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