group(cache);
2873         }
2874
2875         while (1) {
2876                 if (last == 0) {
2877                         err = btrfs_run_delayed_refs(trans, root,
2878                                                      (unsigned long)-1);
2879                         BUG_ON(err);
2880                 }
2881
2882                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2883                 while (cache) {
2884                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2885                                 btrfs_put_block_group(cache);
2886                                 goto again;
2887                         }
2888
2889                         if (cache->dirty)
2890                                 break;
2891                         cache = next_block_group(root, cache);
2892                 }
2893                 if (!cache) {
2894                         if (last == 0)
2895                                 break;
2896                         last = 0;
2897                         continue;
2898                 }
2899
2900                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2901                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2902                 cache->dirty = 0;
2903                 last = cache->key.objectid + cache->key.offset;
2904
2905                 err = write_one_cache_group(trans, root, path, cache);
2906                 BUG_ON(err);
2907                 btrfs_put_block_group(cache);
2908         }
2909
2910         while (1) {
2911                 /*
2912                  * I don't think this is needed since we're just marking our
2913                  * preallocated extent as written, but just in case it can't
2914                  * hurt.
2915                  */
2916                 if (last == 0) {
2917                         err = btrfs_run_delayed_refs(trans, root,
2918                                                      (unsigned long)-1);
2919                         BUG_ON(err);
2920                 }
2921
2922                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2923                 while (cache) {
2924                         /*
2925                          * Really this shouldn't happen, but it could if we
2926                          * couldn't write the entire preallocated extent and
2927                          * splitting the extent resulted in a new block.
2928                          */
2929                         if (cache->dirty) {
2930                                 btrfs_put_block_group(cache);
2931                                 goto again;
2932                         }
2933                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2934                                 break;
2935                         cache = next_block_group(root, cache);
2936                 }
2937                 if (!cache) {
2938                         if (last == 0)
2939                                 break;
2940                         last = 0;
2941                         continue;
2942                 }
2943
2944                 btrfs_write_out_cache(root, trans, cache, path);
2945
2946                 /*
2947                  * If we didn't have an error then the cache state is still
2948                  * NEED_WRITE, so we can set it to WRITTEN.
2949                  */
2950                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2951                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2952                 last = cache->key.objectid + cache->key.offset;
2953                 btrfs_put_block_group(cache);
2954         }
2955
2956         btrfs_free_path(path);
2957         return 0;
2958 }
2959
2960 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2961 {
2962         struct btrfs_block_group_cache *block_group;
2963         int readonly = 0;
2964
2965         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2966         if (!block_group || block_group->ro)
2967                 readonly = 1;
2968         if (block_group)
2969                 btrfs_put_block_group(block_group);
2970         return readonly;
2971 }
2972
2973 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2974                              u64 total_bytes, u64 bytes_used,
2975                              struct btrfs_space_info **space_info)
2976 {
2977         struct btrfs_space_info *found;
2978         int i;
2979         int factor;
2980
2981         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2982                      BTRFS_BLOCK_GROUP_RAID10))
2983                 factor = 2;
2984         else
2985                 factor = 1;
2986
2987         found = __find_space_info(info, flags);
2988         if (found) {
2989                 spin_lock(&found->lock);
2990                 found->total_bytes += total_bytes;
2991                 found->disk_total += total_bytes * factor;
2992                 found->bytes_used += bytes_used;
2993                 found->disk_used += bytes_used * factor;
2994                 found->full = 0;
2995                 spin_unlock(&found->lock);
2996                 *space_info = found;
2997                 return 0;
2998         }
2999         found = kzalloc(sizeof(*found), GFP_NOFS);
3000         if (!found)
3001                 return -ENOMEM;
3002
3003         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3004                 INIT_LIST_HEAD(&found->block_groups[i]);
3005         init_rwsem(&found->groups_sem);
3006         spin_lock_init(&found->lock);
3007         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3008                                 BTRFS_BLOCK_GROUP_SYSTEM |
3009                                 BTRFS_BLOCK_GROUP_METADATA);
3010         found->total_bytes = total_bytes;
3011         found->disk_total = total_bytes * factor;
3012         found->bytes_used = bytes_used;
3013         found->disk_used = bytes_used * factor;
3014         found->bytes_pinned = 0;
3015         found->bytes_reserved = 0;
3016         found->bytes_readonly = 0;
3017         found->bytes_may_use = 0;
3018         found->full = 0;
3019         found->force_alloc = 0;
3020         *space_info = found;
3021         list_add_rcu(&found->list, &info->space_info);
3022         atomic_set(&found->caching_threads, 0);
3023         return 0;
3024 }
3025
3026 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3027 {
3028         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3029                                    BTRFS_BLOCK_GROUP_RAID1 |
3030                                    BTRFS_BLOCK_GROUP_RAID10 |
3031                                    BTRFS_BLOCK_GROUP_DUP);
3032         if (extra_flags) {
3033                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3034                         fs_info->avail_data_alloc_bits |= extra_flags;
3035                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3036                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3037                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3038                         fs_info->avail_system_alloc_bits |= extra_flags;
3039         }
3040 }
3041
3042 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3043 {
3044         /*
3045          * we add in the count of missing devices because we want
3046          * to make sure that any RAID levels on a degraded FS
3047          * continue to be honored.
3048          */
3049         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3050                 root->fs_info->fs_devices->missing_devices;
3051
3052         if (num_devices == 1)
3053                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3054         if (num_devices < 4)
3055                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3056
3057         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3058             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3059                       BTRFS_BLOCK_GROUP_RAID10))) {
3060                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3061         }
3062
3063         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3064             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3065                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3066         }
3067
3068         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3069             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3070              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3071              (flags & BTRFS_BLOCK_GROUP_DUP)))
3072                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3073         return flags;
3074 }
3075
3076 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3077 {
3078         if (flags & BTRFS_BLOCK_GROUP_DATA)
3079                 flags |= root->fs_info->avail_data_alloc_bits &
3080                          root->fs_info->data_alloc_profile;
3081         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3082                 flags |= root->fs_info->avail_system_alloc_bits &
3083                          root->fs_info->system_alloc_profile;
3084         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3085                 flags |= root->fs_info->avail_metadata_alloc_bits &
3086                          root->fs_info->metadata_alloc_profile;
3087         return btrfs_reduce_alloc_profile(root, flags);
3088 }
3089
3090 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3091 {
3092         u64 flags;
3093
3094         if (data)
3095                 flags = BTRFS_BLOCK_GROUP_DATA;
3096         else if (root == root->fs_info->chunk_root)
3097                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3098         else
3099                 flags = BTRFS_BLOCK_GROUP_METADATA;
3100
3101         return get_alloc_profile(root, flags);
3102 }
3103
3104 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3105 {
3106         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3107                                                        BTRFS_BLOCK_GROUP_DATA);
3108 }
3109
3110 /*
3111  * This will check the space that the inode allocates from to make sure we have
3112  * enough space for bytes.
3113  */
3114 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3115 {
3116         struct btrfs_space_info *data_sinfo;
3117         struct btrfs_root *root = BTRFS_I(inode)->root;
3118         u64 used;
3119         int ret = 0, committed = 0, alloc_chunk = 1;
3120
3121         /* make sure bytes are sectorsize aligned */
3122         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3123
3124         if (root == root->fs_info->tree_root) {
3125                 alloc_chunk = 0;
3126                 committed = 1;
3127         }
3128
3129         data_sinfo = BTRFS_I(inode)->space_info;
3130         if (!data_sinfo)
3131                 goto alloc;
3132
3133 again:
3134         /* make sure we have enough space to handle the data first */
3135         spin_lock(&data_sinfo->lock);
3136         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3137                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3138                 data_sinfo->bytes_may_use;
3139
3140         if (used + bytes > data_sinfo->total_bytes) {
3141                 struct btrfs_trans_handle *trans;
3142
3143                 /*
3144                  * if we don't have enough free bytes in this space then we need
3145                  * to alloc a new chunk.
3146                  */
3147                 if (!data_sinfo->full && alloc_chunk) {
3148                         u64 alloc_target;
3149
3150                         data_sinfo->force_alloc = 1;
3151                         spin_unlock(&data_sinfo->lock);
3152 alloc:
3153                         alloc_target = btrfs_get_alloc_profile(root, 1);
3154                         trans = btrfs_join_transaction(root, 1);
3155                         if (IS_ERR(trans))
3156                                 return PTR_ERR(trans);
3157
3158                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3159                                              bytes + 2 * 1024 * 1024,
3160                                              alloc_target, 0);
3161                         btrfs_end_transaction(trans, root);
3162                         if (ret < 0) {
3163                                 if (ret != -ENOSPC)
3164                                         return ret;
3165                                 else
3166                                         goto commit_trans;
3167                         }
3168
3169                         if (!data_sinfo) {
3170                                 btrfs_set_inode_space_info(root, inode);
3171                                 data_sinfo = BTRFS_I(inode)->space_info;
3172                         }
3173                         goto again;
3174                 }
3175                 spin_unlock(&data_sinfo->lock);
3176
3177                 /* commit the current transaction and try again */
3178 commit_trans:
3179                 if (!committed && !root->fs_info->open_ioctl_trans) {
3180                         committed = 1;
3181                         trans = btrfs_join_transaction(root, 1);
3182                         if (IS_ERR(trans))
3183                                 return PTR_ERR(trans);
3184                         ret = btrfs_commit_transaction(trans, root);
3185                         if (ret)
3186                                 return ret;
3187                         goto again;
3188                 }
3189
3190 #if 0 /* I hope we never need this code again, just in case */
3191                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3192                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3193                        "%llu bytes_readonly, %llu may use %llu total\n",
3194                        (unsigned long long)bytes,
3195                        (unsigned long long)data_sinfo->bytes_used,
3196                        (unsigned long long)data_sinfo->bytes_reserved,
3197                        (unsigned long long)data_sinfo->bytes_pinned,
3198                        (unsigned long long)data_sinfo->bytes_readonly,
3199                        (unsigned long long)data_sinfo->bytes_may_use,
3200                        (unsigned long long)data_sinfo->total_bytes);
3201 #endif
3202                 return -ENOSPC;
3203         }
3204         data_sinfo->bytes_may_use += bytes;
3205         BTRFS_I(inode)->reserved_bytes += bytes;
3206         spin_unlock(&data_sinfo->lock);
3207
3208         return 0;
3209 }
3210
3211 /*
3212  * called when we are clearing an delalloc extent from the
3213  * inode's io_tree or there was an error for whatever reason
3214  * after calling btrfs_check_data_free_space
3215  */
3216 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3217 {
3218         struct btrfs_root *root = BTRFS_I(inode)->root;
3219         struct btrfs_space_info *data_sinfo;
3220
3221         /* make sure bytes are sectorsize aligned */
3222         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3223
3224         data_sinfo = BTRFS_I(inode)->space_info;
3225         spin_lock(&data_sinfo->lock);
3226         data_sinfo->bytes_may_use -= bytes;
3227         BTRFS_I(inode)->reserved_bytes -= bytes;
3228         spin_unlock(&data_sinfo->lock);
3229 }
3230
3231 static void force_metadata_allocation(struct btrfs_fs_info *info)
3232 {
3233         struct list_head *head = &info->space_info;
3234         struct btrfs_space_info *found;
3235
3236         rcu_read_lock();
3237         list_for_each_entry_rcu(found, head, list) {
3238                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3239                         found->force_alloc = 1;
3240         }
3241         rcu_read_unlock();
3242 }
3243
3244 static int should_alloc_chunk(struct btrfs_root *root,
3245                               struct btrfs_space_info *sinfo, u64 alloc_bytes)
3246 {
3247         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3248         u64 thresh;
3249
3250         if (sinfo->bytes_used + sinfo->bytes_reserved +
3251             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3252                 return 0;
3253
3254         if (sinfo->bytes_used + sinfo->bytes_reserved +
3255             alloc_bytes < div_factor(num_bytes, 8))
3256                 return 0;
3257
3258         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3259         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3260
3261         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3262                 return 0;
3263
3264         return 1;
3265 }
3266
3267 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3268                           struct btrfs_root *extent_root, u64 alloc_bytes,
3269                           u64 flags, int force)
3270 {
3271         struct btrfs_space_info *space_info;
3272         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3273         int ret = 0;
3274
3275         mutex_lock(&fs_info->chunk_mutex);
3276
3277         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3278
3279         space_info = __find_space_info(extent_root->fs_info, flags);
3280         if (!space_info) {
3281                 ret = update_space_info(extent_root->fs_info, flags,
3282                                         0, 0, &space_info);
3283                 BUG_ON(ret);
3284         }
3285         BUG_ON(!space_info);
3286
3287         spin_lock(&space_info->lock);
3288         if (space_info->force_alloc)
3289                 force = 1;
3290         if (space_info->full) {
3291                 spin_unlock(&space_info->lock);
3292                 goto out;
3293         }
3294
3295         if (!force && !should_alloc_chunk(extent_root, space_info,
3296                                           alloc_bytes)) {
3297                 spin_unlock(&space_info->lock);
3298                 goto out;
3299         }
3300         spin_unlock(&space_info->lock);
3301
3302         /*
3303          * If we have mixed data/metadata chunks we want to make sure we keep
3304          * allocating mixed chunks instead of individual chunks.
3305          */
3306         if (btrfs_mixed_space_info(space_info))
3307                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3308
3309         /*
3310          * if we're doing a data chunk, go ahead and make sure that
3311          * we keep a reasonable number of metadata chunks allocated in the
3312          * FS as well.
3313          */
3314         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3315                 fs_info->data_chunk_allocations++;
3316                 if (!(fs_info->data_chunk_allocations %
3317                       fs_info->metadata_ratio))
3318                         force_metadata_allocation(fs_info);
3319         }
3320
3321         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3322         spin_lock(&space_info->lock);
3323         if (ret)
3324                 space_info->full = 1;
3325         else
3326                 ret = 1;
3327         space_info->force_alloc = 0;
3328         spin_unlock(&space_info->lock);
3329 out:
3330         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3331         return ret;
3332 }
3333
3334 /*
3335  * shrink metadata reservation for delalloc
3336  */
3337 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3338                            struct btrfs_root *root, u64 to_reclaim, int sync)
3339 {
3340         struct btrfs_block_rsv *block_rsv;
3341         struct btrfs_space_info *space_info;
3342         u64 reserved;
3343         u64 max_reclaim;
3344         u64 reclaimed = 0;
3345         long time_left;
3346         int pause = 1;
3347         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3348         int loops = 0;
3349
3350         block_rsv = &root->fs_info->delalloc_block_rsv;
3351         space_info = block_rsv->space_info;
3352
3353         smp_mb();
3354         reserved = space_info->bytes_reserved;
3355
3356         if (reserved == 0)
3357                 return 0;
3358
3359         max_reclaim = min(reserved, to_reclaim);
3360
3361         while (loops < 1024) {
3362                 /* have the flusher threads jump in and do some IO */
3363                 smp_mb();
3364                 nr_pages = min_t(unsigned long, nr_pages,
3365                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3366                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3367
3368                 spin_lock(&space_info->lock);
3369                 if (reserved > space_info->bytes_reserved) {
3370                         loops = 0;
3371                         reclaimed += reserved - space_info->bytes_reserved;
3372                 } else {
3373                         loops++;
3374                 }
3375                 reserved = space_info->bytes_reserved;
3376                 spin_unlock(&space_info->lock);
3377
3378                 if (reserved == 0 || reclaimed >= max_reclaim)
3379                         break;
3380
3381                 if (trans && trans->transaction->blocked)
3382                         return -EAGAIN;
3383
3384                 __set_current_state(TASK_INTERRUPTIBLE);
3385                 time_left = schedule_timeout(pause);
3386
3387                 /* We were interrupted, exit */
3388                 if (time_left)
3389                         break;
3390
3391                 pause <<= 1;
3392                 if (pause > HZ / 10)
3393                         pause = HZ / 10;
3394
3395         }
3396         return reclaimed >= to_reclaim;
3397 }
3398
3399 /*
3400  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3401  * idea is if this is the first call (retries == 0) then we will add to our
3402  * reserved count if we can't make the allocation in order to hold our place
3403  * while we go and try and free up space.  That way for retries > 1 we don't try
3404  * and add space, we just check to see if the amount of unused space is >= the
3405  * total space, meaning that our reservation is valid.
3406  *
3407  * However if we don't intend to retry this reservation, pass -1 as retries so
3408  * that it short circuits this logic.
3409  */
3410 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3411                                   struct btrfs_root *root,
3412                                   struct btrfs_block_rsv *block_rsv,
3413                                   u64 orig_bytes, int flush)
3414 {
3415         struct btrfs_space_info *space_info = block_rsv->space_info;
3416         u64 unused;
3417         u64 num_bytes = orig_bytes;
3418         int retries = 0;
3419         int ret = 0;
3420         bool reserved = false;
3421         bool committed = false;
3422
3423 again:
3424         ret = -ENOSPC;
3425         if (reserved)
3426                 num_bytes = 0;
3427
3428         spin_lock(&space_info->lock);
3429         unused = space_info->bytes_used + space_info->bytes_reserved +
3430                  space_info->bytes_pinned + space_info->bytes_readonly +
3431                  space_info->bytes_may_use;
3432
3433         /*
3434          * The idea here is that we've not already over-reserved the block group
3435          * then we can go ahead and save our reservation first and then start
3436          * flushing if we need to.  Otherwise if we've already overcommitted
3437          * lets start flushing stuff first and then come back and try to make
3438          * our reservation.
3439          */
3440         if (unused <= space_info->total_bytes) {
3441                 unused = space_info->total_bytes - unused;
3442                 if (unused >= num_bytes) {
3443                         if (!reserved)
3444                                 space_info->bytes_reserved += orig_bytes;
3445                         ret = 0;
3446                 } else {
3447                         /*
3448                          * Ok set num_bytes to orig_bytes since we aren't
3449                          * overocmmitted, this way we only try and reclaim what
3450                          * we need.
3451                          */
3452                         num_bytes = orig_bytes;
3453                 }
3454         } else {
3455                 /*
3456                  * Ok we're over committed, set num_bytes to the overcommitted
3457                  * amount plus the amount of bytes that we need for this
3458                  * reservation.
3459                  */
3460                 num_bytes = unused - space_info->total_bytes +
3461                         (orig_bytes * (retries + 1));
3462         }
3463
3464         /*
3465          * Couldn't make our reservation, save our place so while we're trying
3466          * to reclaim space we can actually use it instead of somebody else
3467          * stealing it from us.
3468          */
3469         if (ret && !reserved) {
3470                 space_info->bytes_reserved += orig_bytes;
3471                 reserved = true;
3472         }
3473
3474         spin_unlock(&space_info->lock);
3475
3476         if (!ret)
3477                 return 0;
3478
3479         if (!flush)
3480                 goto out;
3481
3482         /*
3483          * We do synchronous shrinking since we don't actually unreserve
3484          * metadata until after the IO is completed.
3485          */
3486         ret = shrink_delalloc(trans, root, num_bytes, 1);
3487         if (ret > 0)
3488                 return 0;
3489         else if (ret < 0)
3490                 goto out;
3491
3492         /*
3493          * So if we were overcommitted it's possible that somebody else flushed
3494          * out enough space and we simply didn't have enough space to reclaim,
3495          * so go back around and try again.
3496          */
3497         if (retries < 2) {
3498                 retries++;
3499                 goto again;
3500         }
3501
3502         spin_lock(&space_info->lock);
3503         /*
3504          * Not enough space to be reclaimed, don't bother committing the
3505          * transaction.
3506          */
3507         if (space_info->bytes_pinned < orig_bytes)
3508                 ret = -ENOSPC;
3509         spin_unlock(&space_info->lock);
3510         if (ret)
3511                 goto out;
3512
3513         ret = -EAGAIN;
3514         if (trans || committed)
3515                 goto out;
3516
3517         ret = -ENOSPC;
3518         trans = btrfs_join_transaction(root, 1);
3519         if (IS_ERR(trans))
3520                 goto out;
3521         ret = btrfs_commit_transaction(trans, root);
3522         if (!ret) {
3523                 trans = NULL;
3524                 committed = true;
3525                 goto again;
3526         }
3527
3528 out:
3529         if (reserved) {
3530                 spin_lock(&space_info->lock);
3531                 space_info->bytes_reserved -= orig_bytes;
3532                 spin_unlock(&space_info->lock);
3533         }
3534
3535         return ret;
3536 }
3537
3538 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3539                                              struct btrfs_root *root)
3540 {
3541         struct btrfs_block_rsv *block_rsv;
3542         if (root->ref_cows)
3543                 block_rsv = trans->block_rsv;
3544         else
3545                 block_rsv = root->block_rsv;
3546
3547         if (!block_rsv)
3548                 block_rsv = &root->fs_info->empty_block_rsv;
3549
3550         return block_rsv;
3551 }
3552
3553 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3554                                u64 num_bytes)
3555 {
3556         int ret = -ENOSPC;
3557         spin_lock(&block_rsv->lock);
3558         if (block_rsv->reserved >= num_bytes) {
3559                 block_rsv->reserved -= num_bytes;
3560                 if (block_rsv->reserved < block_rsv->size)
3561                         block_rsv->full = 0;
3562                 ret = 0;
3563         }
3564         spin_unlock(&block_rsv->lock);
3565         return ret;
3566 }
3567
3568 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3569                                 u64 num_bytes, int update_size)
3570 {
3571         spin_lock(&block_rsv->lock);
3572         block_rsv->reserved += num_bytes;
3573         if (update_size)
3574                 block_rsv->size += num_bytes;
3575         else if (block_rsv->reserved >= block_rsv->size)
3576                 block_rsv->full = 1;
3577         spin_unlock(&block_rsv->lock);
3578 }
3579
3580 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3581                              struct btrfs_block_rsv *dest, u64 num_bytes)
3582 {
3583         struct btrfs_space_info *space_info = block_rsv->space_info;
3584
3585         spin_lock(&block_rsv->lock);
3586         if (num_bytes == (u64)-1)
3587                 num_bytes = block_rsv->size;
3588         block_rsv->size -= num_bytes;
3589         if (block_rsv->reserved >= block_rsv->size) {
3590                 num_bytes = block_rsv->reserved - block_rsv->size;
3591                 block_rsv->reserved = block_rsv->size;
3592                 block_rsv->full = 1;
3593         } else {
3594                 num_bytes = 0;
3595         }
3596         spin_unlock(&block_rsv->lock);
3597
3598         if (num_bytes > 0) {
3599                 if (dest) {
3600                         spin_lock(&dest->lock);
3601                         if (!dest->full) {
3602                                 u64 bytes_to_add;
3603
3604                                 bytes_to_add = dest->size - dest->reserved;
3605                                 bytes_to_add = min(num_bytes, bytes_to_add);
3606                                 dest->reserved += bytes_to_add;
3607                                 if (dest->reserved >= dest->size)
3608                                         dest->full = 1;
3609                                 num_bytes -= bytes_to_add;
3610                         }
3611                         spin_unlock(&dest->lock);
3612                 }
3613                 if (num_bytes) {
3614                         spin_lock(&space_info->lock);
3615                         space_info->bytes_reserved -= num_bytes;
3616                         spin_unlock(&space_info->lock);
3617                 }
3618         }
3619 }
3620
3621 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3622                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3623 {
3624         int ret;
3625
3626         ret = block_rsv_use_bytes(src, num_bytes);
3627         if (ret)
3628                 return ret;
3629
3630         block_rsv_add_bytes(dst, num_bytes, 1);
3631         return 0;
3632 }
3633
3634 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3635 {
3636         memset(rsv, 0, sizeof(*rsv));
3637         spin_lock_init(&rsv->lock);
3638         atomic_set(&rsv->usage, 1);
3639         rsv->priority = 6;
3640         INIT_LIST_HEAD(&rsv->list);
3641 }
3642
3643 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3644 {
3645         struct btrfs_block_rsv *block_rsv;
3646         struct btrfs_fs_info *fs_info = root->fs_info;
3647
3648         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3649         if (!block_rsv)
3650                 return NULL;
3651
3652         btrfs_init_block_rsv(block_rsv);
3653         block_rsv->space_info = __find_space_info(fs_info,
3654                                                   BTRFS_BLOCK_GROUP_METADATA);
3655         return block_rsv;
3656 }
3657
3658 void btrfs_free_block_rsv(struct btrfs_root *root,
3659                           struct btrfs_block_rsv *rsv)
3660 {
3661         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3662                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3663                 if (!rsv->durable)
3664                         kfree(rsv);
3665         }
3666 }
3667
3668 /*
3669  * make the block_rsv struct be able to capture freed space.
