fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/kthread.h>
  25 #include <linux/slab.h>
  26 #include "compat.h"
  27 #include "hash.h"
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "print-tree.h"
  31 #include "transaction.h"
  32 #include "volumes.h"
  33 #include "locking.h"
  34 #include "free-space-cache.h"
  35
  36 static int update_block_group(struct btrfs_trans_handle *trans,
  37                               struct btrfs_root *root,
  38                               u64 bytenr, u64 num_bytes, int alloc);
  39 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
  40                                  u64 num_bytes, int reserve, int sinfo);
  41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  42                                 struct btrfs_root *root,
  43                                 u64 bytenr, u64 num_bytes, u64 parent,
  44                                 u64 root_objectid, u64 owner_objectid,
  45                                 u64 owner_offset, int refs_to_drop,
  46                                 struct btrfs_delayed_extent_op *extra_op);
  47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  48                                     struct extent_buffer *leaf,
  49                                     struct btrfs_extent_item *ei);
  50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  51                                       struct btrfs_root *root,
  52                                       u64 parent, u64 root_objectid,
  53                                       u64 flags, u64 owner, u64 offset,
  54                                       struct btrfs_key *ins, int ref_mod);
  55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  56                                      struct btrfs_root *root,
  57                                      u64 parent, u64 root_objectid,
  58                                      u64 flags, struct btrfs_disk_key *key,
  59                                      int level, struct btrfs_key *ins);
  60 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  61                           struct btrfs_root *extent_root, u64 alloc_bytes,
  62                           u64 flags, int force);
  63 static int find_next_key(struct btrfs_path *path, int level,
  64                          struct btrfs_key *key);
  65 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  66                             int dump_block_groups);
  67
  68 static noinline int
  69 block_group_cache_done(struct btrfs_block_group_cache *cache)
  70 {
  71         smp_mb();
  72         return cache->cached == BTRFS_CACHE_FINISHED;
  73 }
  74
  75 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  76 {
  77         return (cache->flags & bits) == bits;
  78 }
  79
  80 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
  81 {
  82         atomic_inc(&cache->count);
  83 }
  84
  85 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
  86 {
  87         if (atomic_dec_and_test(&cache->count)) {
  88                 WARN_ON(cache->pinned > 0);
  89                 WARN_ON(cache->reserved > 0);
  90                 WARN_ON(cache->reserved_pinned > 0);
  91                 kfree(cache);
  92         }
  93 }
  94
  95 /*
  96  * this adds the block group to the fs_info rb tree for the block group
  97  * cache
  98  */
  99 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 100                                 struct btrfs_block_group_cache *block_group)
 101 {
 102         struct rb_node **p;
 103         struct rb_node *parent = NULL;
 104         struct btrfs_block_group_cache *cache;
 105
 106         spin_lock(&info->block_group_cache_lock);
 107         p = &info->block_group_cache_tree.rb_node;
 108
 109         while (*p) {
 110                 parent = *p;
 111                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 112                                  cache_node);
 113                 if (block_group->key.objectid < cache->key.objectid) {
 114                         p = &(*p)->rb_left;
 115                 } else if (block_group->key.objectid > cache->key.objectid) {
 116                         p = &(*p)->rb_right;
 117                 } else {
 118                         spin_unlock(&info->block_group_cache_lock);
 119                         return -EEXIST;
 120                 }
 121         }
 122
 123         rb_link_node(&block_group->cache_node, parent, p);
 124         rb_insert_color(&block_group->cache_node,
 125                         &info->block_group_cache_tree);
 126         spin_unlock(&info->block_group_cache_lock);
 127
 128         return 0;
 129 }
 130
 131 /*
 132  * This will return the block group at or after bytenr if contains is 0, else
 133  * it will return the block group that contains the bytenr
 134  */
 135 static struct btrfs_block_group_cache *
 136 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 137                               int contains)
 138 {
 139         struct btrfs_block_group_cache *cache, *ret = NULL;
 140         struct rb_node *n;
 141         u64 end, start;
 142
 143         spin_lock(&info->block_group_cache_lock);
 144         n = info->block_group_cache_tree.rb_node;
 145
 146         while (n) {
 147                 cache = rb_entry(n, struct btrfs_block_group_cache,
 148                                  cache_node);
 149                 end = cache->key.objectid + cache->key.offset - 1;
 150                 start = cache->key.objectid;
 151
 152                 if (bytenr < start) {
 153                         if (!contains && (!ret || start < ret->key.objectid))
 154                                 ret = cache;
 155                         n = n->rb_left;
 156                 } else if (bytenr > start) {
 157                         if (contains && bytenr <= end) {
 158                                 ret = cache;
 159                                 break;
 160                         }
 161                         n = n->rb_right;
 162                 } else {
 163                         ret = cache;
 164                         break;
 165                 }
 166         }
 167         if (ret)
 168                 btrfs_get_block_group(ret);
 169         spin_unlock(&info->block_group_cache_lock);
 170
 171         return ret;
 172 }
 173
 174 static int add_excluded_extent(struct btrfs_root *root,
 175                                u64 start, u64 num_bytes)
