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