3670  * the captured space will re-add to the the block_rsv struct
3671  * after transaction commit
3672  */
3673 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3674                                  struct btrfs_block_rsv *block_rsv)
3675 {
3676         block_rsv->durable = 1;
3677         mutex_lock(&fs_info->durable_block_rsv_mutex);
3678         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3679         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3680 }
3681
3682 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3683                         struct btrfs_root *root,
3684                         struct btrfs_block_rsv *block_rsv,
3685                         u64 num_bytes)
3686 {
3687         int ret;
3688
3689         if (num_bytes == 0)
3690                 return 0;
3691
3692         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3693         if (!ret) {
3694                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3695                 return 0;
3696         }
3697
3698         return ret;
3699 }
3700
3701 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3702                           struct btrfs_root *root,
3703                           struct btrfs_block_rsv *block_rsv,
3704                           u64 min_reserved, int min_factor)
3705 {
3706         u64 num_bytes = 0;
3707         int commit_trans = 0;
3708         int ret = -ENOSPC;
3709
3710         if (!block_rsv)
3711                 return 0;
3712
3713         spin_lock(&block_rsv->lock);
3714         if (min_factor > 0)
3715                 num_bytes = div_factor(block_rsv->size, min_factor);
3716         if (min_reserved > num_bytes)
3717                 num_bytes = min_reserved;
3718
3719         if (block_rsv->reserved >= num_bytes) {
3720                 ret = 0;
3721         } else {
3722                 num_bytes -= block_rsv->reserved;
3723                 if (block_rsv->durable &&
3724                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3725                         commit_trans = 1;
3726         }
3727         spin_unlock(&block_rsv->lock);
3728         if (!ret)
3729                 return 0;
3730
3731         if (block_rsv->refill_used) {
3732                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3733                                              num_bytes, 0);
3734                 if (!ret) {
3735                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3736                         return 0;
3737                 }
3738         }
3739
3740         if (commit_trans) {
3741                 if (trans)
3742                         return -EAGAIN;
3743
3744                 trans = btrfs_join_transaction(root, 1);
3745                 BUG_ON(IS_ERR(trans));
3746                 ret = btrfs_commit_transaction(trans, root);
3747                 return 0;
3748         }
3749
3750         return -ENOSPC;
3751 }
3752
3753 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3754                             struct btrfs_block_rsv *dst_rsv,
3755                             u64 num_bytes)
3756 {
3757         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3758 }
3759
3760 void btrfs_block_rsv_release(struct btrfs_root *root,
3761                              struct btrfs_block_rsv *block_rsv,
3762                              u64 num_bytes)
3763 {
3764         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3765         if (global_rsv->full || global_rsv == block_rsv ||
3766             block_rsv->space_info != global_rsv->space_info)
3767                 global_rsv = NULL;
3768         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3769 }
3770
3771 /*
3772  * helper to calculate size of global block reservation.
3773  * the desired value is sum of space used by extent tree,
3774  * checksum tree and root tree
3775  */
3776 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3777 {
3778         struct btrfs_space_info *sinfo;
3779         u64 num_bytes;
3780         u64 meta_used;
3781         u64 data_used;
3782         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3783 #if 0
3784         /*
3785          * per tree used space accounting can be inaccuracy, so we
3786          * can't rely on it.
3787          */
3788         spin_lock(&fs_info->extent_root->accounting_lock);
3789         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3790         spin_unlock(&fs_info->extent_root->accounting_lock);
3791
3792         spin_lock(&fs_info->csum_root->accounting_lock);
3793         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3794         spin_unlock(&fs_info->csum_root->accounting_lock);
3795
3796         spin_lock(&fs_info->tree_root->accounting_lock);
3797         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3798         spin_unlock(&fs_info->tree_root->accounting_lock);
3799 #endif
3800         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3801         spin_lock(&sinfo->lock);
3802         data_used = sinfo->bytes_used;
3803         spin_unlock(&sinfo->lock);
3804
3805         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3806         spin_lock(&sinfo->lock);
3807         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3808                 data_used = 0;
3809         meta_used = sinfo->bytes_used;
3810         spin_unlock(&sinfo->lock);
3811
3812         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3813                     csum_size * 2;
3814         num_bytes += div64_u64(data_used + meta_used, 50);
3815
3816         if (num_bytes * 3 > meta_used)
3817                 num_bytes = div64_u64(meta_used, 3);
3818
3819         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3820 }
3821
3822 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3823 {
3824         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3825         struct btrfs_space_info *sinfo = block_rsv->space_info;
3826         u64 num_bytes;
3827
3828         num_bytes = calc_global_metadata_size(fs_info);
3829
3830         spin_lock(&block_rsv->lock);
3831         spin_lock(&sinfo->lock);
3832
3833         block_rsv->size = num_bytes;
3834
3835         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3836                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3837                     sinfo->bytes_may_use;
3838
3839         if (sinfo->total_bytes > num_bytes) {
3840                 num_bytes = sinfo->total_bytes - num_bytes;
3841                 block_rsv->reserved += num_bytes;
3842                 sinfo->bytes_reserved += num_bytes;
3843         }
3844
3845         if (block_rsv->reserved >= block_rsv->size) {
3846                 num_bytes = block_rsv->reserved - block_rsv->size;
3847                 sinfo->bytes_reserved -= num_bytes;
3848                 block_rsv->reserved = block_rsv->size;
3849                 block_rsv->full = 1;
3850         }
3851 #if 0
3852         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3853                 block_rsv->size, block_rsv->reserved);
3854 #endif
3855         spin_unlock(&sinfo->lock);
3856         spin_unlock(&block_rsv->lock);
3857 }
3858
3859 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3860 {
3861         struct btrfs_space_info *space_info;
3862
3863         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3864         fs_info->chunk_block_rsv.space_info = space_info;
3865         fs_info->chunk_block_rsv.priority = 10;
3866
3867         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3868         fs_info->global_block_rsv.space_info = space_info;
3869         fs_info->global_block_rsv.priority = 10;
3870         fs_info->global_block_rsv.refill_used = 1;
3871         fs_info->delalloc_block_rsv.space_info = space_info;
3872         fs_info->trans_block_rsv.space_info = space_info;
3873         fs_info->empty_block_rsv.space_info = space_info;
3874         fs_info->empty_block_rsv.priority = 10;
3875
3876         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3877         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3878         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3879         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3880         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3881
3882         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3883
3884         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3885
3886         update_global_block_rsv(fs_info);
3887 }
3888
3889 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3890 {
3891         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3892         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3893         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3894         WARN_ON(fs_info->trans_block_rsv.size > 0);
3895         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3896         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3897         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3898 }
3899
3900 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3901 {
3902         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3903                 3 * num_items;
3904 }
3905
3906 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3907                                  struct btrfs_root *root,
3908                                  int num_items)
3909 {
3910         u64 num_bytes;
3911         int ret;
3912
3913         if (num_items == 0 || root->fs_info->chunk_root == root)
3914                 return 0;
3915
3916         num_bytes = calc_trans_metadata_size(root, num_items);
3917         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3918                                   num_bytes);
3919         if (!ret) {
3920                 trans->bytes_reserved += num_bytes;
3921                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3922         }
3923         return ret;
3924 }
3925
3926 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3927                                   struct btrfs_root *root)
3928 {
3929         if (!trans->bytes_reserved)
3930                 return;
3931
3932         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3933         btrfs_block_rsv_release(root, trans->block_rsv,
3934                                 trans->bytes_reserved);
3935         trans->bytes_reserved = 0;
3936 }
3937
3938 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3939                                   struct inode *inode)
3940 {
3941         struct btrfs_root *root = BTRFS_I(inode)->root;
3942         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3943         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3944
3945         /*
3946          * one for deleting orphan item, one for updating inode and
3947          * two for calling btrfs_truncate_inode_items.
3948          *
3949          * btrfs_truncate_inode_items is a delete operation, it frees
3950          * more space than it uses in most cases. So two units of
3951          * metadata space should be enough for calling it many times.
3952          * If all of the metadata space is used, we can commit
3953          * transaction and use space it freed.
3954          */
3955         u64 num_bytes = calc_trans_metadata_size(root, 4);
3956         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3957 }
3958
3959 void btrfs_orphan_release_metadata(struct inode *inode)
3960 {
3961         struct btrfs_root *root = BTRFS_I(inode)->root;
3962         u64 num_bytes = calc_trans_metadata_size(root, 4);
3963         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3964 }
3965
3966 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3967                                 struct btrfs_pending_snapshot *pending)
3968 {
3969         struct btrfs_root *root = pending->root;
3970         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3971         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3972         /*
3973          * two for root back/forward refs, two for directory entries
3974          * and one for root of the snapshot.
3975          */
3976         u64 num_bytes = calc_trans_metadata_size(root, 5);
3977         dst_rsv->space_info = src_rsv->space_info;
3978         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3979 }
3980
3981 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3982 {
3983         return num_bytes >>= 3;
3984 }
3985
3986 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3987 {
3988         struct btrfs_root *root = BTRFS_I(inode)->root;
3989         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3990         u64 to_reserve;
3991         int nr_extents;
3992         int ret;
3993
3994         if (btrfs_transaction_in_commit(root->fs_info))
3995                 schedule_timeout(1);
3996
3997         num_bytes = ALIGN(num_bytes, root->sectorsize);
3998
3999         spin_lock(&BTRFS_I(inode)->accounting_lock);
4000         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
4001         if (nr_extents > BTRFS_I(inode)->reserved_extents) {
4002                 nr_extents -= BTRFS_I(inode)->reserved_extents;
4003                 to_reserve = calc_trans_metadata_size(root, nr_extents);
4004         } else {
4005                 nr_extents = 0;
4006                 to_reserve = 0;
4007         }
4008         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4009
4010         to_reserve += calc_csum_metadata_size(inode, num_bytes);
4011         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4012         if (ret)
4013                 return ret;
4014
4015         spin_lock(&BTRFS_I(inode)->accounting_lock);
4016         BTRFS_I(inode)->reserved_extents += nr_extents;
4017         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4018         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4019
4020         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4021
4022         if (block_rsv->size > 512 * 1024 * 1024)
4023                 shrink_delalloc(NULL, root, to_reserve, 0);
4024
4025         return 0;
4026 }
4027
4028 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4029 {
4030         struct btrfs_root *root = BTRFS_I(inode)->root;
4031         u64 to_free;
4032         int nr_extents;
4033
4034         num_bytes = ALIGN(num_bytes, root->sectorsize);
4035         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4036         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
4037
4038         spin_lock(&BTRFS_I(inode)->accounting_lock);
4039         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4040         if (nr_extents < BTRFS_I(inode)->reserved_extents) {
4041                 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
4042                 BTRFS_I(inode)->reserved_extents -= nr_extents;
4043         } else {
4044                 nr_extents = 0;
4045         }
4046         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4047
4048         to_free = calc_csum_metadata_size(inode, num_bytes);
4049         if (nr_extents > 0)
4050                 to_free += calc_trans_metadata_size(root, nr_extents);
4051
4052         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4053                                 to_free);
4054 }
4055
4056 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4057 {
4058         int ret;
4059
4060         ret = btrfs_check_data_free_space(inode, num_bytes);
4061         if (ret)
4062                 return ret;
4063
4064         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4065         if (ret) {
4066                 btrfs_free_reserved_data_space(inode, num_bytes);
4067                 return ret;
4068         }
4069
4070         return 0;
4071 }
4072
4073 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4074 {
4075         btrfs_delalloc_release_metadata(inode, num_bytes);
4076         btrfs_free_reserved_data_space(inode, num_bytes);
4077 }
4078
4079 static int update_block_group(struct btrfs_trans_handle *trans,
4080                               struct btrfs_root *root,
4081                               u64 bytenr, u64 num_bytes, int alloc)
4082 {
4083         struct btrfs_block_group_cache *cache = NULL;
4084         struct btrfs_fs_info *info = root->fs_info;
4085         u64 total = num_bytes;
4086         u64 old_val;
4087         u64 byte_in_group;
4088         int factor;
4089
4090         /* block accounting for super block */
4091         spin_lock(&info->delalloc_lock);
4092         old_val = btrfs_super_bytes_used(&info->super_copy);
4093         if (alloc)
4094                 old_val += num_bytes;
4095         else
4096                 old_val -= num_bytes;
4097         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4098         spin_unlock(&info->delalloc_lock);
4099
4100         while (total) {
4101                 cache = btrfs_lookup_block_group(info, bytenr);
4102                 if (!cache)
4103                         return -1;
4104                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4105                                     BTRFS_BLOCK_GROUP_RAID1 |
4106                                     BTRFS_BLOCK_GROUP_RAID10))
4107                         factor = 2;
4108                 else
4109                         factor = 1;
4110                 /*
4111                  * If this block group has free space cache written out, we
4112                  * need to make sure to load it if we are removing space.  This
4113                  * is because we need the unpinning stage to actually add the
4114                  * space back to the block group, otherwise we will leak space.
4115                  */
4116                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4117                         cache_block_group(cache, trans, NULL, 1);
4118
4119                 byte_in_group = bytenr - cache->key.objectid;
4120                 WARN_ON(byte_in_group > cache->key.offset);
4121
4122                 spin_lock(&cache->space_info->lock);
4123                 spin_lock(&cache->lock);
4124
4125                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4126                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4127                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4128
4129                 cache->dirty = 1;
4130                 old_val = btrfs_block_group_used(&cache->item);
4131                 num_bytes = min(total, cache->key.offset - byte_in_group);
4132                 if (alloc) {
4133                         old_val += num_bytes;
4134                         btrfs_set_block_group_used(&cache->item, old_val);
4135                         cache->reserved -= num_bytes;
4136                         cache->space_info->bytes_reserved -= num_bytes;
4137                         cache->space_info->bytes_used += num_bytes;
4138                         cache->space_info->disk_used += num_bytes * factor;
4139                         spin_unlock(&cache->lock);
4140                         spin_unlock(&cache->space_info->lock);
4141                 } else {
4142                         old_val -= num_bytes;
4143                         btrfs_set_block_group_used(&cache->item, old_val);
4144                         cache->pinned += num_bytes;
4145                         cache->space_info->bytes_pinned += num_bytes;
4146                         cache->space_info->bytes_used -= num_bytes;
4147                         cache->space_info->disk_used -= num_bytes * factor;
4148                         spin_unlock(&cache->lock);
4149                         spin_unlock(&cache->space_info->lock);
4150
4151                         set_extent_dirty(info->pinned_extents,
4152                                          bytenr, bytenr + num_bytes - 1,
4153                                          GFP_NOFS | __GFP_NOFAIL);
4154                 }
4155                 btrfs_put_block_group(cache);
4156                 total -= num_bytes;
4157                 bytenr += num_bytes;
4158         }
4159         return 0;
4160 }
4161
4162 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4163 {
4164         struct btrfs_block_group_cache *cache;
4165         u64 bytenr;
4166
4167         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4168         if (!cache)
4169                 return 0;
4170
4171         bytenr = cache->key.objectid;
4172         btrfs_put_block_group(cache);
4173
4174         return bytenr;
4175 }
4176
4177 static int pin_down_extent(struct btrfs_root *root,
4178                            struct btrfs_block_group_cache *cache,
4179                            u64 bytenr, u64 num_bytes, int reserved)
4180 {
4181         spin_lock(&cache->space_info->lock);
4182         spin_lock(&cache->lock);
4183         cache->pinned += num_bytes;
4184         cache->space_info->bytes_pinned += num_bytes;
4185         if (reserved) {
4186                 cache->reserved -= num_bytes;
4187                 cache->space_info->bytes_reserved -= num_bytes;
4188         }
4189         spin_unlock(&cache->lock);
4190         spin_unlock(&cache->space_info->lock);
4191
4192         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4193                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4194         return 0;
4195 }
4196
4197 /*
4198  * this function must be called within transaction
4199  */
4200 int btrfs_pin_extent(struct btrfs_root *root,
4201                      u64 bytenr, u64 num_bytes, int reserved)
4202 {
4203         struct btrfs_block_group_cache *cache;
4204
4205         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4206         BUG_ON(!cache);
4207
4208         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4209
4210         btrfs_put_block_group(cache);
4211         return 0;
4212 }
4213
4214 /*
4215  * update size of reserved extents. this function may return -EAGAIN
4216  * if 'reserve' is true or 'sinfo' is false.
