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