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