4217  */
4218 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4219                                  u64 num_bytes, int reserve, int sinfo)
4220 {
4221         int ret = 0;
4222         if (sinfo) {
4223                 struct btrfs_space_info *space_info = cache->space_info;
4224                 spin_lock(&space_info->lock);
4225                 spin_lock(&cache->lock);
4226                 if (reserve) {
4227                         if (cache->ro) {
4228                                 ret = -EAGAIN;
4229                         } else {
4230                                 cache->reserved += num_bytes;
4231                                 space_info->bytes_reserved += num_bytes;
4232                         }
4233                 } else {
4234                         if (cache->ro)
4235                                 space_info->bytes_readonly += num_bytes;
4236                         cache->reserved -= num_bytes;
4237                         space_info->bytes_reserved -= num_bytes;
4238                 }
4239                 spin_unlock(&cache->lock);
4240                 spin_unlock(&space_info->lock);
4241         } else {
4242                 spin_lock(&cache->lock);
4243                 if (cache->ro) {
4244                         ret = -EAGAIN;
4245                 } else {
4246                         if (reserve)
4247                                 cache->reserved += num_bytes;
4248                         else
4249                                 cache->reserved -= num_bytes;
4250                 }
4251                 spin_unlock(&cache->lock);
4252         }
4253         return ret;
4254 }
4255
4256 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4257                                 struct btrfs_root *root)
4258 {
4259         struct btrfs_fs_info *fs_info = root->fs_info;
4260         struct btrfs_caching_control *next;
4261         struct btrfs_caching_control *caching_ctl;
4262         struct btrfs_block_group_cache *cache;
4263
4264         down_write(&fs_info->extent_commit_sem);
4265
4266         list_for_each_entry_safe(caching_ctl, next,
4267                                  &fs_info->caching_block_groups, list) {
4268                 cache = caching_ctl->block_group;
4269                 if (block_group_cache_done(cache)) {
4270                         cache->last_byte_to_unpin = (u64)-1;
4271                         list_del_init(&caching_ctl->list);
4272                         put_caching_control(caching_ctl);
4273                 } else {
4274                         cache->last_byte_to_unpin = caching_ctl->progress;
4275                 }
4276         }
4277
4278         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4279                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4280         else
4281                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4282
4283         up_write(&fs_info->extent_commit_sem);
4284
4285         update_global_block_rsv(fs_info);
4286         return 0;
4287 }
4288
4289 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4290 {
4291         struct btrfs_fs_info *fs_info = root->fs_info;
4292         struct btrfs_block_group_cache *cache = NULL;
4293         u64 len;
4294
4295         while (start <= end) {
4296                 if (!cache ||
4297                     start >= cache->key.objectid + cache->key.offset) {
4298                         if (cache)
4299                                 btrfs_put_block_group(cache);
4300                         cache = btrfs_lookup_block_group(fs_info, start);
4301                         BUG_ON(!cache);
4302                 }
4303
4304                 len = cache->key.objectid + cache->key.offset - start;
4305                 len = min(len, end + 1 - start);
4306
4307                 if (start < cache->last_byte_to_unpin) {
4308                         len = min(len, cache->last_byte_to_unpin - start);
4309                         btrfs_add_free_space(cache, start, len);
4310                 }
4311
4312                 start += len;
4313
4314                 spin_lock(&cache->space_info->lock);
4315                 spin_lock(&cache->lock);
4316                 cache->pinned -= len;
4317                 cache->space_info->bytes_pinned -= len;
4318                 if (cache->ro) {
4319                         cache->space_info->bytes_readonly += len;
4320                 } else if (cache->reserved_pinned > 0) {
4321                         len = min(len, cache->reserved_pinned);
4322                         cache->reserved_pinned -= len;
4323                         cache->space_info->bytes_reserved += len;
4324                 }
4325                 spin_unlock(&cache->lock);
4326                 spin_unlock(&cache->space_info->lock);
4327         }
4328
4329         if (cache)
4330                 btrfs_put_block_group(cache);
4331         return 0;
4332 }
4333
4334 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4335                                struct btrfs_root *root)
4336 {
4337         struct btrfs_fs_info *fs_info = root->fs_info;
4338         struct extent_io_tree *unpin;
4339         struct btrfs_block_rsv *block_rsv;
4340         struct btrfs_block_rsv *next_rsv;
4341         u64 start;
4342         u64 end;
4343         int idx;
4344         int ret;
4345
4346         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4347                 unpin = &fs_info->freed_extents[1];
4348         else
4349                 unpin = &fs_info->freed_extents[0];
4350
4351         while (1) {
4352                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4353                                             EXTENT_DIRTY);
4354                 if (ret)
4355                         break;
4356
4357                 ret = btrfs_discard_extent(root, start, end + 1 - start);
4358
4359                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4360                 unpin_extent_range(root, start, end);
4361                 cond_resched();
4362         }
4363
4364         mutex_lock(&fs_info->durable_block_rsv_mutex);
4365         list_for_each_entry_safe(block_rsv, next_rsv,
4366                                  &fs_info->durable_block_rsv_list, list) {
4367
4368                 idx = trans->transid & 0x1;
4369                 if (block_rsv->freed[idx] > 0) {
4370                         block_rsv_add_bytes(block_rsv,
4371                                             block_rsv->freed[idx], 0);
4372                         block_rsv->freed[idx] = 0;
4373                 }
4374                 if (atomic_read(&block_rsv->usage) == 0) {
4375                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4376
4377                         if (block_rsv->freed[0] == 0 &&
4378                             block_rsv->freed[1] == 0) {
4379                                 list_del_init(&block_rsv->list);
4380                                 kfree(block_rsv);
4381                         }
4382                 } else {
4383                         btrfs_block_rsv_release(root, block_rsv, 0);
4384                 }
4385         }
4386         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4387
4388         return 0;
4389 }
4390
4391 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4392                                 struct btrfs_root *root,
4393                                 u64 bytenr, u64 num_bytes, u64 parent,
4394                                 u64 root_objectid, u64 owner_objectid,
4395                                 u64 owner_offset, int refs_to_drop,
4396                                 struct btrfs_delayed_extent_op *extent_op)
4397 {
4398         struct btrfs_key key;
4399         struct btrfs_path *path;
4400         struct btrfs_fs_info *info = root->fs_info;
4401         struct btrfs_root *extent_root = info->extent_root;
4402         struct extent_buffer *leaf;
4403         struct btrfs_extent_item *ei;
4404         struct btrfs_extent_inline_ref *iref;
4405         int ret;
4406         int is_data;
4407         int extent_slot = 0;
4408         int found_extent = 0;
4409         int num_to_del = 1;
4410         u32 item_size;
4411         u64 refs;
4412
4413         path = btrfs_alloc_path();
4414         if (!path)
4415                 return -ENOMEM;
4416
4417         path->reada = 1;
4418         path->leave_spinning = 1;
4419
4420         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4421         BUG_ON(!is_data && refs_to_drop != 1);
4422
4423         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4424                                     bytenr, num_bytes, parent,
4425                                     root_objectid, owner_objectid,
4426                                     owner_offset);
4427         if (ret == 0) {
4428                 extent_slot = path->slots[0];
4429                 while (extent_slot >= 0) {
4430                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4431                                               extent_slot);
4432                         if (key.objectid != bytenr)
4433                                 break;
4434                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4435                             key.offset == num_bytes) {
4436                                 found_extent = 1;
4437                                 break;
4438                         }
4439                         if (path->slots[0] - extent_slot > 5)
4440                                 break;
4441                         extent_slot--;
4442                 }
4443 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4444                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4445                 if (found_extent && item_size < sizeof(*ei))
4446                         found_extent = 0;
4447 #endif
4448                 if (!found_extent) {
4449                         BUG_ON(iref);
4450                         ret = remove_extent_backref(trans, extent_root, path,
4451                                                     NULL, refs_to_drop,
4452                                                     is_data);
4453                         BUG_ON(ret);
4454                         btrfs_release_path(extent_root, path);
4455                         path->leave_spinning = 1;
4456
4457                         key.objectid = bytenr;
4458                         key.type = BTRFS_EXTENT_ITEM_KEY;
4459                         key.offset = num_bytes;
4460
4461                         ret = btrfs_search_slot(trans, extent_root,
4462                                                 &key, path, -1, 1);
4463                         if (ret) {
4464                                 printk(KERN_ERR "umm, got %d back from search"
4465                                        ", was looking for %llu\n", ret,
4466                                        (unsigned long long)bytenr);
4467                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4468                         }
4469                         BUG_ON(ret);
4470                         extent_slot = path->slots[0];
4471                 }
4472         } else {
4473                 btrfs_print_leaf(extent_root, path->nodes[0]);
4474                 WARN_ON(1);
4475                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4476                        "parent %llu root %llu  owner %llu offset %llu\n",
4477                        (unsigned long long)bytenr,
4478                        (unsigned long long)parent,
4479                        (unsigned long long)root_objectid,
4480                        (unsigned long long)owner_objectid,
4481                        (unsigned long long)owner_offset);
4482         }
4483
4484         leaf = path->nodes[0];
4485         item_size = btrfs_item_size_nr(leaf, extent_slot);
4486 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4487         if (item_size < sizeof(*ei)) {
4488                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4489                 ret = convert_extent_item_v0(trans, extent_root, path,
4490                                              owner_objectid, 0);
4491                 BUG_ON(ret < 0);
4492
4493                 btrfs_release_path(extent_root, path);
4494                 path->leave_spinning = 1;
4495
4496                 key.objectid = bytenr;
4497                 key.type = BTRFS_EXTENT_ITEM_KEY;
4498                 key.offset = num_bytes;
4499
4500                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4501                                         -1, 1);
4502                 if (ret) {
4503                         printk(KERN_ERR "umm, got %d back from search"
4504                                ", was looking for %llu\n", ret,
4505                                (unsigned long long)bytenr);
4506                         btrfs_print_leaf(extent_root, path->nodes[0]);
4507                 }
4508                 BUG_ON(ret);
4509                 extent_slot = path->slots[0];
4510                 leaf = path->nodes[0];
4511                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4512         }
4513 #endif
4514         BUG_ON(item_size < sizeof(*ei));
4515         ei = btrfs_item_ptr(leaf, extent_slot,
4516                             struct btrfs_extent_item);
4517         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4518                 struct btrfs_tree_block_info *bi;
4519                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4520                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4521                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4522         }
4523
4524         refs = btrfs_extent_refs(leaf, ei);
4525         BUG_ON(refs < refs_to_drop);
4526         refs -= refs_to_drop;
4527
4528         if (refs > 0) {
4529                 if (extent_op)
4530                         __run_delayed_extent_op(extent_op, leaf, ei);
4531                 /*
4532                  * In the case of inline back ref, reference count will
4533                  * be updated by remove_extent_backref
4534                  */
4535                 if (iref) {
4536                         BUG_ON(!found_extent);
4537                 } else {
4538                         btrfs_set_extent_refs(leaf, ei, refs);
4539                         btrfs_mark_buffer_dirty(leaf);
4540                 }
4541                 if (found_extent) {
4542                         ret = remove_extent_backref(trans, extent_root, path,
4543                                                     iref, refs_to_drop,
4544                                                     is_data);
4545                         BUG_ON(ret);
4546                 }
4547         } else {
4548                 if (found_extent) {
4549                         BUG_ON(is_data && refs_to_drop !=
4550                                extent_data_ref_count(root, path, iref));
4551                         if (iref) {
4552                                 BUG_ON(path->slots[0] != extent_slot);
4553                         } else {
4554                                 BUG_ON(path->slots[0] != extent_slot + 1);
4555                                 path->slots[0] = extent_slot;
4556                                 num_to_del = 2;
4557                         }
4558                 }
4559
4560                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4561                                       num_to_del);
4562                 BUG_ON(ret);
4563                 btrfs_release_path(extent_root, path);
4564
4565                 if (is_data) {
4566                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4567                         BUG_ON(ret);
4568                 } else {
4569                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4570                              bytenr >> PAGE_CACHE_SHIFT,
4571                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4572                 }
4573
4574                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4575                 BUG_ON(ret);
4576         }
4577         btrfs_free_path(path);
4578         return ret;
4579 }
4580
4581 /*
4582  * when we free an block, it is possible (and likely) that we free the last
4583  * delayed ref for that extent as well.  This searches the delayed ref tree for
4584  * a given extent, and if there are no other delayed refs to be processed, it
4585  * removes it from the tree.
4586  */
4587 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4588                                       struct btrfs_root *root, u64 bytenr)
4589 {
4590         struct btrfs_delayed_ref_head *head;
4591         struct btrfs_delayed_ref_root *delayed_refs;
4592         struct btrfs_delayed_ref_node *ref;
4593         struct rb_node *node;
4594         int ret = 0;
4595
4596         delayed_refs = &trans->transaction->delayed_refs;
4597         spin_lock(&delayed_refs->lock);
4598         head = btrfs_find_delayed_ref_head(trans, bytenr);
4599         if (!head)
4600                 goto out;
4601
4602         node = rb_prev(&head->node.rb_node);
4603         if (!node)
4604                 goto out;
4605
4606         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4607
4608         /* there are still entries for this ref, we can't drop it */
4609         if (ref->bytenr == bytenr)
4610                 goto out;
4611
4612         if (head->extent_op) {
4613                 if (!head->must_insert_reserved)
4614                         goto out;
4615                 kfree(head->extent_op);
4616                 head->extent_op = NULL;
4617         }
4618
4619         /*
4620          * waiting for the lock here would deadlock.  If someone else has it
4621          * locked they are already in the process of dropping it anyway
4622          */
4623         if (!mutex_trylock(&head->mutex))
4624                 goto out;
4625
4626         /*
4627          * at this point we have a head with no other entries.  Go
4628          * ahead and process it.
4629          */
4630         head->node.in_tree = 0;
4631         rb_erase(&head->node.rb_node, &delayed_refs->root);
4632
4633         delayed_refs->num_entries--;
4634
4635         /*
4636          * we don't take a ref on the node because we're removing it from the
4637          * tree, so we just steal the ref the tree was holding.
4638          */
4639         delayed_refs->num_heads--;
4640         if (list_empty(&head->cluster))
4641                 delayed_refs->num_heads_ready--;
4642
4643         list_del_init(&head->cluster);
4644         spin_unlock(&delayed_refs->lock);
4645
4646         BUG_ON(head->extent_op);
4647         if (head->must_insert_reserved)
4648                 ret = 1;
4649
4650         mutex_unlock(&head->mutex);
4651         btrfs_put_delayed_ref(&head->node);
4652         return ret;
4653 out:
4654         spin_unlock(&delayed_refs->lock);
4655         return 0;
4656 }
4657
4658 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4659                            struct btrfs_root *root,
4660                            struct extent_buffer *buf,
4661                            u64 parent, int last_ref)
4662 {
4663         struct btrfs_block_rsv *block_rsv;
4664         struct btrfs_block_group_cache *cache = NULL;
4665         int ret;
4666
4667         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4668                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4669                                                 parent, root->root_key.objectid,
4670                                                 btrfs_header_level(buf),
4671                                                 BTRFS_DROP_DELAYED_REF, NULL);
4672                 BUG_ON(ret);
4673         }
4674
4675         if (!last_ref)
4676                 return;
4677
4678         block_rsv = get_block_rsv(trans, root);
4679         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4680         if (block_rsv->space_info != cache->space_info)
4681                 goto out;
4682
4683         if (btrfs_header_generation(buf) == trans->transid) {
4684                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4685                         ret = check_ref_cleanup(trans, root, buf->start);
4686                         if (!ret)
4687                                 goto pin;
4688                 }
4689
4690                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4691                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4692                         goto pin;
4693                 }
4694
4695                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4696
4697                 btrfs_add_free_space(cache, buf->start, buf->len);
4698                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4699                 if (ret == -EAGAIN) {
4700                         /* block group became read-only */
4701                         update_reserved_bytes(cache, buf->len, 0, 1);
4702                         goto out;
4703                 }
4704
4705                 ret = 1;
4706                 spin_lock(&block_rsv->lock);
4707                 if (block_rsv->reserved < block_rsv->size) {
4708                         block_rsv->reserved += buf->len;
4709                         ret = 0;
4710                 }
4711                 spin_unlock(&block_rsv->lock);
4712
4713                 if (ret) {
4714                         spin_lock(&cache->space_info->lock);
4715                         cache->space_info->bytes_reserved -= buf->len;
4716                         spin_unlock(&cache->space_info->lock);
4717                 }
4718                 goto out;
4719         }
4720 pin:
4721         if (block_rsv->durable && !cache->ro) {
4722                 ret = 0;
4723                 spin_lock(&cache->lock);
4724                 if (!cache->ro) {
4725                         cache->reserved_pinned += buf->len;
4726                         ret = 1;
4727                 }
4728                 spin_unlock(&cache->lock);
4729
4730                 if (ret) {
4731                         spin_lock(&block_rsv->lock);
4732                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4733                         spin_unlock(&block_rsv->lock);
4734                 }
4735         }
4736 out:
4737         btrfs_put_block_group(cache);
4738 }
4739
4740 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4741                       struct btrfs_root *root,
4742                       u64 bytenr, u64 num_bytes, u64 parent,
4743                       u64 root_objectid, u64 owner, u64 offset)
4744 {
4745         int ret;
4746
4747         /*
4748          * tree log blocks never actually go into the extent allocation
4749          * tree, just update pinning info and exit early.
4750          */
4751         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4752                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4753                 /* unlocks the pinned mutex */
4754                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4755                 ret = 0;
4756         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4757                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4758                                         parent, root_objectid, (int)owner,
4759                                         BTRFS_DROP_DELAYED_REF, NULL);
4760                 BUG_ON(ret);
4761         } else {
4762                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4763                                         parent, root_objectid, owner,
4764                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4765                 BUG_ON(ret);
4766         }
4767         return ret;
4768 }
4769
4770 static u64 stripe_align(struct btrfs_root *root, u64 val)
4771 {
4772         u64 mask = ((u64)root->stripesize - 1);
4773         u64 ret = (val + mask) & ~mask;
4774         return ret;
4775 }
4776
4777 /*
4778  * when we wait for progress in the block group caching, its because
4779  * our allocation attempt failed at least once.  So, we must sleep
4780  * and let some progress happen before we try again.
4781  *
4782  * This function will sleep at least once waiting for new free space to
4783  * show up, and then it will check the block group free space numbers
4784  * for our min num_bytes.  Another option is to have it go ahead
4785  * and look in the rbtree for a free extent of a given size, but this
4786  * is a good start.
4787  */
4788 static noinline int
4789 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4790                                 u64 num_bytes)
4791 {
4792         struct btrfs_caching_control *caching_ctl;
4793         DEFINE_WAIT(wait);
4794
4795         caching_ctl = get_caching_control(cache);
4796         if (!caching_ctl)
4797                 return 0;
4798
4799         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4800                    (cache->free_space >= num_bytes));
4801
4802         put_caching_control(caching_ctl);
4803         return 0;
4804 }
4805
4806 static noinline int
4807 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4808 {
4809         struct btrfs_caching_control *caching_ctl;
4810         DEFINE_WAIT(wait);
4811
4812         caching_ctl = get_caching_control(cache);
4813         if (!caching_ctl)
4814                 return 0;
4815
4816         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4817
4818         put_caching_control(caching_ctl);
4819         return 0;
4820 }
4821
4822 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4823 {
4824         int index;
4825         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4826                 index = 0;
4827         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4828                 index = 1;
4829         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4830                 index = 2;
4831         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4832                 index = 3;
4833         else
4834                 index = 4;
4835         return index;
4836 }
4837
4838 enum btrfs_loop_type {
4839         LOOP_FIND_IDEAL = 0,
4840         LOOP_CACHING_NOWAIT = 1,
4841         LOOP_CACHING_WAIT = 2,
4842         LOOP_ALLOC_CHUNK = 3,
4843         LOOP_NO_EMPTY_SIZE = 4,
4844 };
4845
4846 /*
4847  * walks the btree of allocated extents and find a hole of a given size.
4848  * The key ins is changed to record the hole:
4849  * ins->objectid == block start
4850  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4851  * ins->offset == number of blocks
4852  * Any available blocks before search_start are skipped.
4853  */
4854 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4855                                      struct btrfs_root *orig_root,
4856                                      u64 num_bytes, u64 empty_size,
4857                                      u64 search_start, u64 search_end,
4858                                      u64 hint_byte, struct btrfs_key *ins,
4859                                      int data)
4860 {
4861         int ret = 0;
4862         struct btrfs_root *root = orig_root->fs_info->extent_root;
4863         struct btrfs_free_cluster *last_ptr = NULL;
4864         struct btrfs_block_group_cache *block_group = NULL;
4865         int empty_cluster = 2 * 1024 * 1024;
4866         int allowed_chunk_alloc = 0;
4867         int done_chunk_alloc = 0;
4868         struct btrfs_space_info *space_info;
4869         int last_ptr_loop = 0;
4870         int loop = 0;
4871         int index = 0;
4872         bool found_uncached_bg = false;
4873         bool failed_cluster_refill = false;
4874         bool failed_alloc = false;
4875         bool use_cluster = true;
4876         u64 ideal_cache_percent = 0;
4877         u64 ideal_cache_offset = 0;
4878
4879         WARN_ON(num_bytes < root->sectorsize);
4880         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4881         ins->objectid = 0;
4882         ins->offset = 0;
4883
4884         space_info = __find_space_info(root->fs_info, data);
4885         if (!space_info) {
4886                 printk(KERN_ERR "No space info for %d\n", data);
4887                 return -ENOSPC;
4888         }
4889
4890         /*
4891          * If the space info is for both data and metadata it means we have a
4892          * small filesystem and we can't use the clustering stuff.
4893          */
4894         if (btrfs_mixed_space_info(space_info))
4895                 use_cluster = false;
4896
4897         if (orig_root->ref_cows || empty_size)
4898                 allowed_chunk_alloc = 1;
4899
4900         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4901                 last_ptr = &root->fs_info->meta_alloc_cluster;
4902                 if (!btrfs_test_opt(root, SSD))
4903                         empty_cluster = 64 * 1024;
4904         }
4905
4906         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4907             btrfs_test_opt(root, SSD)) {
4908                 last_ptr = &root->fs_info->data_alloc_cluster;
4909         }
4910
4911         if (last_ptr) {
4912                 spin_lock(&last_ptr->lock);
4913                 if (last_ptr->block_group)
4914                         hint_byte = last_ptr->window_start;
4915                 spin_unlock(&last_ptr->lock);
4916         }
4917
4918         search_start = max(search_start, first_logical_byte(root, 0));
4919         search_start = max(search_start, hint_byte);
4920
4921         if (!last_ptr)
4922                 empty_cluster = 0;
4923
4924         if (search_start == hint_byte) {
4925 ideal_cache:
4926                 block_group = btrfs_lookup_block_group(root->fs_info,
4927                                                        search_start);
4928                 /*
4929                  * we don't want to use the block group if it doesn't match our
4930                  * allocation bits, or if its not cached.
4931                  *
4932                  * However if we are re-searching with an ideal block group
4933                  * picked out then we don't care that the block group is cached.
4934                  */
4935                 if (block_group && block_group_bits(block_group, data) &&
4936                     (block_group->cached != BTRFS_CACHE_NO ||
4937                      search_start == ideal_cache_offset)) {
4938                         down_read(&space_info->groups_sem);
4939                         if (list_empty(&block_group->list) ||
4940                             block_group->ro) {
4941                                 /*
4942                                  * someone is removing this block group,
4943                                  * we can't jump into the have_block_group
4944                                  * target because our list pointers are not
4945                                  * valid
4946                                  */
4947                                 btrfs_put_block_group(block_group);
4948                                 up_read(&space_info->groups_sem);
4949                         } else {
4950                                 index = get_block_group_index(block_group);
4951                                 goto have_block_group;
4952                         }
4953                 } else if (block_group) {
4954                         btrfs_put_block_group(block_group);
4955                 }
4956         }
4957 search:
4958         down_read(&space_info->groups_sem);
4959         list_for_each_entry(block_group, &space_info->block_groups[index],
4960                             list) {
4961                 u64 offset;
4962                 int cached;
4963
4964                 btrfs_get_block_group(block_group);
4965                 search_start = block_group->key.objectid;
4966
4967                 /*
4968                  * this can happen if we end up cycling through all the
4969                  * raid types, but we want to make sure we only allocate
4970                  * for the proper type.
4971                  */
4972                 if (!block_group_bits(block_group, data)) {
4973                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
4974                                 BTRFS_BLOCK_GROUP_RAID1 |
4975                                 BTRFS_BLOCK_GROUP_RAID10;
4976
4977                         /*
4978                          * if they asked for extra copies and this block group
4979                          * doesn't provide them, bail.  This does allow us to
4980                          * fill raid0 from raid1.
4981                          */
4982                         if ((data & extra) && !(block_group->flags & extra))
4983                                 goto loop;
4984                 }
4985
4986 have_block_group:
4987                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4988                         u64 free_percent;
4989
4990                         ret = cache_block_group(block_group, trans,
4991                                                 orig_root, 1);
4992                         if (block_group->cached == BTRFS_CACHE_FINISHED)
4993                                 goto have_block_group;
4994
4995                         free_percent = btrfs_block_group_used(&block_group->item);
4996                         free_percent *= 100;
4997                         free_percent = div64_u64(free_percent,
4998                                                  block_group->key.offset);
4999                         free_percent = 100 - free_percent;
5000                         if (free_percent > ideal_cache_percent &&
5001                             likely(!block_group->ro)) {
5002                                 ideal_cache_offset = block_group->key.objectid;
5003                                 ideal_cache_percent = free_percent;
5004                         }
5005
5006                         /*
5007                          * We only want to start kthread caching if we are at
5008                          * the point where we will wait for caching to make
5009                          * progress, or if our ideal search is over and we've
5010                          * found somebody to start caching.
5011                          */
5012                         if (loop > LOOP_CACHING_NOWAIT ||
5013                             (loop > LOOP_FIND_IDEAL &&
5014                              atomic_read(&space_info->caching_threads) < 2)) {
5015                                 ret = cache_block_group(block_group, trans,
5016                                                         orig_root, 0);
5017                                 BUG_ON(ret);
5018                         }
5019                         found_uncached_bg = true;
5020
5021                         /*
5022                          * If loop is set for cached only, try the next block
5023                          * group.
5024                          */
5025                         if (loop == LOOP_FIND_IDEAL)
5026                                 goto loop;
5027                 }
5028
5029                 cached = block_group_cache_done(block_group);
5030                 if (unlikely(!cached))
5031                         found_uncached_bg = true;
5032
5033                 if (unlikely(block_group->ro))
5034                         goto loop;
5035
5036                 /*
5037                  * Ok we want to try and use the cluster allocator, so lets look
5038                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5039                  * have tried the cluster allocator plenty of times at this
5040                  * point and not have found anything, so we are likely way too
5041                  * fragmented for the clustering stuff to find anything, so lets
5042                  * just skip it and let the allocator find whatever block it can
5043                  * find
5044                  */
5045                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5046                         /*
5047                          * the refill lock keeps out other
5048                          * people trying to start a new cluster
5049                          */
5050                         spin_lock(&last_ptr->refill_lock);
5051                         if (last_ptr->block_group &&
5052                             (last_ptr->block_group->ro ||
5053                             !block_group_bits(last_ptr->block_group, data))) {
5054                                 offset = 0;
5055                                 goto refill_cluster;
5056                         }
5057
5058                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5059                                                  num_bytes, search_start);
5060                         if (offset) {
5061                                 /* we have a block, we're done */
5062                                 spin_unlock(&last_ptr->refill_lock);
5063                                 goto checks;
5064                         }
5065
5066                         spin_lock(&last_ptr->lock);
5067                         /*
5068                          * whoops, this cluster doesn't actually point to
5069                          * this block group.  Get a ref on the block
5070                          * group is does point to and try again
5071                          */
5072                         if (!last_ptr_loop && last_ptr->block_group &&
5073                             last_ptr->block_group != block_group) {
5074
5075                                 btrfs_put_block_group(block_group);
5076                                 block_group = last_ptr->block_group;
5077                                 btrfs_get_block_group(block_group);
5078                                 spin_unlock(&last_ptr->lock);
5079                                 spin_unlock(&last_ptr->refill_lock);
5080
5081                                 last_ptr_loop = 1;
5082                                 search_start = block_group->key.objectid;
5083                                 /*
5084                                  * we know this block group is properly
5085                                  * in the list because
5086                                  * btrfs_remove_block_group, drops the
5087                                  * cluster before it removes the block
5088                                  * group from the list
5089                                  */
5090                                 goto have_block_group;
5091                         }
5092                         spin_unlock(&last_ptr->lock);
5093 refill_cluster:
5094                         /*
5095                          * this cluster didn't work out, free it and
5096                          * start over
5097                          */
5098                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5099
5100                         last_ptr_loop = 0;
5101
5102                         /* allocate a cluster in this block group */
5103                         ret = btrfs_find_space_cluster(trans, root,
5104                                                block_group, last_ptr,
5105                                                offset, num_bytes,
5106                                                empty_cluster + empty_size);
5107                         if (ret == 0) {
5108                                 /*
5109                                  * now pull our allocation out of this
5110                                  * cluster
5111                                  */
5112                                 offset = btrfs_alloc_from_cluster(block_group,
5113                                                   last_ptr, num_bytes,
5114                                                   search_start);
5115                                 if (offset) {
5116                                         /* we found one, proceed */
5117                                         spin_unlock(&last_ptr->refill_lock);
5118                                         goto checks;
5119                                 }
5120                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5121                                    && !failed_cluster_refill) {
5122                                 spin_unlock(&last_ptr->refill_lock);
5123
5124                                 failed_cluster_refill = true;
5125                                 wait_block_group_cache_progress(block_group,
5126                                        num_bytes + empty_cluster + empty_size);
5127                                 goto have_block_group;
5128                         }
5129
5130                         /*
5131                          * at this point we either didn't find a cluster
5132                          * or we weren't able to allocate a block from our
5133                          * cluster.  Free the cluster we've been trying
5134                          * to use, and go to the next block group
5135                          */
5136                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5137                         spin_unlock(&last_ptr->refill_lock);
5138                         goto loop;
5139                 }
5140
5141                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5142                                                     num_bytes, empty_size);
5143                 /*
5144                  * If we didn't find a chunk, and we haven't failed on this
5145                  * block group before, and this block group is in the middle of
5146                  * caching and we are ok with waiting, then go ahead and wait
5147                  * for progress to be made, and set failed_alloc to true.
5148                  *
5149                  * If failed_alloc is true then we've already waited on this
5150                  * block group once and should move on to the next block group.
5151                  */
5152                 if (!offset && !failed_alloc && !cached &&
5153                     loop > LOOP_CACHING_NOWAIT) {
5154                         wait_block_group_cache_progress(block_group,
5155                                                 num_bytes + empty_size);
5156                         failed_alloc = true;
5157                         goto have_block_group;
5158                 } else if (!offset) {
5159                         goto loop;
5160                 }
5161 checks:
5162                 search_start = stripe_align(root, offset);
5163                 /* move on to the next group */
5164                 if (search_start + num_bytes >= search_end) {
5165                         btrfs_add_free_space(block_group, offset, num_bytes);
5166                         goto loop;
5167                 }
5168
5169                 /* move on to the next group */
5170                 if (search_start + num_bytes >
5171                     block_group->key.objectid + block_group->key.offset) {
5172                         btrfs_add_free_space(block_group, offset, num_bytes);
5173                         goto loop;
5174                 }
5175
5176                 ins->objectid = search_start;
5177                 ins->offset = num_bytes;
5178
5179                 if (offset < search_start)
5180                         btrfs_add_free_space(block_group, offset,
5181                                              search_start - offset);
5182                 BUG_ON(offset > search_start);
5183
5184                 ret = update_reserved_bytes(block_group, num_bytes, 1,
5185                                             (data & BTRFS_BLOCK_GROUP_DATA));
5186                 if (ret == -EAGAIN) {
5187                         btrfs_add_free_space(block_group, offset, num_bytes);
5188                         goto loop;
5189                 }
5190
5191                 /* we are all good, lets return */
5192                 ins->objectid = search_start;
5193                 ins->offset = num_bytes;
5194
5195                 if (offset < search_start)
5196                         btrfs_add_free_space(block_group, offset,
5197                                              search_start - offset);
5198                 BUG_ON(offset > search_start);
5199                 break;
5200 loop:
5201                 failed_cluster_refill = false;
5202                 failed_alloc = false;
5203                 BUG_ON(index != get_block_group_index(block_group));
5204                 btrfs_put_block_group(block_group);
5205         }
5206         up_read(&space_info->groups_sem);
5207
5208         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5209                 goto search;
5210
5211         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5212          *                      for them to make caching progress.  Also
5213          *                      determine the best possible bg to cache
5214          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5215          *                      caching kthreads as we move along
5216          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5217          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5218          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5219          *                      again
5220          */
5221         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5222             (found_uncached_bg || empty_size || empty_cluster ||
5223              allowed_chunk_alloc)) {
5224                 index = 0;
5225                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5226                         found_uncached_bg = false;
5227                         loop++;
5228                         if (!ideal_cache_percent &&
5229                             atomic_read(&space_info->caching_threads))
5230                                 goto search;
5231
5232                         /*
5233                          * 1 of the following 2 things have happened so far
5234                          *
5235                          * 1) We found an ideal block group for caching that
5236                          * is mostly full and will cache quickly, so we might
5237                          * as well wait for it.
5238                          *
5239                          * 2) We searched for cached only and we didn't find
5240                          * anything, and we didn't start any caching kthreads
5241                          * either, so chances are we will loop through and
5242                          * start a couple caching kthreads, and then come back
5243                          * around and just wait for them.  This will be slower
5244                          * because we will have 2 caching kthreads reading at
5245                          * the same time when we could have just started one
5246                          * and waited for it to get far enough to give us an
5247                          * allocation, so go ahead and go to the wait caching
5248                          * loop.
5249                          */
5250                         loop = LOOP_CACHING_WAIT;
5251                         search_start = ideal_cache_offset;
5252                         ideal_cache_percent = 0;
5253                         goto ideal_cache;
5254                 } else if (loop == LOOP_FIND_IDEAL) {
5255                         /*
5256                          * Didn't find a uncached bg, wait on anything we find
5257                          * next.
5258                          */
5259                         loop = LOOP_CACHING_WAIT;
5260                         goto search;
5261                 }
5262
5263                 if (loop < LOOP_CACHING_WAIT) {
5264                         loop++;
5265                         goto search;
5266                 }
5267
5268                 if (loop == LOOP_ALLOC_CHUNK) {
5269                         empty_size = 0;
5270                         empty_cluster = 0;
5271                 }
5272
5273                 if (allowed_chunk_alloc) {
5274                         ret = do_chunk_alloc(trans, root, num_bytes +
5275                                              2 * 1024 * 1024, data, 1);
5276                         allowed_chunk_alloc = 0;
5277                         done_chunk_alloc = 1;
5278                 } else if (!done_chunk_alloc) {
5279                         space_info->force_alloc = 1;
5280                 }
5281
5282                 if (loop < LOOP_NO_EMPTY_SIZE) {
5283                         loop++;
5284                         goto search;
5285                 }
5286                 ret = -ENOSPC;
5287         } else if (!ins->objectid) {
5288                 ret = -ENOSPC;
5289         }
5290
5291         /* we found what we needed */
5292         if (ins->objectid) {
5293                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5294                         trans->block_group = block_group->key.objectid;
5295
5296                 btrfs_put_block_group(block_group);
5297                 ret = 0;
5298         }
5299
5300         return ret;
5301 }
5302
5303 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5304                             int dump_block_groups)
5305 {
5306         struct btrfs_block_group_cache *cache;
5307         int index = 0;
5308
5309         spin_lock(&info->lock);
5310         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5311                (unsigned long long)(info->total_bytes - info->bytes_used -
5312                                     info->bytes_pinned - info->bytes_reserved -
5313                                     info->bytes_readonly),
5314                (info->full) ? "" : "not ");
5315         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5316                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5317                (unsigned long long)info->total_bytes,
5318                (unsigned long long)info->bytes_used,
5319                (unsigned long long)info->bytes_pinned,
5320                (unsigned long long)info->bytes_reserved,
5321                (unsigned long long)info->bytes_may_use,
5322                (unsigned long long)info->bytes_readonly);
5323         spin_unlock(&info->lock);
5324
5325         if (!dump_block_groups)
5326                 return;
5327
5328         down_read(&info->groups_sem);
5329 again:
5330         list_for_each_entry(cache, &info->block_groups[index], list) {
5331                 spin_lock(&cache->lock);
5332                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5333                        "%llu pinned %llu reserved\n",
5334                        (unsigned long long)cache->key.objectid,
5335                        (unsigned long long)cache->key.offset,
5336                        (unsigned long long)btrfs_block_group_used(&cache->item),
5337                        (unsigned long long)cache->pinned,
5338                        (unsigned long long)cache->reserved);
5339                 btrfs_dump_free_space(cache, bytes);
5340                 spin_unlock(&cache->lock);
5341         }
5342         if (++index < BTRFS_NR_RAID_TYPES)
5343                 goto again;
5344         up_read(&info->groups_sem);
5345 }
5346
5347 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5348                          struct btrfs_root *root,
5349                          u64 num_bytes, u64 min_alloc_size,
5350                          u64 empty_size, u64 hint_byte,
5351                          u64 search_end, struct btrfs_key *ins,
5352                          u64 data)
5353 {
5354         int ret;
5355         u64 search_start = 0;
5356
5357         data = btrfs_get_alloc_profile(root, data);
5358 again:
5359         /*
5360          * the only place that sets empty_size is btrfs_realloc_node, which
5361          * is not called recursively on allocations
5362          */
5363         if (empty_size || root->ref_cows)
5364                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5365                                      num_bytes + 2 * 1024 * 1024, data, 0);
5366
5367         WARN_ON(num_bytes < root->sectorsize);
5368         ret = find_free_extent(trans, root, num_bytes, empty_size,
5369                                search_start, search_end, hint_byte,
5370                                ins, data);
5371
5372         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5373                 num_bytes = num_bytes >> 1;
5374                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5375                 num_bytes = max(num_bytes, min_alloc_size);
5376                 do_chunk_alloc(trans, root->fs_info->extent_root,
5377                                num_bytes, data, 1);
5378                 goto again;
5379         }
5380         if (ret == -ENOSPC) {
5381                 struct btrfs_space_info *sinfo;
5382
5383                 sinfo = __find_space_info(root->fs_info, data);
5384                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5385                        "wanted %llu\n", (unsigned long long)data,
5386                        (unsigned long long)num_bytes);
5387                 dump_space_info(sinfo, num_bytes, 1);
5388         }
5389
5390         return ret;
5391 }
5392
5393 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5394 {
5395         struct btrfs_block_group_cache *cache;
5396         int ret = 0;
5397
5398         cache = btrfs_lookup_block_group(root->fs_info, start);
5399         if (!cache) {
5400                 printk(KERN_ERR "Unable to find block group for %llu\n",
5401                        (unsigned long long)start);
5402                 return -ENOSPC;
5403         }
5404
5405         ret = btrfs_discard_extent(root, start, len);
5406
5407         btrfs_add_free_space(cache, start, len);
5408         update_reserved_bytes(cache, len, 0, 1);
5409         btrfs_put_block_group(cache);
5410
5411         return ret;
5412 }
5413
5414 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5415                                       struct btrfs_root *root,
5416                                       u64 parent, u64 root_objectid,
5417                                       u64 flags, u64 owner, u64 offset,
5418                                       struct btrfs_key *ins, int ref_mod)
5419 {
5420         int ret;
5421         struct btrfs_fs_info *fs_info = root->fs_info;
5422         struct btrfs_extent_item *extent_item;
5423         struct btrfs_extent_inline_ref *iref;
5424         struct btrfs_path *path;
5425         struct extent_buffer *leaf;
5426         int type;
5427         u32 size;
5428
5429         if (parent > 0)
5430                 type = BTRFS_SHARED_DATA_REF_KEY;
5431         else
5432                 type = BTRFS_EXTENT_DATA_REF_KEY;
5433
5434         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5435
5436         path = btrfs_alloc_path();
5437         BUG_ON(!path);
5438
5439         path->leave_spinning = 1;
5440         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5441                                       ins, size);
5442         BUG_ON(ret);
5443
5444         leaf = path->nodes[0];
5445         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5446                                      struct btrfs_extent_item);
5447         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5448         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5449         btrfs_set_extent_flags(leaf, extent_item,
5450                                flags | BTRFS_EXTENT_FLAG_DATA);
5451
5452         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5453         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5454         if (parent > 0) {
5455                 struct btrfs_shared_data_ref *ref;
5456                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5457                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5458                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5459         } else {
5460                 struct btrfs_extent_data_ref *ref;
5461                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5462                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5463                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5464                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5465                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5466         }
5467
5468         btrfs_mark_buffer_dirty(path->nodes[0]);
5469         btrfs_free_path(path);
5470
5471         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5472         if (ret) {
5473                 printk(KERN_ERR "btrfs update block group failed for %llu "
5474                        "%llu\n", (unsigned long long)ins->objectid,
5475                        (unsigned long long)ins->offset);
5476                 BUG();
5477         }
5478         return ret;
5479 }
5480
5481 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5482                                      struct btrfs_root *root,
5483                                      u64 parent, u64 root_objectid,
5484                                      u64 flags, struct btrfs_disk_key *key,
5485                                      int level, struct btrfs_key *ins)
5486 {
5487         int ret;
5488         struct btrfs_fs_info *fs_info = root->fs_info;
5489         struct btrfs_extent_item *extent_item;
5490         struct btrfs_tree_block_info *block_info;
5491         struct btrfs_extent_inline_ref *iref;
5492         struct btrfs_path *path;
5493         struct extent_buffer *leaf;
5494         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5495
5496         path = btrfs_alloc_path();
5497         BUG_ON(!path);
5498
5499         path->leave_spinning = 1;
5500         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5501                                       ins, size);
5502         BUG_ON(ret);
5503
5504         leaf = path->nodes[0];
5505         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5506                                      struct btrfs_extent_item);
5507         btrfs_set_extent_refs(leaf, extent_item, 1);
5508         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5509         btrfs_set_extent_flags(leaf, extent_item,
5510                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5511         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5512
5513         btrfs_set_tree_block_key(leaf, block_info, key);
5514         btrfs_set_tree_block_level(leaf, block_info, level);
5515
5516         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5517         if (parent > 0) {
5518                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5519                 btrfs_set_extent_inline_ref_type(leaf, iref,
5520                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5521                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5522         } else {
5523                 btrfs_set_extent_inline_ref_type(leaf, iref,
5524                                                  BTRFS_TREE_BLOCK_REF_KEY);
5525                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5526         }
5527
5528         btrfs_mark_buffer_dirty(leaf);
5529         btrfs_free_path(path);
5530
5531         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5532         if (ret) {
5533                 printk(KERN_ERR "btrfs update block group failed for %llu "
5534                        "%llu\n", (unsigned long long)ins->objectid,
5535                        (unsigned long long)ins->offset);
5536                 BUG();
5537         }
5538         return ret;
5539 }
5540
5541 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5542                                      struct btrfs_root *root,
5543                                      u64 root_objectid, u64 owner,
5544                                      u64 offset, struct btrfs_key *ins)
5545 {
5546         int ret;
5547
5548         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5549
5550         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5551                                          0, root_objectid, owner, offset,
5552                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5553         return ret;
5554 }
5555
5556 /*
5557  * this is used by the tree logging recovery code.  It records that
5558  * an extent has been allocated and makes sure to clear the free
5559  * space cache bits as well
5560  */
5561 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5562                                    struct btrfs_root *root,
5563                                    u64 root_objectid, u64 owner, u64 offset,
5564                                    struct btrfs_key *ins)
5565 {
5566         int ret;
5567         struct btrfs_block_group_cache *block_group;
5568         struct btrfs_caching_control *caching_ctl;
5569         u64 start = ins->objectid;
5570         u64 num_bytes = ins->offset;
5571
5572         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5573         cache_block_group(block_group, trans, NULL, 0);
5574         caching_ctl = get_caching_control(block_group);
5575
5576         if (!caching_ctl) {
5577                 BUG_ON(!block_group_cache_done(block_group));
5578                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5579                 BUG_ON(ret);
5580         } else {
5581                 mutex_lock(&caching_ctl->mutex);
5582
5583                 if (start >= caching_ctl->progress) {
5584                         ret = add_excluded_extent(root, start, num_bytes);
5585                         BUG_ON(ret);
5586                 } else if (start + num_bytes <= caching_ctl->progress) {
5587                         ret = btrfs_remove_free_space(block_group,
5588                                                       start, num_bytes);
5589                         BUG_ON(ret);
5590                 } else {
5591                         num_bytes = caching_ctl->progress - start;
5592                         ret = btrfs_remove_free_space(block_group,
5593                                                       start, num_bytes);
5594                         BUG_ON(ret);
5595
5596                         start = caching_ctl->progress;
5597                         num_bytes = ins->objectid + ins->offset -
5598                                     caching_ctl->progress;
5599                         ret = add_excluded_extent(root, start, num_bytes);
5600                         BUG_ON(ret);
5601                 }
5602
5603                 mutex_unlock(&caching_ctl->mutex);
5604                 put_caching_control(caching_ctl);
5605         }
5606
5607         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5608         BUG_ON(ret);
5609         btrfs_put_block_group(block_group);
5610         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5611                                          0, owner, offset, ins, 1);
5612         return ret;
5613 }
5614
5615 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5616                                             struct btrfs_root *root,
5617                                             u64 bytenr, u32 blocksize,
5618                                             int level)
5619 {
5620         struct extent_buffer *buf;
5621
5622         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5623         if (!buf)
5624                 return ERR_PTR(-ENOMEM);
5625         btrfs_set_header_generation(buf, trans->transid);
5626         btrfs_set_buffer_lockdep_class(buf, level);
5627         btrfs_tree_lock(buf);
5628         clean_tree_block(trans, root, buf);
5629
5630         btrfs_set_lock_blocking(buf);
5631         btrfs_set_buffer_uptodate(buf);
5632
5633         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5634                 /*
5635                  * we allow two log transactions at a time, use different
5636                  * EXENT bit to differentiate dirty pages.
5637                  */
5638                 if (root->log_transid % 2 == 0)
5639                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5640                                         buf->start + buf->len - 1, GFP_NOFS);
5641                 else
5642                         set_extent_new(&root->dirty_log_pages, buf->start,
5643                                         buf->start + buf->len - 1, GFP_NOFS);
5644         } else {
5645                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5646                          buf->start + buf->len - 1, GFP_NOFS);
5647         }
5648         trans->blocks_used++;
5649         /* this returns a buffer locked for blocking */
5650         return buf;
5651 }
5652
5653 static struct btrfs_block_rsv *
5654 use_block_rsv(struct btrfs_trans_handle *trans,
5655               struct btrfs_root *root, u32 blocksize)
5656 {
5657         struct btrfs_block_rsv *block_rsv;
5658         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5659         int ret;
5660
5661         block_rsv = get_block_rsv(trans, root);
5662
5663         if (block_rsv->size == 0) {
5664                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5665                                              blocksize, 0);
5666                 /*
5667                  * If we couldn't reserve metadata bytes try and use some from
5668                  * the global reserve.
5669                  */
5670                 if (ret && block_rsv != global_rsv) {
5671                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5672                         if (!ret)
5673                                 return global_rsv;
5674                         return ERR_PTR(ret);
5675                 } else if (ret) {
5676                         return ERR_PTR(ret);
5677                 }
5678                 return block_rsv;
5679         }
5680
5681         ret = block_rsv_use_bytes(block_rsv, blocksize);
5682         if (!ret)
5683                 return block_rsv;
5684         if (ret) {
5685                 WARN_ON(1);
5686                 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5687                                              0);
5688                 if (!ret) {
5689                         spin_lock(&block_rsv->lock);
5690                         block_rsv->size += blocksize;
5691                         spin_unlock(&block_rsv->lock);
5692                         return block_rsv;
5693                 } else if (ret && block_rsv != global_rsv) {
5694                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5695                         if (!ret)
5696                                 return global_rsv;
5697                 }
5698         }
5699
5700         return ERR_PTR(-ENOSPC);
5701 }
5702
5703 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5704 {
5705         block_rsv_add_bytes(block_rsv, blocksize, 0);
5706         block_rsv_release_bytes(block_rsv, NULL, 0);
5707 }
5708
5709 /*
5710  * finds a free extent and does all the dirty work required for allocation
5711  * returns the key for the extent through ins, and a tree buffer for
5712  * the first block of the extent through buf.
5713  *
5714  * returns the tree buffer or NULL.
5715  */
5716 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5717                                         struct btrfs_root *root, u32 blocksize,
5718                                         u64 parent, u64 root_objectid,
5719                                         struct btrfs_disk_key *key, int level,
5720                                         u64 hint, u64 empty_size)
5721 {
5722         struct btrfs_key ins;
5723         struct btrfs_block_rsv *block_rsv;
5724         struct extent_buffer *buf;
5725         u64 flags = 0;
5726         int ret;
5727
5728
5729         block_rsv = use_block_rsv(trans, root, blocksize);
5730         if (IS_ERR(block_rsv))
5731                 return ERR_CAST(block_rsv);
5732
5733         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5734                                    empty_size, hint, (u64)-1, &ins, 0);
5735         if (ret) {
5736                 unuse_block_rsv(block_rsv, blocksize);
5737                 return ERR_PTR(ret);
5738         }
5739
5740         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5741                                     blocksize, level);
5742         BUG_ON(IS_ERR(buf));
5743
5744         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5745                 if (parent == 0)
5746                         parent = ins.objectid;
5747                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5748         } else
5749                 BUG_ON(parent > 0);
5750
5751         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5752                 struct btrfs_delayed_extent_op *extent_op;
5753                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5754                 BUG_ON(!extent_op);
5755                 if (key)
5756                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5757                 else
5758                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5759                 extent_op->flags_to_set = flags;
5760                 extent_op->update_key = 1;
5761                 extent_op->update_flags = 1;
5762                 extent_op->is_data = 0;
5763
5764                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5765                                         ins.offset, parent, root_objectid,
5766                                         level, BTRFS_ADD_DELAYED_EXTENT,
5767                                         extent_op);
5768                 BUG_ON(ret);
5769         }
5770         return buf;
5771 }
5772
5773 struct walk_control {
5774         u64 refs[BTRFS_MAX_LEVEL];
5775         u64 flags[BTRFS_MAX_LEVEL];
5776         struct btrfs_key update_progress;
5777         int stage;
5778         int level;
5779         int shared_level;
5780         int update_ref;
5781         int keep_locks;
5782         int reada_slot;
5783         int reada_count;
5784 };
5785
5786 #define DROP_REFERENCE  1
5787 #define UPDATE_BACKREF  2
5788
5789 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5790                                      struct btrfs_root *root,
5791                                      struct walk_control *wc,
5792                                      struct btrfs_path *path)
5793 {
5794         u64 bytenr;
5795         u64 generation;
5796         u64 refs;
5797         u64 flags;
5798         u32 nritems;
5799         u32 blocksize;
5800         struct btrfs_key key;
5801         struct extent_buffer *eb;
5802         int ret;
5803         int slot;
5804         int nread = 0;
5805
5806         if (path->slots[wc->level] < wc->reada_slot) {
5807                 wc->reada_count = wc->reada_count * 2 / 3;
5808                 wc->reada_count = max(wc->reada_count, 2);
5809         } else {
5810                 wc->reada_count = wc->reada_count * 3 / 2;
5811                 wc->reada_count = min_t(int, wc->reada_count,
5812                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5813         }
5814
5815         eb = path->nodes[wc->level];
5816         nritems = btrfs_header_nritems(eb);
5817         blocksize = btrfs_level_size(root, wc->level - 1);
5818
5819         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5820                 if (nread >= wc->reada_count)
5821                         break;
5822
5823                 cond_resched();
5824                 bytenr = btrfs_node_blockptr(eb, slot);
5825                 generation = btrfs_node_ptr_generation(eb, slot);
5826
5827                 if (slot == path->slots[wc->level])
5828                         goto reada;
5829
5830                 if (wc->stage == UPDATE_BACKREF &&
5831                     generation <= root->root_key.offset)
5832                         continue;
5833
5834                 /* We don't lock the tree block, it's OK to be racy here */
5835                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5836                                                &refs, &flags);
5837                 BUG_ON(ret);
5838                 BUG_ON(refs == 0);
5839
5840                 if (wc->stage == DROP_REFERENCE) {
5841                         if (refs == 1)
5842                                 goto reada;
5843
5844                         if (wc->level == 1 &&
5845                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5846                                 continue;
5847                         if (!wc->update_ref ||
5848                             generation <= root->root_key.offset)
5849                                 continue;
5850                         btrfs_node_key_to_cpu(eb, &key, slot);
5851                         ret = btrfs_comp_cpu_keys(&key,
5852                                                   &wc->update_progress);
5853                         if (ret < 0)
5854                                 continue;
5855                 } else {
5856                         if (wc->level == 1 &&
5857                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5858                                 continue;
5859                 }
5860 reada:
5861                 ret = readahead_tree_block(root, bytenr, blocksize,
5862                                            generation);
5863                 if (ret)
5864                         break;
5865                 nread++;
5866         }
5867         wc->reada_slot = slot;
5868 }
5869
5870 /*
5871  * hepler to process tree block while walking down the tree.
5872  *
5873  * when wc->stage == UPDATE_BACKREF, this function updates
5874  * back refs for pointers in the block.
5875  *
5876  * NOTE: return value 1 means we should stop walking down.
5877  */
5878 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5879                                    struct btrfs_root *root,
5880                                    struct btrfs_path *path,
5881                                    struct walk_control *wc, int lookup_info)
5882 {
5883         int level = wc->level;
5884         struct extent_buffer *eb = path->nodes[level];
5885         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5886         int ret;
5887
5888         if (wc->stage == UPDATE_BACKREF &&
5889             btrfs_header_owner(eb) != root->root_key.objectid)
5890                 return 1;
5891
5892         /*
5893          * when reference count of tree block is 1, it won't increase
5894          * again. once full backref flag is set, we never clear it.
5895          */
5896         if (lookup_info &&
5897             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5898              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5899                 BUG_ON(!path->locks[level]);
5900                 ret = btrfs_lookup_extent_info(trans, root,
5901                                                eb->start, eb->len,
5902                                                &wc->refs[level],
5903                                                &wc->flags[level]);
5904                 BUG_ON(ret);
5905                 BUG_ON(wc->refs[level] == 0);
5906         }
5907
5908         if (wc->stage == DROP_REFERENCE) {
5909                 if (wc->refs[level] > 1)
5910                         return 1;
5911
5912                 if (path->locks[level] && !wc->keep_locks) {
5913                         btrfs_tree_unlock(eb);
5914                         path->locks[level] = 0;
5915                 }
5916                 return 0;
5917         }
5918
5919         /* wc->stage == UPDATE_BACKREF */
5920         if (!(wc->flags[level] & flag)) {
5921                 BUG_ON(!path->locks[level]);
5922                 ret = btrfs_inc_ref(trans, root, eb, 1);
5923                 BUG_ON(ret);
5924                 ret = btrfs_dec_ref(trans, root, eb, 0);
5925                 BUG_ON(ret);
5926                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5927                                                   eb->len, flag, 0);
5928                 BUG_ON(ret);
5929                 wc->flags[level] |= flag;
5930         }
5931
5932         /*
5933          * the block is shared by multiple trees, so it's not good to
5934          * keep the tree lock
5935          */
5936         if (path->locks[level] && level > 0) {
5937                 btrfs_tree_unlock(eb);
5938                 path->locks[level] = 0;
5939         }
5940         return 0;
5941 }
5942
5943 /*
5944  * hepler to process tree block pointer.
5945  *
5946  * when wc->stage == DROP_REFERENCE, this function checks
5947  * reference count of the block pointed to. if the block
5948  * is shared and we need update back refs for the subtree
5949  * rooted at the block, this function changes wc->stage to
5950  * UPDATE_BACKREF. if the block is shared and there is no
5951  * need to update back, this function drops the reference
5952  * to the block.
5953  *
5954  * NOTE: return value 1 means we should stop walking down.
5955  */
5956 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5957                                  struct btrfs_root *root,
5958                                  struct btrfs_path *path,
5959                                  struct walk_control *wc, int *lookup_info)
5960 {
5961         u64 bytenr;
5962         u64 generation;
5963         u64 parent;
5964         u32 blocksize;
5965         struct btrfs_key key;
5966         struct extent_buffer *next;
5967         int level = wc->level;
5968         int reada = 0;
5969         int ret = 0;
5970
5971         generation = btrfs_node_ptr_generation(path->nodes[level],
5972                                                path->slots[level]);
5973         /*
5974          * if the lower level block was created before the snapshot
5975          * was created, we know there is no need to update back refs
5976          * for the subtree
5977          */
5978         if (wc->stage == UPDATE_BACKREF &&
5979             generation <= root->root_key.offset) {
5980                 *lookup_info = 1;
5981                 return 1;
5982         }
5983
5984         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5985         blocksize = btrfs_level_size(root, level - 1);
5986
5987         next = btrfs_find_tree_block(root, bytenr, blocksize);
5988         if (!next) {
5989                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5990                 if (!next)
5991                         return -ENOMEM;
5992                 reada = 1;
5993         }
5994         btrfs_tree_lock(next);
5995         btrfs_set_lock_blocking(next);
5996
5997         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5998                                        &wc->refs[level - 1],
5999                                        &wc->flags[level - 1]);
6000         BUG_ON(ret);
6001         BUG_ON(wc->refs[level - 1] == 0);
6002         *lookup_info = 0;
6003
6004         if (wc->stage == DROP_REFERENCE) {
6005                 if (wc->refs[level - 1] > 1) {
6006                         if (level == 1 &&
6007                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6008                                 goto skip;
6009
6010                         if (!wc->update_ref ||
6011                             generation <= root->root_key.offset)
6012                                 goto skip;
6013
6014                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6015                                               path->slots[level]);
6016                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6017                         if (ret < 0)
6018                                 goto skip;
6019
6020                         wc->stage = UPDATE_BACKREF;
6021                         wc->shared_level = level - 1;
6022                 }
6023         } else {
6024                 if (level == 1 &&
6025                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6026                         goto skip;
6027         }
6028
6029         if (!btrfs_buffer_uptodate(next, generation)) {
6030                 btrfs_tree_unlock(next);
6031                 free_extent_buffer(next);
6032                 next = NULL;
6033                 *lookup_info = 1;
6034         }
6035
6036         if (!next) {
6037                 if (reada && level == 1)
6038                         reada_walk_down(trans, root, wc, path);
6039                 next = read_tree_block(root, bytenr, blocksize, generation);
6040                 btrfs_tree_lock(next);
6041                 btrfs_set_lock_blocking(next);
6042         }
6043
6044         level--;
6045         BUG_ON(level != btrfs_header_level(next));
6046         path->nodes[level] = next;
6047         path->slots[level] = 0;
6048         path->locks[level] = 1;
6049         wc->level = level;
6050         if (wc->level == 1)
6051                 wc->reada_slot = 0;
6052         return 0;
6053 skip:
6054         wc->refs[level - 1] = 0;
6055         wc->flags[level - 1] = 0;
6056         if (wc->stage == DROP_REFERENCE) {
6057                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6058                         parent = path->nodes[level]->start;
6059                 } else {
6060                         BUG_ON(root->root_key.objectid !=
6061                                btrfs_header_owner(path->nodes[level]));
6062                         parent = 0;
6063                 }
6064
6065                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6066                                         root->root_key.objectid, level - 1, 0);
6067                 BUG_ON(ret);
6068         }
6069         btrfs_tree_unlock(next);
6070         free_extent_buffer(next);
6071         *lookup_info = 1;
6072         return 1;
6073 }
6074
6075 /*
6076  * hepler to process tree block while walking up the tree.
6077  *
6078  * when wc->stage == DROP_REFERENCE, this function drops
6079  * reference count on the block.
6080  *
6081  * when wc->stage == UPDATE_BACKREF, this function changes
6082  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6083  * to UPDATE_BACKREF previously while processing the block.
6084  *
6085  * NOTE: return value 1 means we should stop walking up.
6086  */
6087 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6088                                  struct btrfs_root *root,
6089                                  struct btrfs_path *path,
6090                                  struct walk_control *wc)
6091 {
6092         int ret;
6093         int level = wc->level;
6094         struct extent_buffer *eb = path->nodes[level];
6095         u64 parent = 0;
6096
6097         if (wc->stage == UPDATE_BACKREF) {
6098                 BUG_ON(wc->shared_level < level);
6099                 if (level < wc->shared_level)
6100                         goto out;
6101
6102                 ret = find_next_key(path, level + 1, &wc->update_progress);
6103                 if (ret > 0)
6104                         wc->update_ref = 0;
6105
6106                 wc->stage = DROP_REFERENCE;
6107                 wc->shared_level = -1;
6108                 path->slots[level] = 0;
6109
6110                 /*
6111                  * check reference count again if the block isn't locked.
6112                  * we should start walking down the tree again if reference
6113                  * count is one.
6114                  */
6115                 if (!path->locks[level]) {
6116                         BUG_ON(level == 0);
6117                         btrfs_tree_lock(eb);
6118                         btrfs_set_lock_blocking(eb);
6119                         path->locks[level] = 1;
6120
6121                         ret = btrfs_lookup_extent_info(trans, root,
6122                                                        eb->start, eb->len,
6123                                                        &wc->refs[level],
6124                                                        &wc->flags[level]);
6125                         BUG_ON(ret);
6126                         BUG_ON(wc->refs[level] == 0);
6127                         if (wc->refs[level] == 1) {
6128                                 btrfs_tree_unlock(eb);
6129                                 path->locks[level] = 0;
6130                                 return 1;
6131                         }
6132                 }
6133         }
6134
6135         /* wc->stage == DROP_REFERENCE */
6136         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6137
6138         if (wc->refs[level] == 1) {
6139                 if (level == 0) {
6140                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6141                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6142                         else
6143                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6144                         BUG_ON(ret);
6145                 }
6146                 /* make block locked assertion in clean_tree_block happy */
6147                 if (!path->locks[level] &&
6148                     btrfs_header_generation(eb) == trans->transid) {
6149                         btrfs_tree_lock(eb);
6150                         btrfs_set_lock_blocking(eb);
6151                         path->locks[level] = 1;
6152                 }
6153                 clean_tree_block(trans, root, eb);
6154         }
6155
6156         if (eb == root->node) {
6157                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6158                         parent = eb->start;
6159                 else
6160                         BUG_ON(root->root_key.objectid !=
6161                                btrfs_header_owner(eb));
6162         } else {
6163                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6164                         parent = path->nodes[level + 1]->start;
6165                 else
6166                         BUG_ON(root->root_key.objectid !=
6167                                btrfs_header_owner(path->nodes[level + 1]));
6168         }
6169
6170         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6171 out:
6172         wc->refs[level] = 0;
6173         wc->flags[level] = 0;
6174         return 0;
6175 }
6176
6177 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6178                                    struct btrfs_root *root,
6179                                    struct btrfs_path *path,
6180                                    struct walk_control *wc)
6181 {
6182         int level = wc->level;
6183         int lookup_info = 1;
6184         int ret;
6185
6186         while (level >= 0) {
6187                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6188                 if (ret > 0)
6189                         break;
6190
6191                 if (level == 0)
6192                         break;
6193
6194                 if (path->slots[level] >=
6195                     btrfs_header_nritems(path->nodes[level]))
6196                         break;
6197
6198                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6199                 if (ret > 0) {
6200                         path->slots[level]++;
6201                         continue;
6202                 } else if (ret < 0)
6203                         return ret;
6204                 level = wc->level;
6205         }
6206         return 0;
6207 }
6208
6209 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6210                                  struct btrfs_root *root,
6211                                  struct btrfs_path *path,
6212                                  struct walk_control *wc, int max_level)
6213 {
6214         int level = wc->level;
6215         int ret;
6216
6217         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6218         while (level < max_level && path->nodes[level]) {
6219                 wc->level = level;
6220                 if (path->slots[level] + 1 <
6221                     btrfs_header_nritems(path->nodes[level])) {
6222                         path->slots[level]++;
6223                         return 0;
6224                 } else {
6225                         ret = walk_up_proc(trans, root, path, wc);
6226                         if (ret > 0)
6227                                 return 0;
6228
6229                         if (path->locks[level]) {
6230                                 btrfs_tree_unlock(path->nodes[level]);
6231                                 path->locks[level] = 0;
6232                         }
6233                         free_extent_buffer(path->nodes[level]);
6234                         path->nodes[level] = NULL;
6235                         level++;
6236                 }
6237         }
6238         return 1;
6239 }
6240
6241 /*
6242  * drop a subvolume tree.
6243  *
6244  * this function traverses the tree freeing any blocks that only
6245  * referenced by the tree.
6246  *
6247  * when a shared tree block is found. this function decreases its
6248  * reference count by one. if update_ref is true, this function
6249  * also make sure backrefs for the shared block and all lower level
6250  * blocks are properly updated.
6251  */
6252 int btrfs_drop_snapshot(struct btrfs_root *root,
6253                         struct btrfs_block_rsv *block_rsv, int update_ref)
6254 {
6255         struct btrfs_path *path;
6256         struct btrfs_trans_handle *trans;
6257         struct btrfs_root *tree_root = root->fs_info->tree_root;
6258         struct btrfs_root_item *root_item = &root->root_item;
6259         struct walk_control *wc;
6260         struct btrfs_key key;
6261         int err = 0;
6262         int ret;
6263         int level;
6264
6265         path = btrfs_alloc_path();
6266         BUG_ON(!path);
6267
6268         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6269         BUG_ON(!wc);
6270
6271         trans = btrfs_start_transaction(tree_root, 0);
6272         BUG_ON(IS_ERR(trans));
6273
6274         if (block_rsv)
6275                 trans->block_rsv = block_rsv;
6276
6277         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6278                 level = btrfs_header_level(root->node);
6279                 path->nodes[level] = btrfs_lock_root_node(root);
6280                 btrfs_set_lock_blocking(path->nodes[level]);
6281                 path->slots[level] = 0;
6282                 path->locks[level] = 1;
6283                 memset(&wc->update_progress, 0,
6284                        sizeof(wc->update_progress));
6285         } else {
6286                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6287                 memcpy(&wc->update_progress, &key,
6288                        sizeof(wc->update_progress));
6289
6290                 level = root_item->drop_level;
6291                 BUG_ON(level == 0);
6292                 path->lowest_level = level;
6293                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6294                 path->lowest_level = 0;
6295                 if (ret < 0) {
6296                         err = ret;
6297                         goto out;
6298                 }
6299                 WARN_ON(ret > 0);
6300
6301                 /*
6302                  * unlock our path, this is safe because only this
6303                  * function is allowed to delete this snapshot
6304                  */
6305                 btrfs_unlock_up_safe(path, 0);
6306
6307                 level = btrfs_header_level(root->node);
6308                 while (1) {
6309                         btrfs_tree_lock(path->nodes[level]);
6310                         btrfs_set_lock_blocking(path->nodes[level]);
6311
6312                         ret = btrfs_lookup_extent_info(trans, root,
6313                                                 path->nodes[level]->start,
6314                                                 path->nodes[level]->len,
6315                                                 &wc->refs[level],
6316                                                 &wc->flags[level]);
6317                         BUG_ON(ret);
6318                         BUG_ON(wc->refs[level] == 0);
6319
6320                         if (level == root_item->drop_level)
6321                                 break;
6322
6323                         btrfs_tree_unlock(path->nodes[level]);
6324                         WARN_ON(wc->refs[level] != 1);
6325                         level--;
6326                 }
6327         }
6328
6329         wc->level = level;
6330         wc->shared_level = -1;
6331         wc->stage = DROP_REFERENCE;
6332         wc->update_ref = update_ref;
6333         wc->keep_locks = 0;
6334         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6335
6336         while (1) {
6337                 ret = walk_down_tree(trans, root, path, wc);
6338                 if (ret < 0) {
6339                         err = ret;
6340                         break;
6341                 }
6342
6343                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6344                 if (ret < 0) {
6345                         err = ret;
6346                         break;
6347                 }
6348
6349                 if (ret > 0) {
6350                         BUG_ON(wc->stage != DROP_REFERENCE);
6351                         break;
6352                 }
6353
6354                 if (wc->stage == DROP_REFERENCE) {
6355                         level = wc->level;
6356                         btrfs_node_key(path->nodes[level],
6357                                        &root_item->drop_progress,
6358                                        path->slots[level]);
6359                         root_item->drop_level = level;
6360                 }
6361
6362                 BUG_ON(wc->level == 0);
6363                 if (btrfs_should_end_transaction(trans, tree_root)) {
6364                         ret = btrfs_update_root(trans, tree_root,
6365                                                 &root->root_key,
6366                                                 root_item);
6367                         BUG_ON(ret);
6368
6369                         btrfs_end_transaction_throttle(trans, tree_root);
6370                         trans = btrfs_start_transaction(tree_root, 0);
6371                         BUG_ON(IS_ERR(trans));
6372                         if (block_rsv)
6373                                 trans->block_rsv = block_rsv;
6374                 }
6375         }
6376         btrfs_release_path(root, path);
6377         BUG_ON(err);
6378
6379         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6380         BUG_ON(ret);
6381
6382         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6383                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6384                                            NULL, NULL);
6385                 BUG_ON(ret < 0);
6386                 if (ret > 0) {
6387                         /* if we fail to delete the orphan item this time
6388                          * around, it'll get picked up the next time.
6389                          *
6390                          * The most common failure here is just -ENOENT.
6391                          */
6392                         btrfs_del_orphan_item(trans, tree_root,
6393                                               root->root_key.objectid);
6394                 }
6395         }
6396
6397         if (root->in_radix) {
6398                 btrfs_free_fs_root(tree_root->fs_info, root);
6399         } else {
6400                 free_extent_buffer(root->node);
6401                 free_extent_buffer(root->commit_root);
6402                 kfree(root);
6403         }
6404 out:
6405         btrfs_end_transaction_throttle(trans, tree_root);
6406         kfree(wc);
6407         btrfs_free_path(path);
6408         return err;
6409 }
6410
6411 /*
6412  * drop subtree rooted at tree block 'node'.
6413  *
6414  * NOTE: this function will unlock and release tree block 'node'
6415  */
6416 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6417                         struct btrfs_root *root,
6418                         struct extent_buffer *node,
6419                         struct extent_buffer *parent)
6420 {
6421         struct btrfs_path *path;
6422         struct walk_control *wc;
6423         int level;
6424         int parent_level;
6425         int ret = 0;
6426         int wret;
6427
6428         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6429
6430         path = btrfs_alloc_path();
6431         BUG_ON(!path);
6432
6433         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6434         BUG_ON(!wc);
6435
6436         btrfs_assert_tree_locked(parent);
6437         parent_level = btrfs_header_level(parent);
6438         extent_buffer_get(parent);
6439         path->nodes[parent_level] = parent;
6440         path->slots[parent_level] = btrfs_header_nritems(parent);
6441
6442         btrfs_assert_tree_locked(node);
6443         level = btrfs_header_level(node);
6444         path->nodes[level] = node;
6445         path->slots[level] = 0;
6446         path->locks[level] = 1;
6447
6448         wc->refs[parent_level] = 1;
6449         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6450         wc->level = level;
6451         wc->shared_level = -1;
6452         wc->stage = DROP_REFERENCE;
6453         wc->update_ref = 0;
6454         wc->keep_locks = 1;
6455         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6456
6457         while (1) {
6458                 wret = walk_down_tree(trans, root, path, wc);
6459                 if (wret < 0) {
6460                         ret = wret;
6461                         break;
6462                 }
6463
6464                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6465                 if (wret < 0)
6466                         ret = wret;
6467                 if (wret != 0)
6468                         break;
6469         }
6470
6471         kfree(wc);
6472         btrfs_free_path(path);
6473         return ret;
6474 }
6475
6476 #if 0
6477 static unsigned long calc_ra(unsigned long start, unsigned long last,
6478                              unsigned long nr)
6479 {
6480         return min(last, start + nr - 1);
6481 }
6482
6483 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6484                                          u64 len)
6485 {
6486         u64 page_start;
6487         u64 page_end;
6488         unsigned long first_index;
6489         unsigned long last_index;
6490         unsigned long i;
6491         struct page *page;
6492         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6493         struct file_ra_state *ra;
6494         struct btrfs_ordered_extent *ordered;
6495         unsigned int total_read = 0;
6496         unsigned int total_dirty = 0;
6497         int ret = 0;
6498
6499         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6500         if (!ra)
6501                 return -ENOMEM;
6502
6503         mutex_lock(&inode->i_mutex);
6504         first_index = start >> PAGE_CACHE_SHIFT;
6505         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6506
6507         /* make sure the dirty trick played by the caller work */
6508         ret = invalidate_inode_pages2_range(inode->i_mapping,
6509                                             first_index, last_index);
6510         if (ret)
6511                 goto out_unlock;
6512
6513         file_ra_state_init(ra, inode->i_mapping);
6514
6515         for (i = first_index ; i <= last_index; i++) {
6516                 if (total_read % ra->ra_pages == 0) {
6517                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6518                                        calc_ra(i, last_index, ra->ra_pages));
6519                 }
6520                 total_read++;
6521 again:
6522                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6523                         BUG_ON(1);
6524                 page = grab_cache_page(inode->i_mapping, i);
6525                 if (!page) {
6526                         ret = -ENOMEM;
6527                         goto out_unlock;
6528                 }
6529                 if (!PageUptodate(page)) {
6530                         btrfs_readpage(NULL, page);
6531                         lock_page(page);
6532                         if (!PageUptodate(page)) {
6533                                 unlock_page(page);
6534                                 page_cache_release(page);
6535                                 ret = -EIO;
6536                                 goto out_unlock;
6537                         }
6538                 }
6539                 wait_on_page_writeback(page);
6540
6541                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6542                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6543                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6544
6545                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6546                 if (ordered) {
6547                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6548                         unlock_page(page);
6549                         page_cache_release(page);
6550                         btrfs_start_ordered_extent(inode, ordered, 1);
6551                         btrfs_put_ordered_extent(ordered);
6552                         goto again;
6553                 }
6554                 set_page_extent_mapped(page);
6555
6556                 if (i == first_index)
6557                         set_extent_bits(io_tree, page_start, page_end,
6558                                         EXTENT_BOUNDARY, GFP_NOFS);
6559                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6560
6561                 set_page_dirty(page);
6562                 total_dirty++;
6563
6564                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6565                 unlock_page(page);
6566                 page_cache_release(page);
6567         }
6568
6569 out_unlock:
6570         kfree(ra);
6571         mutex_unlock(&inode->i_mutex);
6572         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6573         return ret;
6574 }
6575
6576 static noinline int relocate_data_extent(struct inode *reloc_inode,
6577                                          struct btrfs_key *extent_key,
6578                                          u64 offset)
6579 {
6580         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6581         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6582         struct extent_map *em;
6583         u64 start = extent_key->objectid - offset;
6584         u64 end = start + extent_key->offset - 1;
6585
6586         em = alloc_extent_map(GFP_NOFS);
6587         BUG_ON(!em);
6588
6589         em->start = start;
6590         em->len = extent_key->offset;
6591         em->block_len = extent_key->offset;
6592         em->block_start = extent_key->objectid;
6593         em->bdev = root->fs_info->fs_devices->latest_bdev;
6594         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6595
6596         /* setup extent map to cheat btrfs_readpage */
6597         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6598         while (1) {
6599                 int ret;
6600                 write_lock(&em_tree->lock);
6601                 ret = add_extent_mapping(em_tree, em);
6602                 write_unlock(&em_tree->lock);
6603                 if (ret != -EEXIST) {
6604                         free_extent_map(em);
6605                         break;
6606                 }
6607                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6608         }
6609         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6610
6611         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6612 }
6613
6614 struct btrfs_ref_path {
6615         u64 extent_start;
6616         u64 nodes[BTRFS_MAX_LEVEL];
6617         u64 root_objectid;
6618         u64 root_generation;
6619         u64 owner_objectid;
6620         u32 num_refs;
6621         int lowest_level;
6622         int current_level;
6623         int shared_level;
6624
6625         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6626         u64 new_nodes[BTRFS_MAX_LEVEL];
6627 };
6628
6629 struct disk_extent {
6630         u64 ram_bytes;
6631         u64 disk_bytenr;
6632         u64 disk_num_bytes;
6633         u64 offset;
6634         u64 num_bytes;
6635         u8 compression;
6636         u8 encryption;
6637         u16 other_encoding;
6638 };
6639
6640 static int is_cowonly_root(u64 root_objectid)
6641 {
6642         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6643             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6644             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6645             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6646             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6647             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6648                 return 1;
6649         return 0;
6650 }
6651
6652 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6653                                     struct btrfs_root *extent_root,
6654                                     struct btrfs_ref_path *ref_path,
6655                                     int first_time)
6656 {
6657         struct extent_buffer *leaf;
6658         struct btrfs_path *path;
6659         struct btrfs_extent_ref *ref;
6660         struct btrfs_key key;
6661         struct btrfs_key found_key;
6662         u64 bytenr;
6663         u32 nritems;
6664         int level;
6665         int ret = 1;
6666
6667         path = btrfs_alloc_path();
6668         if (!path)
6669                 return -ENOMEM;
6670
6671         if (first_time) {
6672                 ref_path->lowest_level = -1;
6673                 ref_path->current_level = -1;
6674                 ref_path->shared_level = -1;
6675                 goto walk_up;
6676         }
6677 walk_down:
6678         level = ref_path->current_level - 1;
6679         while (level >= -1) {
6680                 u64 parent;
6681                 if (level < ref_path->lowest_level)
6682                         break;
6683
6684                 if (level >= 0)
6685                         bytenr = ref_path->nodes[level];
6686                 else
6687                         bytenr = ref_path->extent_start;
6688                 BUG_ON(bytenr == 0);
6689
6690                 parent = ref_path->nodes[level + 1];
6691                 ref_path->nodes[level + 1] = 0;
6692                 ref_path->current_level = level;
6693                 BUG_ON(parent == 0);
6694
6695                 key.objectid = bytenr;
6696                 key.offset = parent + 1;
6697                 key.type = BTRFS_EXTENT_REF_KEY;
6698
6699                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6700                 if (ret < 0)
6701                         goto out;
6702                 BUG_ON(ret == 0);
6703
6704                 leaf = path->nodes[0];
6705                 nritems = btrfs_header_nritems(leaf);
6706                 if (path->slots[0] >= nritems) {
6707                         ret = btrfs_next_leaf(extent_root, path);
6708                         if (ret < 0)
6709                                 goto out;
6710                         if (ret > 0)
6711                                 goto next;
6712                         leaf = path->nodes[0];
6713                 }
6714
6715                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6716                 if (found_key.objectid == bytenr &&
6717                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6718                         if (level < ref_path->shared_level)
6719                                 ref_path->shared_level = level;
6720                         goto found;
6721                 }
6722 next:
6723                 level--;
6724                 btrfs_release_path(extent_root, path);
6725                 cond_resched();
6726         }
6727         /* reached lowest level */
6728         ret = 1;
6729         goto out;
6730 walk_up:
6731         level = ref_path->current_level;
6732         while (level < BTRFS_MAX_LEVEL - 1) {
6733                 u64 ref_objectid;
6734
6735                 if (level >= 0)
6736                         bytenr = ref_path->nodes[level];
6737                 else
6738                         bytenr = ref_path->extent_start;
6739
6740                 BUG_ON(bytenr == 0);
6741
6742                 key.objectid = bytenr;
6743                 key.offset = 0;
6744                 key.type = BTRFS_EXTENT_REF_KEY;
6745
6746                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6747                 if (ret < 0)
6748                         goto out;
6749
6750                 leaf = path->nodes[0];
6751                 nritems = btrfs_header_nritems(leaf);
6752                 if (path->slots[0] >= nritems) {
6753                         ret = btrfs_next_leaf(extent_root, path);
6754                         if (ret < 0)
6755                                 goto out;
6756                         if (ret > 0) {
6757                                 /* the extent was freed by someone */
6758                                 if (ref_path->lowest_level == level)
6759                                         goto out;
6760                                 btrfs_release_path(extent_root, path);
6761                                 goto walk_down;
6762                         }
6763                         leaf = path->nodes[0];
6764                 }
6765
6766                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6767                 if (found_key.objectid != bytenr ||
6768                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6769                         /* the extent was freed by someone */
6770                         if (ref_path->lowest_level == level) {
6771                                 ret = 1;
6772                                 goto out;
6773                         }
6774                         btrfs_release_path(extent_root, path);
6775                         goto walk_down;
6776                 }
6777 found:
6778                 ref = btrfs_item_ptr(leaf, path->slots[0],
6779                                 struct btrfs_extent_ref);
6780                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6781                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6782                         if (first_time) {
6783                                 level = (int)ref_objectid;
6784                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6785                                 ref_path->lowest_level = level;
6786                                 ref_path->current_level = level;
6787                                 ref_path->nodes[level] = bytenr;
6788                         } else {
6789                                 WARN_ON(ref_objectid != level);
6790                         }
6791                 } else {
6792                         WARN_ON(level != -1);
6793                 }
6794                 first_time = 0;
6795
6796                 if (ref_path->lowest_level == level) {
6797                         ref_path->owner_objectid = ref_objectid;
6798                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6799                 }
6800
6801                 /*
6802                  * the block is tree root or the block isn't in reference
6803                  * counted tree.
6804                  */
6805                 if (found_key.objectid == found_key.offset ||
6806                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6807                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6808                         ref_path->root_generation =
6809                                 btrfs_ref_generation(leaf, ref);
6810                         if (level < 0) {
6811                                 /* special reference from the tree log */
6812                                 ref_path->nodes[0] = found_key.offset;
6813                                 ref_path->current_level = 0;
6814                         }
6815                         ret = 0;
6816                         goto out;
6817                 }
6818
6819                 level++;
6820                 BUG_ON(ref_path->nodes[level] != 0);
6821                 ref_path->nodes[level] = found_key.offset;
6822                 ref_path->current_level = level;
6823
6824                 /*
6825                  * the reference was created in the running transaction,
6826                  * no need to continue walking up.
6827                  */
6828                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6829                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6830                         ref_path->root_generation =
6831                                 btrfs_ref_generation(leaf, ref);
6832                         ret = 0;
6833                         goto out;
6834                 }
6835
6836                 btrfs_release_path(extent_root, path);
6837                 cond_resched();
6838         }
6839         /* reached max tree level, but no tree root found. */
6840         BUG();
6841 out:
6842         btrfs_free_path(path);
6843         return ret;
6844 }
6845
6846 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6847                                 struct btrfs_root *extent_root,
6848                                 struct btrfs_ref_path *ref_path,
6849                                 u64 extent_start)
6850 {
6851         memset(ref_path, 0, sizeof(*ref_path));
6852         ref_path->extent_start = extent_start;
6853
6854         return __next_ref_path(trans, extent_root, ref_path, 1);
6855 }
6856
6857 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6858                                struct btrfs_root *extent_root,
6859                                struct btrfs_ref_path *ref_path)
6860 {
6861         return __next_ref_path(trans, extent_root, ref_path, 0);
6862 }
6863
6864 static noinline int get_new_locations(struct inode *reloc_inode,
6865                                       struct btrfs_key *extent_key,
6866                                       u64 offset, int no_fragment,
6867                                       struct disk_extent **extents,
6868                                       int *nr_extents)
6869 {
6870         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6871         struct btrfs_path *path;
6872         struct btrfs_file_extent_item *fi;
6873         struct extent_buffer *leaf;
6874         struct disk_extent *exts = *extents;
6875         struct btrfs_key found_key;
6876         u64 cur_pos;
6877         u64 last_byte;
6878         u32 nritems;
6879         int nr = 0;
6880         int max = *nr_extents;
6881         int ret;
6882
6883         WARN_ON(!no_fragment && *extents);
6884         if (!exts) {
6885                 max = 1;
6886                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6887                 if (!exts)
6888                         return -ENOMEM;
6889         }
6890
6891         path = btrfs_alloc_path();
6892         BUG_ON(!path);
6893
6894         cur_pos = extent_key->objectid - offset;
6895         last_byte = extent_key->objectid + extent_key->offset;
6896         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6897                                        cur_pos, 0);
6898         if (ret < 0)
6899                 goto out;
6900         if (ret > 0) {
6901                 ret = -ENOENT;
6902                 goto out;
6903         }
6904
6905         while (1) {
6906                 leaf = path->nodes[0];
6907                 nritems = btrfs_header_nritems(leaf);
6908                 if (path->slots[0] >= nritems) {
6909                         ret = btrfs_next_leaf(root, path);
6910                         if (ret < 0)
6911                                 goto out;
6912                         if (ret > 0)
6913                                 break;
6914                         leaf = path->nodes[0];
6915                 }
6916
6917                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6918                 if (found_key.offset != cur_pos ||
6919                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6920                     found_key.objectid != reloc_inode->i_ino)
6921                         break;
6922
6923                 fi = btrfs_item_ptr(leaf, path->slots[0],
6924                                     struct btrfs_file_extent_item);
6925                 if (btrfs_file_extent_type(leaf, fi) !=
6926                     BTRFS_FILE_EXTENT_REG ||
6927                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6928                         break;
6929
6930                 if (nr == max) {
6931                         struct disk_extent *old = exts;
6932                         max *= 2;
6933                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6934                         memcpy(exts, old, sizeof(*exts) * nr);
6935                         if (old != *extents)
6936                                 kfree(old);
6937                 }
6938
6939                 exts[nr].disk_bytenr =
6940                         btrfs_file_extent_disk_bytenr(leaf, fi);
6941                 exts[nr].disk_num_bytes =
6942                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6943                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6944                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6945                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6946                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6947                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6948                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6949                                                                            fi);
6950                 BUG_ON(exts[nr].offset > 0);
6951                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6952                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6953
6954                 cur_pos += exts[nr].num_bytes;
6955                 nr++;
6956
6957                 if (cur_pos + offset >= last_byte)
6958                         break;
6959
6960                 if (no_fragment) {
6961                         ret = 1;
6962                         goto out;
6963                 }
6964                 path->slots[0]++;
6965         }
6966
6967         BUG_ON(cur_pos + offset > last_byte);
6968         if (cur_pos + offset < last_byte) {
6969                 ret = -ENOENT;
6970                 goto out;
6971         }
6972         ret = 0;
6973 out:
6974         btrfs_free_path(path);
6975         if (ret) {
6976                 if (exts != *extents)
6977                         kfree(exts);
6978         } else {
6979                 *extents = exts;
6980                 *nr_extents = nr;
6981         }
6982         return ret;
6983 }
6984
6985 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6986                                         struct btrfs_root *root,
6987                                         struct btrfs_path *path,
6988                                         struct btrfs_key *extent_key,
6989                                         struct btrfs_key *leaf_key,
6990                                         struct btrfs_ref_path *ref_path,
6991                                         struct disk_extent *new_extents,
6992                                         int nr_extents)
6993 {
6994         struct extent_buffer *leaf;
6995         struct btrfs_file_extent_item *fi;
6996         struct inode *inode = NULL;
6997         struct btrfs_key key;
6998         u64 lock_start = 0;
6999         u64 lock_end = 0;
7000         u64 num_bytes;
7001         u64 ext_offset;
7002         u64 search_end = (u64)-1;
7003         u32 nritems;
7004         int nr_scaned = 0;
7005         int extent_locked = 0;
7006         int extent_type;
7007         int ret;
7008
7009         memcpy(&key, leaf_key, sizeof(key));
7010         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7011                 if (key.objectid < ref_path->owner_objectid ||
7012                     (key.objectid == ref_path->owner_objectid &&
7013                      key.type < BTRFS_EXTENT_DATA_KEY)) {
7014                         key.objectid = ref_path->owner_objectid;
7015                         key.type = BTRFS_EXTENT_DATA_KEY;
7016                         key.offset = 0;
7017                 }
7018         }
7019
7020         while (1) {
7021                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7022                 if (ret < 0)
7023                         goto out;
7024
7025                 leaf = path->nodes[0];
7026                 nritems = btrfs_header_nritems(leaf);
7027 next:
7028                 if (extent_locked && ret > 0) {
7029                         /*
7030                          * the file extent item was modified by someone
7031                          * before the extent got locked.
7032                          */
7033                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7034                                       lock_end, GFP_NOFS);
7035                         extent_locked = 0;
7036                 }
7037
7038                 if (path->slots[0] >= nritems) {
7039                         if (++nr_scaned > 2)
7040                                 break;
7041
7042                         BUG_ON(extent_locked);
7043                         ret = btrfs_next_leaf(root, path);
7044                         if (ret < 0)
7045                                 goto out;
7046                         if (ret > 0)
7047                                 break;
7048                         leaf = path->nodes[0];
7049                         nritems = btrfs_header_nritems(leaf);
7050                 }
7051
7052                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7053
7054                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7055                         if ((key.objectid > ref_path->owner_objectid) ||
7056                             (key.objectid == ref_path->owner_objectid &&
7057                              key.type > BTRFS_EXTENT_DATA_KEY) ||
7058                             key.offset >= search_end)
7059                                 break;
7060                 }
7061
7062                 if (inode && key.objectid != inode->i_ino) {
7063                         BUG_ON(extent_locked);
7064                         btrfs_release_path(root, path);
7065                         mutex_unlock(&inode->i_mutex);
7066                         iput(inode);
7067                         inode = NULL;
7068                         continue;
7069                 }
7070
7071                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7072                         path->slots[0]++;
7073                         ret = 1;
7074                         goto next;
7075                 }
7076                 fi = btrfs_item_ptr(leaf, path->slots[0],
7077                                     struct btrfs_file_extent_item);
7078                 extent_type = btrfs_file_extent_type(leaf, fi);
7079                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7080                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7081                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7082                      extent_key->objectid)) {
7083                         path->slots[0]++;
7084                         ret = 1;
7085                         goto next;
7086                 }
7087
7088                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7089                 ext_offset = btrfs_file_extent_offset(leaf, fi);
7090
7091                 if (search_end == (u64)-1) {
7092                         search_end = key.offset - ext_offset +
7093                                 btrfs_file_extent_ram_bytes(leaf, fi);
7094                 }
7095
7096                 if (!extent_locked) {
7097                         lock_start = key.offset;
7098                         lock_end = lock_start + num_bytes - 1;
7099                 } else {
7100                         if (lock_start > key.offset ||
7101                             lock_end + 1 < key.offset + num_bytes) {
7102                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7103                                               lock_start, lock_end, GFP_NOFS);
7104                                 extent_locked = 0;
7105                         }
7106                 }
7107
7108                 if (!inode) {
7109                         btrfs_release_path(root, path);
7110
7111                         inode = btrfs_iget_locked(root->fs_info->sb,
7112                                                   key.objectid, root);
7113                         if (inode->i_state & I_NEW) {
7114                                 BTRFS_I(inode)->root = root;
7115                                 BTRFS_I(inode)->location.objectid =
7116                                         key.objectid;
7117                                 BTRFS_I(inode)->location.type =
7118                                         BTRFS_INODE_ITEM_KEY;
7119                                 BTRFS_I(inode)->location.offset = 0;
7120                                 btrfs_read_locked_inode(inode);
7121                                 unlock_new_inode(inode);
7122                         }
7123                         /*
7124                          * some code call btrfs_commit_transaction while
7125                          * holding the i_mutex, so we can't use mutex_lock
7126                          * here.
7127                          */
7128                         if (is_bad_inode(inode) ||
7129                             !mutex_trylock(&inode->i_mutex)) {
7130                                 iput(inode);
7131                                 inode = NULL;
7132                                 key.offset = (u64)-1;
7133                                 goto skip;
7134                         }
7135                 }
7136
7137                 if (!extent_locked) {
7138                         struct btrfs_ordered_extent *ordered;
7139
7140                         btrfs_release_path(root, path);
7141
7142                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7143                                     lock_end, GFP_NOFS);
7144                         ordered = btrfs_lookup_first_ordered_extent(inode,
7145                                                                     lock_end);
7146                         if (ordered &&
7147                             ordered->file_offset <= lock_end &&
7148                             ordered->file_offset + ordered->len > lock_start) {
7149                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7150                                               lock_start, lock_end, GFP_NOFS);
7151                                 btrfs_start_ordered_extent(inode, ordered, 1);
7152                                 btrfs_put_ordered_extent(ordered);
7153                                 key.offset += num_bytes;
7154                                 goto skip;
7155                         }
7156                         if (ordered)
7157                                 btrfs_put_ordered_extent(ordered);
7158
7159                         extent_locked = 1;
7160                         continue;
7161                 }
7162
7163                 if (nr_extents == 1) {
7164                         /* update extent pointer in place */
7165                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7166                                                 new_extents[0].disk_bytenr);
7167                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7168                                                 new_extents[0].disk_num_bytes);
7169                         btrfs_mark_buffer_dirty(leaf);
7170
7171                         btrfs_drop_extent_cache(inode, key.offset,
7172                                                 key.offset + num_bytes - 1, 0);
7173
7174                         ret = btrfs_inc_extent_ref(trans, root,
7175                                                 new_extents[0].disk_bytenr,
7176                                                 new_extents[0].disk_num_bytes,
7177                                                 leaf->start,
7178                                                 root->root_key.objectid,
7179                                                 trans->transid,
7180                                                 key.objectid);
7181                         BUG_ON(ret);
7182
7183                         ret = btrfs_free_extent(trans, root,
7184                                                 extent_key->objectid,
7185                                                 extent_key->offset,
7186                                                 leaf->start,
7187                                                 btrfs_header_owner(leaf),
7188                                                 btrfs_header_generation(leaf),
7189                                                 key.objectid, 0);
7190                         BUG_ON(ret);
7191
7192                         btrfs_release_path(root, path);
7193                         key.offset += num_bytes;
7194                 } else {
7195                         BUG_ON(1);
7196 #if 0
7197                         u64 alloc_hint;
7198                         u64 extent_len;
7199                         int i;
7200                         /*
7201                          * drop old extent pointer at first, then insert the
7202                          * new pointers one bye one
7203                          */
7204                         btrfs_release_path(root, path);
7205                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7206                                                  key.offset + num_bytes,
7207                                                  key.offset, &alloc_hint);
7208                         BUG_ON(ret);
7209
7210                         for (i = 0; i < nr_extents; i++) {
7211                                 if (ext_offset >= new_extents[i].num_bytes) {
7212                                         ext_offset -= new_extents[i].num_bytes;
7213                                         continue;
7214                                 }
7215                                 extent_len = min(new_extents[i].num_bytes -
7216                                                  ext_offset, num_bytes);
7217
7218                                 ret = btrfs_insert_empty_item(trans, root,
7219                                                               path, &key,
7220                                                               sizeof(*fi));
7221                                 BUG_ON(ret);
7222
7223                                 leaf = path->nodes[0];
7224                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7225                                                 struct btrfs_file_extent_item);
7226                                 btrfs_set_file_extent_generation(leaf, fi,
7227                                                         trans->transid);
7228                                 btrfs_set_file_extent_type(leaf, fi,
7229                                                         BTRFS_FILE_EXTENT_REG);
7230                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7231                                                 new_extents[i].disk_bytenr);
7232                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7233                                                 new_extents[i].disk_num_bytes);
7234                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7235                                                 new_extents[i].ram_bytes);
7236
7237                                 btrfs_set_file_extent_compression(leaf, fi,
7238                                                 new_extents[i].compression);
7239                                 btrfs_set_file_extent_encryption(leaf, fi,
7240                                                 new_extents[i].encryption);
7241                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7242                                                 new_extents[i].other_encoding);
7243
7244                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7245                                                         extent_len);
7246                                 ext_offset += new_extents[i].offset;
7247                                 btrfs_set_file_extent_offset(leaf, fi,
7248                                                         ext_offset);
7249                                 btrfs_mark_buffer_dirty(leaf);
7250
7251                                 btrfs_drop_extent_cache(inode, key.offset,
7252                                                 key.offset + extent_len - 1, 0);
7253
7254                                 ret = btrfs_inc_extent_ref(trans, root,
7255                                                 new_extents[i].disk_bytenr,
7256                                                 new_extents[i].disk_num_bytes,
7257                                                 leaf->start,
7258                                                 root->root_key.objectid,
7259                                                 trans->transid, key.objectid);
7260                                 BUG_ON(ret);
7261                                 btrfs_release_path(root, path);
7262
7263                                 inode_add_bytes(inode, extent_len);
7264
7265                                 ext_offset = 0;
7266                                 num_bytes -= extent_len;
7267                                 key.offset += extent_len;
7268
7269                                 if (num_bytes == 0)
7270                                         break;
7271                         }
7272                         BUG_ON(i >= nr_extents);
7273 #endif
7274                 }
7275
7276                 if (extent_locked) {
7277                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7278                                       lock_end, GFP_NOFS);
7279                         extent_locked = 0;
7280                 }
7281 skip:
7282                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7283                     key.offset >= search_end)
7284                         break;
7285
7286                 cond_resched();
7287         }
7288         ret = 0;
7289 out:
7290         btrfs_release_path(root, path);
7291         if (inode) {
7292                 mutex_unlock(&inode->i_mutex);
7293                 if (extent_locked) {
7294                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7295                                       lock_end, GFP_NOFS);
7296                 }
7297                 iput(inode);
7298         }
7299         return ret;
7300 }
7301
7302 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7303                                struct btrfs_root *root,
7304                                struct extent_buffer *buf, u64 orig_start)
7305 {
7306         int level;
7307         int ret;
7308
7309         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7310         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7311
7312         level = btrfs_header_level(buf);
7313         if (level == 0) {
7314                 struct btrfs_leaf_ref *ref;
7315                 struct btrfs_leaf_ref *orig_ref;
7316
7317                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7318                 if (!orig_ref)
7319                         return -ENOENT;
7320
7321                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7322                 if (!ref) {
7323                         btrfs_free_leaf_ref(root, orig_ref);
7324                         return -ENOMEM;
7325                 }
7326
7327                 ref->nritems = orig_ref->nritems;
7328                 memcpy(ref->extents, orig_ref->extents,
7329                         sizeof(ref->extents[0]) * ref->nritems);
7330
7331                 btrfs_free_leaf_ref(root, orig_ref);
7332
7333                 ref->root_gen = trans->transid;
7334                 ref->bytenr = buf->start;
7335                 ref->owner = btrfs_header_owner(buf);
7336                 ref->generation = btrfs_header_generation(buf);
7337
7338                 ret = btrfs_add_leaf_ref(root, ref, 0);
7339                 WARN_ON(ret);
7340                 btrfs_free_leaf_ref(root, ref);
7341         }
7342         return 0;
7343 }
7344
7345 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7346                                         struct extent_buffer *leaf,
7347                                         struct btrfs_block_group_cache *group,
7348                                         struct btrfs_root *target_root)
7349 {
7350         struct btrfs_key key;
7351         struct inode *inode = NULL;
7352         struct btrfs_file_extent_item *fi;
7353         struct extent_state *cached_state = NULL;
7354         u64 num_bytes;
7355         u64 skip_objectid = 0;
7356         u32 nritems;
7357         u32 i;
7358
7359         nritems = btrfs_header_nritems(leaf);
7360         for (i = 0; i < nritems; i++) {
7361                 btrfs_item_key_to_cpu(leaf, &key, i);
7362                 if (key.objectid == skip_objectid ||
7363                     key.type != BTRFS_EXTENT_DATA_KEY)
7364                         continue;
7365                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7366                 if (btrfs_file_extent_type(leaf, fi) ==
7367                     BTRFS_FILE_EXTENT_INLINE)
7368                         continue;
7369                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7370                         continue;
7371                 if (!inode || inode->i_ino != key.objectid) {
7372                         iput(inode);
7373                         inode = btrfs_ilookup(target_root->fs_info->sb,
7374                                               key.objectid, target_root, 1);
7375                 }
7376                 if (!inode) {
7377                         skip_objectid = key.objectid;
7378                         continue;
7379                 }
7380                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7381
7382                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7383                                  key.offset + num_bytes - 1, 0, &cached_state,
7384                                  GFP_NOFS);
7385                 btrfs_drop_extent_cache(inode, key.offset,
7386                                         key.offset + num_bytes - 1, 1);
7387                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7388                                      key.offset + num_bytes - 1, &cached_state,
7389                                      GFP_NOFS);
7390                 cond_resched();
7391         }
7392         iput(inode);
7393         return 0;
7394 }
7395
7396 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7397                                         struct btrfs_root *root,
7398                                         struct extent_buffer *leaf,
7399                                         struct btrfs_block_group_cache *group,
7400                                         struct inode *reloc_inode)
7401 {
7402         struct btrfs_key key;
7403         struct btrfs_key extent_key;
7404         struct btrfs_file_extent_item *fi;
7405         struct btrfs_leaf_ref *ref;
7406         struct disk_extent *new_extent;
7407         u64 bytenr;
7408         u64 num_bytes;
7409         u32 nritems;
7410         u32 i;
7411         int ext_index;
7412         int nr_extent;
7413         int ret;
7414
7415         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7416         BUG_ON(!new_extent);
7417
7418         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7419         BUG_ON(!ref);
7420
7421         ext_index = -1;
7422         nritems = btrfs_header_nritems(leaf);
7423         for (i = 0; i < nritems; i++) {
7424                 btrfs_item_key_to_cpu(leaf, &key, i);
7425                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7426                         continue;
7427                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7428                 if (btrfs_file_extent_type(leaf, fi) ==
7429                     BTRFS_FILE_EXTENT_INLINE)
7430                         continue;
7431                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7432                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7433                 if (bytenr == 0)
7434                         continue;
7435
7436                 ext_index++;
7437                 if (bytenr >= group->key.objectid + group->key.offset ||
7438                     bytenr + num_bytes <= group->key.objectid)
7439                         continue;
7440
7441                 extent_key.objectid = bytenr;
7442                 extent_key.offset = num_bytes;
7443                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7444                 nr_extent = 1;
7445                 ret = get_new_locations(reloc_inode, &extent_key,
7446                                         group->key.objectid, 1,
7447                                         &new_extent, &nr_extent);
7448                 if (ret > 0)
7449                         continue;
7450                 BUG_ON(ret < 0);
7451
7452                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7453                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7454                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7455                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7456
7457                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7458                                                 new_extent->disk_bytenr);
7459                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7460                                                 new_extent->disk_num_bytes);
7461                 btrfs_mark_buffer_dirty(leaf);
7462
7463                 ret = btrfs_inc_extent_ref(trans, root,
7464                                         new_extent->disk_bytenr,
7465                                         new_extent->disk_num_bytes,
7466                                         leaf->start,
7467                                         root->root_key.objectid,
7468                                         trans->transid, key.objectid);
7469                 BUG_ON(ret);
7470
7471                 ret = btrfs_free_extent(trans, root,
7472                                         bytenr, num_bytes, leaf->start,
7473                                         btrfs_header_owner(leaf),
7474                                         btrfs_header_generation(leaf),
7475                                         key.objectid, 0);
7476                 BUG_ON(ret);
7477                 cond_resched();
7478         }
7479         kfree(new_extent);
7480         BUG_ON(ext_index + 1 != ref->nritems);
7481         btrfs_free_leaf_ref(root, ref);
7482         return 0;
7483 }
7484
7485 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7486                           struct btrfs_root *root)
7487 {
7488         struct btrfs_root *reloc_root;
7489         int ret;
7490
7491         if (root->reloc_root) {
7492                 reloc_root = root->reloc_root;
7493                 root->reloc_root = NULL;
7494                 list_add(&reloc_root->dead_list,
7495                          &root->fs_info->dead_reloc_roots);
7496
7497                 btrfs_set_root_bytenr(&reloc_root->root_item,
7498                                       reloc_root->node->start);
7499                 btrfs_set_root_level(&root->root_item,
7500                                      btrfs_header_level(reloc_root->node));
7501                 memset(&reloc_root->root_item.drop_progress, 0,
7502                         sizeof(struct btrfs_disk_key));
7503                 reloc_root->root_item.drop_level = 0;
7504
7505                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7506                                         &reloc_root->root_key,
7507                                         &reloc_root->root_item);
7508                 BUG_ON(ret);
7509         }
7510         return 0;
7511 }
7512
7513 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7514 {
7515         struct btrfs_trans_handle *trans;
7516         struct btrfs_root *reloc_root;
7517         struct btrfs_root *prev_root = NULL;
7518         struct list_head dead_roots;
7519         int ret;
7520         unsigned long nr;
7521
7522         INIT_LIST_HEAD(&dead_roots);
7523         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7524
7525         while (!list_empty(&dead_roots)) {
7526                 reloc_root = list_entry(dead_roots.prev,
7527                                         struct btrfs_root, dead_list);
7528                 list_del_init(&reloc_root->dead_list);
7529
7530                 BUG_ON(reloc_root->commit_root != NULL);
7531                 while (1) {
7532                         trans = btrfs_join_transaction(root, 1);
7533                         BUG_ON(IS_ERR(trans));
7534
7535                         mutex_lock(&root->fs_info->drop_mutex);
7536                         ret = btrfs_drop_snapshot(trans, reloc_root);
7537                         if (ret != -EAGAIN)
7538                                 break;
7539                         mutex_unlock(&root->fs_info->drop_mutex);
7540
7541                         nr = trans->blocks_used;
7542                         ret = btrfs_end_transaction(trans, root);
7543                         BUG_ON(ret);
7544                         btrfs_btree_balance_dirty(root, nr);
7545                 }
7546
7547                 free_extent_buffer(reloc_root->node);
7548
7549                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7550                                      &reloc_root->root_key);
7551                 BUG_ON(ret);
7552                 mutex_unlock(&root->fs_info->drop_mutex);
7553
7554                 nr = trans->blocks_used;
7555                 ret = btrfs_end_transaction(trans, root);
7556                 BUG_ON(ret);
7557                 btrfs_btree_balance_dirty(root, nr);
7558
7559                 kfree(prev_root);
7560                 prev_root = reloc_root;
7561         }
7562         if (prev_root) {
7563                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7564                 kfree(prev_root);
7565         }
7566         return 0;
7567 }
7568
7569 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7570 {
7571         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7572         return 0;
7573 }
7574
7575 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7576 {
7577         struct btrfs_root *reloc_root;
7578         struct btrfs_trans_handle *trans;
7579         struct btrfs_key location;
7580         int found;
7581         int ret;
7582
7583         mutex_lock(&root->fs_info->tree_reloc_mutex);
7584         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7585         BUG_ON(ret);
7586         found = !list_empty(&root->fs_info->dead_reloc_roots);
7587         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7588
7589         if (found) {
7590                 trans = btrfs_start_transaction(root, 1);
7591                 BUG_ON(IS_ERR(trans));
7592                 ret = btrfs_commit_transaction(trans, root);
7593                 BUG_ON(ret);
7594         }
7595
7596         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7597         location.offset = (u64)-1;
7598         location.type = BTRFS_ROOT_ITEM_KEY;
7599
7600         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7601         BUG_ON(!reloc_root);
7602         btrfs_orphan_cleanup(reloc_root);
7603         return 0;
7604 }
7605
7606 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7607                                     struct btrfs_root *root)
7608 {
7609         struct btrfs_root *reloc_root;
7610         struct extent_buffer *eb;
7611         struct btrfs_root_item *root_item;
7612         struct btrfs_key root_key;
7613         int ret;
7614
7615         BUG_ON(!root->ref_cows);
7616         if (root->reloc_root)
7617                 return 0;
7618
7619         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7620         BUG_ON(!root_item);
7621
7622         ret = btrfs_copy_root(trans, root, root->commit_root,
7623                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7624         BUG_ON(ret);
7625
7626         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7627         root_key.offset = root->root_key.objectid;
7628         root_key.type = BTRFS_ROOT_ITEM_KEY;
7629
7630         memcpy(root_item, &root->root_item, sizeof(root_item));
7631         btrfs_set_root_refs(root_item, 0);
7632         btrfs_set_root_bytenr(root_item, eb->start);
7633         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7634         btrfs_set_root_generation(root_item, trans->transid);
7635
7636         btrfs_tree_unlock(eb);
7637         free_extent_buffer(eb);
7638
7639         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7640                                 &root_key, root_item);
7641         BUG_ON(ret);
7642         kfree(root_item);
7643
7644         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7645                                                  &root_key);
7646         BUG_ON(!reloc_root);
7647         reloc_root->last_trans = trans->transid;
7648         reloc_root->commit_root = NULL;
7649         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7650
7651         root->reloc_root = reloc_root;
7652         return 0;
7653 }
7654
7655 /*
7656  * Core function of space balance.
7657  *
7658  * The idea is using reloc trees to relocate tree blocks in reference
7659  * counted roots. There is one reloc tree for each subvol, and all
7660  * reloc trees share same root key objectid. Reloc trees are snapshots
7661  * of the latest committed roots of subvols (root->commit_root).
7662  *
7663  * To relocate a tree block referenced by a subvol, there are two steps.
7664  * COW the block through subvol's reloc tree, then update block pointer
7665  * in the subvol to point to the new block. Since all reloc trees share
7666  * same root key objectid, doing special handing for tree blocks owned
7667  * by them is easy. Once a tree block has been COWed in one reloc tree,
7668  * we can use the resulting new block directly when the same block is
7669  * required to COW again through other reloc trees. By this way, relocated
7670  * tree blocks are shared between reloc trees, so they are also shared
7671  * between subvols.
7672  */
7673 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7674                                       struct btrfs_root *root,
7675                                       struct btrfs_path *path,
7676                                       struct btrfs_key *first_key,
7677                                       struct btrfs_ref_path *ref_path,
7678                                       struct btrfs_block_group_cache *group,
7679                                       struct inode *reloc_inode)
7680 {
7681         struct btrfs_root *reloc_root;
7682         struct extent_buffer *eb = NULL;
7683         struct btrfs_key *keys;
7684         u64 *nodes;
7685         int level;
7686         int shared_level;
7687         int lowest_level = 0;
7688         int ret;
7689
7690         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7691                 lowest_level = ref_path->owner_objectid;
7692
7693         if (!root->ref_cows) {
7694                 path->lowest_level = lowest_level;
7695                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7696                 BUG_ON(ret < 0);
7697                 path->lowest_level = 0;
7698                 btrfs_release_path(root, path);
7699                 return 0;
7700         }
7701
7702         mutex_lock(&root->fs_info->tree_reloc_mutex);
7703         ret = init_reloc_tree(trans, root);
7704         BUG_ON(ret);
7705         reloc_root = root->reloc_root;
7706
7707         shared_level = ref_path->shared_level;
7708         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7709
7710         keys = ref_path->node_keys;
7711         nodes = ref_path->new_nodes;
7712         memset(&keys[shared_level + 1], 0,
7713                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7714         memset(&nodes[shared_level + 1], 0,
7715                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7716
7717         if (nodes[lowest_level] == 0) {
7718                 path->lowest_level = lowest_level;
7719                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7720                                         0, 1);
7721                 BUG_ON(ret);
7722                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7723                         eb = path->nodes[level];
7724                         if (!eb || eb == reloc_root->node)
7725                                 break;
7726                         nodes[level] = eb->start;
7727                         if (level == 0)
7728                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7729                         else
7730                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7731                 }
7732                 if (nodes[0] &&
7733                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7734                         eb = path->nodes[0];
7735                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7736                                                       group, reloc_inode);
7737                         BUG_ON(ret);
7738                 }
7739                 btrfs_release_path(reloc_root, path);
7740         } else {
7741                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7742                                        lowest_level);
7743                 BUG_ON(ret);
7744         }
7745
7746         /*
7747          * replace tree blocks in the fs tree with tree blocks in
7748          * the reloc tree.
7749          */
7750         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7751         BUG_ON(ret < 0);
7752
7753         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7754                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7755                                         0, 0);
7756                 BUG_ON(ret);
7757                 extent_buffer_get(path->nodes[0]);
7758                 eb = path->nodes[0];
7759                 btrfs_release_path(reloc_root, path);
7760                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7761                 BUG_ON(ret);
7762                 free_extent_buffer(eb);
7763         }
7764
7765         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7766         path->lowest_level = 0;
7767         return 0;
7768 }
7769
7770 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7771                                         struct btrfs_root *root,
7772                                         struct btrfs_path *path,
7773                                         struct btrfs_key *first_key,
7774                                         struct btrfs_ref_path *ref_path)
7775 {
7776         int ret;
7777
7778         ret = relocate_one_path(trans, root, path, first_key,
7779                                 ref_path, NULL, NULL);
7780         BUG_ON(ret);
7781
7782         return 0;
7783 }
7784
7785 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7786                                     struct btrfs_root *extent_root,
7787                                     struct btrfs_path *path,
7788                                     struct btrfs_key *extent_key)
7789 {
7790         int ret;
7791
7792         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7793         if (ret)
7794                 goto out;
7795         ret = btrfs_del_item(trans, extent_root, path);
7796 out:
7797         btrfs_release_path(extent_root, path);
7798         return ret;
7799 }
7800
7801 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7802                                                 struct btrfs_ref_path *ref_path)
7803 {
7804         struct btrfs_key root_key;
7805
7806         root_key.objectid = ref_path->root_objectid;
7807         root_key.type = BTRFS_ROOT_ITEM_KEY;
7808         if (is_cowonly_root(ref_path->root_objectid))
7809                 root_key.offset = 0;
7810         else
7811                 root_key.offset = (u64)-1;
7812
7813         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7814 }
7815
7816 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7817                                         struct btrfs_path *path,
7818                                         struct btrfs_key *extent_key,
7819                                         struct btrfs_block_group_cache *group,
7820                                         struct inode *reloc_inode, int pass)
7821 {
7822         struct btrfs_trans_handle *trans;
7823         struct btrfs_root *found_root;
7824         struct btrfs_ref_path *ref_path = NULL;
7825         struct disk_extent *new_extents = NULL;
7826         int nr_extents = 0;
7827         int loops;
7828         int ret;
7829         int level;
7830         struct btrfs_key first_key;
7831         u64 prev_block = 0;
7832
7833
7834         trans = btrfs_start_transaction(extent_root, 1);
7835         BUG_ON(IS_ERR(trans));
7836
7837         if (extent_key->objectid == 0) {
7838                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7839                 goto out;
7840         }
7841
7842         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7843         if (!ref_path) {
7844                 ret = -ENOMEM;
7845                 goto out;
7846         }
7847
7848         for (loops = 0; ; loops++) {
7849                 if (loops == 0) {
7850                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7851                                                    extent_key->objectid);
7852                 } else {
7853                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7854                 }
7855                 if (ret < 0)
7856                         goto out;
7857                 if (ret > 0)
7858                         break;
7859
7860                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7861                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7862                         continue;
7863
7864                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7865                 BUG_ON(!found_root);
7866                 /*
7867                  * for reference counted tree, only process reference paths
7868                  * rooted at the latest committed root.
7869                  */
7870                 if (found_root->ref_cows &&
7871                     ref_path->root_generation != found_root->root_key.offset)
7872                         continue;
7873
7874                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7875                         if (pass == 0) {
7876                                 /*
7877                                  * copy data extents to new locations
7878                                  */
7879                                 u64 group_start = group->key.objectid;
7880                                 ret = relocate_data_extent(reloc_inode,
7881                                                            extent_key,
7882                                                            group_start);
7883                                 if (ret < 0)
7884                                         goto out;
7885                                 break;
7886                         }
7887                         level = 0;
7888                 } else {
7889                         level = ref_path->owner_objectid;
7890                 }
7891
7892                 if (prev_block != ref_path->nodes[level]) {
7893                         struct extent_buffer *eb;
7894                         u64 block_start = ref_path->nodes[level];
7895                         u64 block_size = btrfs_level_size(found_root, level);
7896
7897                         eb = read_tree_block(found_root, block_start,
7898                                              block_size, 0);
7899                         btrfs_tree_lock(eb);
7900                         BUG_ON(level != btrfs_header_level(eb));
7901
7902                         if (level == 0)
7903                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7904                         else
7905                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7906
7907                         btrfs_tree_unlock(eb);
7908                         free_extent_buffer(eb);
7909                         prev_block = block_start;
7910                 }
7911
7912                 mutex_lock(&extent_root->fs_info->trans_mutex);
7913                 btrfs_record_root_in_trans(found_root);
7914                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7915                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7916                         /*
7917                          * try to update data extent references while
7918                          * keeping metadata shared between snapshots.
7919                          */
7920                         if (pass == 1) {
7921                                 ret = relocate_one_path(trans, found_root,
7922                                                 path, &first_key, ref_path,
7923                                                 group, reloc_inode);
7924                                 if (ret < 0)
7925                                         goto out;
7926                                 continue;
7927                         }
7928                         /*
7929                          * use fallback method to process the remaining
7930                          * references.
7931                          */
7932                         if (!new_extents) {
7933                                 u64 group_start = group->key.objectid;
7934                                 new_extents = kmalloc(sizeof(*new_extents),
7935                                                       GFP_NOFS);
7936                                 nr_extents = 1;
7937                                 ret = get_new_locations(reloc_inode,
7938                                                         extent_key,
7939                                                         group_start, 1,
7940                                                         &new_extents,
7941                                                         &nr_extents);
7942                                 if (ret)
7943                                         goto out;
7944                         }
7945                         ret = replace_one_extent(trans, found_root,
7946                                                 path, extent_key,
7947                                                 &first_key, ref_path,
7948                                                 new_extents, nr_extents);
7949                 } else {
7950                         ret = relocate_tree_block(trans, found_root, path,
7951                                                   &first_key, ref_path);
7952                 }
7953                 if (ret < 0)
7954                         goto out;
7955         }
7956         ret = 0;
7957 out:
7958         btrfs_end_transaction(trans, extent_root);
7959         kfree(new_extents);
7960         kfree(ref_path);
7961         return ret;
7962 }
7963 #endif
7964
7965 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7966 {
7967         u64 num_devices;
7968         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7969                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7970
7971         /*
7972          * we add in the count of missing devices because we want
7973          * to make sure that any RAID levels on a degraded FS
7974          * continue to be honored.
7975          */
7976         num_devices = root->fs_info->fs_devices->rw_devices +
7977                 root->fs_info->fs_devices->missing_devices;
7978
7979         if (num_devices == 1) {
7980                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7981                 stripped = flags & ~stripped;
7982
7983                 /* turn raid0 into single device chunks */
7984                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7985                         return stripped;
7986
7987                 /* turn mirroring into duplication */
7988                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7989                              BTRFS_BLOCK_GROUP_RAID10))
7990                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7991                 return flags;
7992         } else {
7993                 /* they already had raid on here, just return */
7994                 if (flags & stripped)
7995                         return flags;
7996
7997                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7998                 stripped = flags & ~stripped;
7999
8000                 /* switch duplicated blocks with raid1 */
8001                 if (flags & BTRFS_BLOCK_GROUP_DUP)
8002                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
8003
8004                 /* turn single device chunks into raid0 */
8005                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8006         }
8007         return flags;
8008 }
8009
8010 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
8011 {
8012         struct btrfs_space_info *sinfo = cache->space_info;
8013         u64 num_bytes;
8014         int ret = -ENOSPC;
8015
8016         if (cache->ro)
8017                 return 0;
8018
8019         spin_lock(&sinfo->lock);
8020         spin_lock(&cache->lock);
8021         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8022                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8023
8024         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8025             sinfo->bytes_may_use + sinfo->bytes_readonly +
8026             cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
8027                 sinfo->bytes_readonly += num_bytes;
8028                 sinfo->bytes_reserved += cache->reserved_pinned;
8029                 cache->reserved_pinned = 0;
8030                 cache->ro = 1;
8031                 ret = 0;
8032         }
8033
8034         spin_unlock(&cache->lock);
8035         spin_unlock(&sinfo->lock);
8036         return ret;
8037 }
8038
8039 int btrfs_set_block_group_ro(struct btrfs_root *root,
8040                              struct btrfs_block_group_cache *cache)
8041
8042 {
8043         struct btrfs_trans_handle *trans;
8044         u64 alloc_flags;
8045         int ret;
8046
8047         BUG_ON(cache->ro);
8048
8049         trans = btrfs_join_transaction(root, 1);
8050         BUG_ON(IS_ERR(trans));
8051
8052         alloc_flags = update_block_group_flags(root, cache->flags);
8053         if (alloc_flags != cache->flags)
8054                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8055
8056         ret = set_block_group_ro(cache);
8057         if (!ret)
8058                 goto out;
8059         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8060         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8061         if (ret < 0)
8062                 goto out;
8063         ret = set_block_group_ro(cache);
8064 out:
8065         btrfs_end_transaction(trans, root);
8066         return ret;
8067 }
8068
8069 /*
8070  * helper to account the unused space of all the readonly block group in the
8071  * list. takes mirrors into account.
8072  */
8073 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8074 {
8075         struct btrfs_block_group_cache *block_group;
8076         u64 free_bytes = 0;
8077         int factor;
8078
8079         list_for_each_entry(block_group, groups_list, list) {
8080                 spin_lock(&block_group->lock);
8081
8082                 if (!block_group->ro) {
8083                         spin_unlock(&block_group->lock);
8084                         continue;
8085                 }
8086
8087                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8088                                           BTRFS_BLOCK_GROUP_RAID10 |
8089                                           BTRFS_BLOCK_GROUP_DUP))
8090                         factor = 2;
8091                 else
8092                         factor = 1;
8093
8094                 free_bytes += (block_group->key.offset -
8095                                btrfs_block_group_used(&block_group->item)) *
8096                                factor;
8097
8098                 spin_unlock(&block_group->lock);
8099         }
8100
8101         return free_bytes;
8102 }
8103
8104 /*
8105  * helper to account the unused space of all the readonly block group in the
8106  * space_info. takes mirrors into account.
8107  */
8108 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8109 {
8110         int i;
8111         u64 free_bytes = 0;
8112
8113         spin_lock(&sinfo->lock);
8114
8115         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8116                 if (!list_empty(&sinfo->block_groups[i]))
8117                         free_bytes += __btrfs_get_ro_block_group_free_space(
8118                                                 &sinfo->block_groups[i]);
8119
8120         spin_unlock(&sinfo->lock);
8121
8122         return free_bytes;
8123 }
8124
8125 int btrfs_set_block_group_rw(struct btrfs_root *root,
8126                               struct btrfs_block_group_cache *cache)
8127 {
8128         struct btrfs_space_info *sinfo = cache->space_info;
8129         u64 num_bytes;
8130
8131         BUG_ON(!cache->ro);
8132
8133         spin_lock(&sinfo->lock);
8134         spin_lock(&cache->lock);
8135         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8136                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8137         sinfo->bytes_readonly -= num_bytes;
8138         cache->ro = 0;
8139         spin_unlock(&cache->lock);
8140         spin_unlock(&sinfo->lock);
8141         return 0;
8142 }
8143
8144 /*
8145  * checks to see if its even possible to relocate this block group.
8146  *
8147  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8148  * ok to go ahead and try.
8149  */
8150 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8151 {
8152         struct btrfs_block_group_cache *block_group;
8153         struct btrfs_space_info *space_info;
8154         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8155         struct btrfs_device *device;
8156         int full = 0;
8157         int ret = 0;
8158
8159         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8160
8161         /* odd, couldn't find the block group, leave it alone */
8162         if (!block_group)
8163                 return -1;
8164
8165         /* no bytes used, we're good */
8166         if (!btrfs_block_group_used(&block_group->item))
8167                 goto out;
8168
8169         space_info = block_group->space_info;
8170         spin_lock(&space_info->lock);
8171
8172         full = space_info->full;
8173
8174         /*
8175          * if this is the last block group we have in this space, we can't
8176          * relocate it unless we're able to allocate a new chunk below.
8177          *
8178          * Otherwise, we need to make sure we have room in the space to handle
8179          * all of the extents from this block group.  If we can, we're good
8180          */
8181         if ((space_info->total_bytes != block_group->key.offset) &&
8182            (space_info->bytes_used + space_info->bytes_reserved +
8183             space_info->bytes_pinned + space_info->bytes_readonly +
8184             btrfs_block_group_used(&block_group->item) <
8185             space_info->total_bytes)) {
8186                 spin_unlock(&space_info->lock);
8187                 goto out;
8188         }
8189         spin_unlock(&space_info->lock);
8190
8191         /*
8192          * ok we don't have enough space, but maybe we have free space on our
8193          * devices to allocate new chunks for relocation, so loop through our
8194          * alloc devices and guess if we have enough space.  However, if we
8195          * were marked as full, then we know there aren't enough chunks, and we
8196          * can just return.
8197          */
8198         ret = -1;
8199         if (full)
8200                 goto out;
8201
8202         mutex_lock(&root->fs_info->chunk_mutex);
8203         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8204                 u64 min_free = btrfs_block_group_used(&block_group->item);
8205                 u64 dev_offset;
8206
8207                 /*
8208                  * check to make sure we can actually find a chunk with enough
8209                  * space to fit our block group in.
8210                  */
8211                 if (device->total_bytes > device->bytes_used + min_free) {
8212                         ret = find_free_dev_extent(NULL, device, min_free,
8213                                                    &dev_offset, NULL);
8214                         if (!ret)
8215                                 break;
8216                         ret = -1;
8217                 }
8218         }
8219         mutex_unlock(&root->fs_info->chunk_mutex);
8220 out:
8221         btrfs_put_block_group(block_group);
8222         return ret;
8223 }
8224
8225 static int find_first_block_group(struct btrfs_root *root,
8226                 struct btrfs_path *path, struct btrfs_key *key)
8227 {
8228         int ret = 0;
8229         struct btrfs_key found_key;
8230         struct extent_buffer *leaf;
8231         int slot;
8232
8233         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8234         if (ret < 0)
8235                 goto out;
8236
8237         while (1) {
8238                 slot = path->slots[0];
8239                 leaf = path->nodes[0];
8240                 if (slot >= btrfs_header_nritems(leaf)) {
8241                         ret = btrfs_next_leaf(root, path);
8242                         if (ret == 0)
8243                                 continue;
8244                         if (ret < 0)
8245                                 goto out;
8246                         break;
8247                 }
8248                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8249
8250                 if (found_key.objectid >= key->objectid &&
8251                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8252                         ret = 0;
8253                         goto out;
8254                 }
8255                 path->slots[0]++;
8256         }
8257 out:
8258         return ret;
8259 }
8260
8261 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8262 {
8263         struct btrfs_block_group_cache *block_group;
8264         u64 last = 0;
8265
8266         while (1) {
8267                 struct inode *inode;
8268
8269                 block_group = btrfs_lookup_first_block_group(info, last);
8270                 while (block_group) {
8271                         spin_lock(&block_group->lock);
8272                         if (block_group->iref)
8273                                 break;
8274                         spin_unlock(&block_group->lock);
8275                         block_group = next_block_group(info->tree_root,
8276                                                        block_group);
8277                 }
8278                 if (!block_group) {
8279                         if (last == 0)
8280                                 break;
8281                         last = 0;
8282                         continue;
8283                 }
8284
8285                 inode = block_group->inode;
8286                 block_group->iref = 0;
8287                 block_group->inode = NULL;
8288                 spin_unlock(&block_group->lock);
8289                 iput(inode);
8290                 last = block_group->key.objectid + block_group->key.offset;
8291                 btrfs_put_block_group(block_group);
8292         }
8293 }
8294
8295 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8296 {
8297         struct btrfs_block_group_cache *block_group;
8298         struct btrfs_space_info *space_info;
8299         struct btrfs_caching_control *caching_ctl;
8300         struct rb_node *n;
8301
8302         down_write(&info->extent_commit_sem);
8303         while (!list_empty(&info->caching_block_groups)) {
8304                 caching_ctl = list_entry(info->caching_block_groups.next,
8305                                          struct btrfs_caching_control, list);
8306                 list_del(&caching_ctl->list);
8307                 put_caching_control(caching_ctl);
8308         }
8309         up_write(&info->extent_commit_sem);
8310
8311         spin_lock(&info->block_group_cache_lock);
8312         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8313                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8314                                        cache_node);
8315                 rb_erase(&block_group->cache_node,
8316                          &info->block_group_cache_tree);
8317                 spin_unlock(&info->block_group_cache_lock);
8318
8319                 down_write(&block_group->space_info->groups_sem);
8320                 list_del(&block_group->list);
8321                 up_write(&block_group->space_info->groups_sem);
8322
8323                 if (block_group->cached == BTRFS_CACHE_STARTED)
8324                         wait_block_group_cache_done(block_group);
8325
8326                 /*
8327                  * We haven't cached this block group, which means we could
8328                  * possibly have excluded extents on this block group.
8329                  */
8330                 if (block_group->cached == BTRFS_CACHE_NO)
8331                         free_excluded_extents(info->extent_root, block_group);
8332
8333                 btrfs_remove_free_space_cache(block_group);
8334                 btrfs_put_block_group(block_group);
8335
8336                 spin_lock(&info->block_group_cache_lock);
8337         }
8338         spin_unlock(&info->block_group_cache_lock);
8339
8340         /* now that all the block groups are freed, go through and
8341          * free all the space_info structs.  This is only called during
8342          * the final stages of unmount, and so we know nobody is
8343          * using them.  We call synchronize_rcu() once before we start,
8344          * just to be on the safe side.
8345          */
8346         synchronize_rcu();
8347
8348         release_global_block_rsv(info);
8349
8350         while(!list_empty(&info->space_info)) {
8351                 space_info = list_entry(info->space_info.next,
8352                                         struct btrfs_space_info,
8353                                         list);
8354                 if (space_info->bytes_pinned > 0 ||
8355                     space_info->bytes_reserved > 0) {
8356                         WARN_ON(1);
8357                         dump_space_info(space_info, 0, 0);
8358                 }
8359                 list_del(&space_info->list);
8360                 kfree(space_info);
8361         }
8362         return 0;
8363 }
8364
8365 static void __link_block_group(struct btrfs_space_info *space_info,
8366                                struct btrfs_block_group_cache *cache)
8367 {
8368         int index = get_block_group_index(cache);
8369
8370         down_write(&space_info->groups_sem);
8371         list_add_tail(&cache->list, &space_info->block_groups[index]);
8372         up_write(&space_info->groups_sem);
8373 }
8374
8375 int btrfs_read_block_groups(struct btrfs_root *root)
8376 {
8377         struct btrfs_path *path;
8378         int ret;
8379         struct btrfs_block_group_cache *cache;
8380         struct btrfs_fs_info *info = root->fs_info;
8381         struct btrfs_space_info *space_info;
8382         struct btrfs_key key;
8383         struct btrfs_key found_key;
8384         struct extent_buffer *leaf;
8385         int need_clear = 0;
8386         u64 cache_gen;
8387
8388         root = info->extent_root;
8389         key.objectid = 0;
8390         key.offset = 0;
8391         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8392         path = btrfs_alloc_path();
8393         if (!path)
8394                 return -ENOMEM;
8395
8396         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8397         if (cache_gen != 0 &&
8398             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8399                 need_clear = 1;
8400         if (btrfs_test_opt(root, CLEAR_CACHE))
8401                 need_clear = 1;
8402         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8403                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8404
8405         while (1) {
8406                 ret = find_first_block_group(root, path, &key);
8407                 if (ret > 0)
8408                         break;
8409                 if (ret != 0)
8410                         goto error;
8411                 leaf = path->nodes[0];
8412                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8413                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8414                 if (!cache) {
8415                         ret = -ENOMEM;
8416                         goto error;
8417                 }
8418
8419                 atomic_set(&cache->count, 1);
8420                 spin_lock_init(&cache->lock);
8421                 spin_lock_init(&cache->tree_lock);
8422                 cache->fs_info = info;
8423                 INIT_LIST_HEAD(&cache->list);
8424                 INIT_LIST_HEAD(&cache->cluster_list);
8425
8426                 if (need_clear)
8427                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8428
8429                 /*
8430                  * we only want to have 32k of ram per block group for keeping
8431                  * track of free space, and if we pass 1/2 of that we want to
8432                  * start converting things over to using bitmaps
8433                  */
8434                 cache->extents_thresh = ((1024 * 32) / 2) /
8435                         sizeof(struct btrfs_free_space);
8436
8437                 read_extent_buffer(leaf, &cache->item,
8438                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8439                                    sizeof(cache->item));
8440                 memcpy(&cache->key, &found_key, sizeof(found_key));
8441
8442                 key.objectid = found_key.objectid + found_key.offset;
8443                 btrfs_release_path(root, path);
8444                 cache->flags = btrfs_block_group_flags(&cache->item);
8445                 cache->sectorsize = root->sectorsize;
8446
8447                 /*
8448                  * We need to exclude the super stripes now so that the space
8449                  * info has super bytes accounted for, otherwise we'll think
8450                  * we have more space than we actually do.
8451                  */
8452                 exclude_super_stripes(root, cache);
8453
8454                 /*
8455                  * check for two cases, either we are full, and therefore
8456                  * don't need to bother with the caching work since we won't
8457                  * find any space, or we are empty, and we can just add all
8458                  * the space in and be done with it.  This saves us _alot_ of
8459                  * time, particularly in the full case.
8460                  */
8461                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8462                         cache->last_byte_to_unpin = (u64)-1;
8463                         cache->cached = BTRFS_CACHE_FINISHED;
8464                         free_excluded_extents(root, cache);
8465                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8466                         cache->last_byte_to_unpin = (u64)-1;
8467                         cache->cached = BTRFS_CACHE_FINISHED;
8468                         add_new_free_space(cache, root->fs_info,
8469                                            found_key.objectid,
8470                                            found_key.objectid +
8471                                            found_key.offset);
8472                         free_excluded_extents(root, cache);
8473                 }
8474
8475                 ret = update_space_info(info, cache->flags, found_key.offset,
8476                                         btrfs_block_group_used(&cache->item),
8477                                         &space_info);
8478                 BUG_ON(ret);
8479                 cache->space_info = space_info;
8480                 spin_lock(&cache->space_info->lock);
8481                 cache->space_info->bytes_readonly += cache->bytes_super;
8482                 spin_unlock(&cache->space_info->lock);
8483
8484                 __link_block_group(space_info, cache);
8485
8486                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8487                 BUG_ON(ret);
8488
8489                 set_avail_alloc_bits(root->fs_info, cache->flags);
8490                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8491                         set_block_group_ro(cache);
8492         }
8493
8494         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8495                 if (!(get_alloc_profile(root, space_info->flags) &
8496                       (BTRFS_BLOCK_GROUP_RAID10 |
8497                        BTRFS_BLOCK_GROUP_RAID1 |
8498                        BTRFS_BLOCK_GROUP_DUP)))
8499                         continue;
8500                 /*
8501                  * avoid allocating from un-mirrored block group if there are
8502                  * mirrored block groups.
8503                  */
8504                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8505                         set_block_group_ro(cache);
8506                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8507                         set_block_group_ro(cache);
8508         }
8509
8510         init_global_block_rsv(info);
8511         ret = 0;
8512 error:
8513         btrfs_free_path(path);
8514         return ret;
8515 }
8516
8517 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8518                            struct btrfs_root *root, u64 bytes_used,
8519                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8520                            u64 size)
8521 {
8522         int ret;
8523         struct btrfs_root *extent_root;
8524         struct btrfs_block_group_cache *cache;
8525
8526         extent_root = root->fs_info->extent_root;
8527
8528         root->fs_info->last_trans_log_full_commit = trans->transid;
8529
8530         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8531         if (!cache)
8532                 return -ENOMEM;
8533
8534         cache->key.objectid = chunk_offset;
8535         cache->key.offset = size;
8536         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8537         cache->sectorsize = root->sectorsize;
8538         cache->fs_info = root->fs_info;
8539
8540         /*
8541          * we only want to have 32k of ram per block group for keeping track
8542          * of free space, and if we pass 1/2 of that we want to start
8543          * converting things over to using bitmaps
8544          */
8545         cache->extents_thresh = ((1024 * 32) / 2) /
8546                 sizeof(struct btrfs_free_space);
8547         atomic_set(&cache->count, 1);
8548         spin_lock_init(&cache->lock);
8549         spin_lock_init(&cache->tree_lock);
8550         INIT_LIST_HEAD(&cache->list);
8551         INIT_LIST_HEAD(&cache->cluster_list);
8552
8553         btrfs_set_block_group_used(&cache->item, bytes_used);
8554         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8555         cache->flags = type;
8556         btrfs_set_block_group_flags(&cache->item, type);
8557
8558         cache->last_byte_to_unpin = (u64)-1;
8559         cache->cached = BTRFS_CACHE_FINISHED;
8560         exclude_super_stripes(root, cache);
8561
8562         add_new_free_space(cache, root->fs_info, chunk_offset,
8563                            chunk_offset + size);
8564
8565         free_excluded_extents(root, cache);
8566
8567         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8568                                 &cache->space_info);
8569         BUG_ON(ret);
8570
8571         spin_lock(&cache->space_info->lock);
8572         cache->space_info->bytes_readonly += cache->bytes_super;
8573         spin_unlock(&cache->space_info->lock);
8574
8575         __link_block_group(cache->space_info, cache);
8576
8577         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8578         BUG_ON(ret);
8579
8580         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8581                                 sizeof(cache->item));
8582         BUG_ON(ret);
8583
8584         set_avail_alloc_bits(extent_root->fs_info, type);
8585
8586         return 0;
8587 }
8588
8589 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8590                              struct btrfs_root *root, u64 group_start)
8591 {
8592         struct btrfs_path *path;
8593         struct btrfs_block_group_cache *block_group;
8594         struct btrfs_free_cluster *cluster;
8595         struct btrfs_root *tree_root = root->fs_info->tree_root;
8596         struct btrfs_key key;
8597         struct inode *inode;
8598         int ret;
8599         int factor;
8600
8601         root = root->fs_info->extent_root;
8602
8603         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8604         BUG_ON(!block_group);
8605         BUG_ON(!block_group->ro);
8606
8607         memcpy(&key, &block_group->key, sizeof(key));
8608         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8609                                   BTRFS_BLOCK_GROUP_RAID1 |
8610                                   BTRFS_BLOCK_GROUP_RAID10))
8611                 factor = 2;
8612         else
8613                 factor = 1;
8614
8615         /* make sure this block group isn't part of an allocation cluster */
8616         cluster = &root->fs_info->data_alloc_cluster;
8617         spin_lock(&cluster->refill_lock);
8618         btrfs_return_cluster_to_free_space(block_group, cluster);
8619         spin_unlock(&cluster->refill_lock);
8620
8621         /*
8622          * make sure this block group isn't part of a metadata
8623          * allocation cluster
8624          */
8625         cluster = &root->fs_info->meta_alloc_cluster;
8626         spin_lock(&cluster->refill_lock);
8627         btrfs_return_cluster_to_free_space(block_group, cluster);
8628         spin_unlock(&cluster->refill_lock);
8629
8630         path = btrfs_alloc_path();
8631         BUG_ON(!path);
8632
8633         inode = lookup_free_space_inode(root, block_group, path);
8634         if (!IS_ERR(inode)) {
8635                 btrfs_orphan_add(trans, inode);
8636                 clear_nlink(inode);
8637                 /* One for the block groups ref */
8638                 spin_lock(&block_group->lock);
8639                 if (block_group->iref) {
8640                         block_group->iref = 0;
8641                         block_group->inode = NULL;
8642                         spin_unlock(&block_group->lock);
8643                         iput(inode);
8644                 } else {
8645                         spin_unlock(&block_group->lock);
8646                 }
8647                 /* One for our lookup ref */
8648                 iput(inode);
8649         }
8650
8651         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8652         key.offset = block_group->key.objectid;
8653         key.type = 0;
8654
8655         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8656         if (ret < 0)
8657                 goto out;
8658         if (ret > 0)
8659                 btrfs_release_path(tree_root, path);
8660         if (ret == 0) {
8661                 ret = btrfs_del_item(trans, tree_root, path);
8662                 if (ret)
8663                         goto out;
8664                 btrfs_release_path(tree_root, path);
8665         }
8666
8667         spin_lock(&root->fs_info->block_group_cache_lock);
8668         rb_erase(&block_group->cache_node,
8669                  &root->fs_info->block_group_cache_tree);
8670         spin_unlock(&root->fs_info->block_group_cache_lock);
8671
8672         down_write(&block_group->space_info->groups_sem);
8673         /*
8674          * we must use list_del_init so people can check to see if they
8675          * are still on the list after taking the semaphore
8676          */
8677         list_del_init(&block_group->list);
8678         up_write(&block_group->space_info->groups_sem);
8679
8680         if (block_group->cached == BTRFS_CACHE_STARTED)
8681                 wait_block_group_cache_done(block_group);
8682
8683         btrfs_remove_free_space_cache(block_group);
8684
8685         spin_lock(&block_group->space_info->lock);
8686         block_group->space_info->total_bytes -= block_group->key.offset;
8687         block_group->space_info->bytes_readonly -= block_group->key.offset;
8688         block_group->space_info->disk_total -= block_group->key.offset * factor;
8689         spin_unlock(&block_group->space_info->lock);
8690
8691         memcpy(&key, &block_group->key, sizeof(key));
8692
8693         btrfs_clear_space_info_full(root->fs_info);
8694
8695         btrfs_put_block_group(block_group);
8696         btrfs_put_block_group(block_group);
8697
8698         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8699         if (ret > 0)
8700                 ret = -EIO;
8701         if (ret < 0)
8702                 goto out;
8703
8704         ret = btrfs_del_item(trans, root, path);
8705 out:
8706         btrfs_free_path(path);
8707         return ret;
8708 }
8709
8710 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8711 {
8712         return unpin_extent_range(root, start, end);
8713 }
8714
8715 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8716                                u64 num_bytes)
8717 {
8718         return btrfs_discard_extent(root, bytenr, num_bytes);
8719 }