 176 {
 177         u64 end = start + num_bytes - 1;
 178         set_extent_bits(&root->fs_info->freed_extents[0],
 179                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 180         set_extent_bits(&root->fs_info->freed_extents[1],
 181                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 182         return 0;
 183 }
 184
 185 static void free_excluded_extents(struct btrfs_root *root,
 186                                   struct btrfs_block_group_cache *cache)
 187 {
 188         u64 start, end;
 189
 190         start = cache->key.objectid;
 191         end = start + cache->key.offset - 1;
 192
 193         clear_extent_bits(&root->fs_info->freed_extents[0],
 194                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 195         clear_extent_bits(&root->fs_info->freed_extents[1],
 196                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 197 }
 198
 199 static int exclude_super_stripes(struct btrfs_root *root,
 200                                  struct btrfs_block_group_cache *cache)
 201 {
 202         u64 bytenr;
 203         u64 *logical;
 204         int stripe_len;
 205         int i, nr, ret;
 206
 207         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 208                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 209                 cache->bytes_super += stripe_len;
 210                 ret = add_excluded_extent(root, cache->key.objectid,
 211                                           stripe_len);
 212                 BUG_ON(ret);
 213         }
 214
 215         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 216                 bytenr = btrfs_sb_offset(i);
 217                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 218                                        cache->key.objectid, bytenr,
 219                                        0, &logical, &nr, &stripe_len);
 220                 BUG_ON(ret);
 221
 222                 while (nr--) {
 223                         cache->bytes_super += stripe_len;
 224                         ret = add_excluded_extent(root, logical[nr],
 225                                                   stripe_len);
 226                         BUG_ON(ret);
 227                 }
 228
 229                 kfree(logical);
 230         }
 231         return 0;
 232 }
 233
 234 static struct btrfs_caching_control *
 235 get_caching_control(struct btrfs_block_group_cache *cache)
 236 {
 237         struct btrfs_caching_control *ctl;
 238
 239         spin_lock(&cache->lock);
 240         if (cache->cached != BTRFS_CACHE_STARTED) {
 241                 spin_unlock(&cache->lock);
 242                 return NULL;
 243         }
 244
 245         /* We're loading it the fast way, so we don't have a caching_ctl. */
 246         if (!cache->caching_ctl) {
 247                 spin_unlock(&cache->lock);
 248                 return NULL;
 249         }
 250
 251         ctl = cache->caching_ctl;
 252         atomic_inc(&ctl->count);
 253         spin_unlock(&cache->lock);
 254         return ctl;
 255 }
 256
 257 static void put_caching_control(struct btrfs_caching_control *ctl)
 258 {
 259         if (atomic_dec_and_test(&ctl->count))
 260                 kfree(ctl);
 261 }
 262
 263 /*
 264  * this is only called by cache_block_group, since we could have freed extents
 265  * we need to check the pinned_extents for any extents that can't be used yet
 266  * since their free space will be released as soon as the transaction commits.
 267  */
 268 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 269                               struct btrfs_fs_info *info, u64 start, u64 end)
 270 {
 271         u64 extent_start, extent_end, size, total_added = 0;
 272         int ret;
 273
 274         while (start < end) {
 275                 ret = find_first_extent_bit(info->pinned_extents, start,
 276                                             &extent_start, &extent_end,
 277                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 278                 if (ret)
 279                         break;
 280
 281                 if (extent_start <= start) {
 282                         start = extent_end + 1;
 283                 } else if (extent_start > start && extent_start < end) {
 284                         size = extent_start - start;
 285                         total_added += size;
 286                         ret = btrfs_add_free_space(block_group, start,
 287                                                    size);
 288                         BUG_ON(ret);
 289                         start = extent_end + 1;
 290                 } else {
 291                         break;
 292                 }
 293         }
 294
 295         if (start < end) {
 296                 size = end - start;
 297                 total_added += size;
 298                 ret = btrfs_add_free_space(block_group, start, size);
 299                 BUG_ON(ret);
 300         }
 301
 302         return total_added;
 303 }
 304
 305 static int caching_kthread(void *data)
 306 {
 307         struct btrfs_block_group_cache *block_group = data;
 308         struct btrfs_fs_info *fs_info = block_group->fs_info;
 309         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 310         struct btrfs_root *extent_root = fs_info->extent_root;
 311         struct btrfs_path *path;
 312         struct extent_buffer *leaf;
 313         struct btrfs_key key;
 314         u64 total_found = 0;
 315         u64 last = 0;
 316         u32 nritems;
 317         int ret = 0;
 318
 319         path = btrfs_alloc_path();
 320         if (!path)
 321                 return -ENOMEM;
 322
 323         exclude_super_stripes(extent_root, block_group);
 324         spin_lock(&block_group->space_info->lock);
 325         block_group->space_info->bytes_readonly += block_group->bytes_super;
 326         spin_unlock(&block_group->space_info->lock);
 327
 328         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 329
 330         /*
 331          * We don't want to deadlock with somebody trying to allocate a new
 332          * extent for the extent root while also trying to search the extent
 333          * root to add free space.  So we skip locking and search the commit
 334          * root, since its read-only
 335          */
 336         path->skip_locking = 1;
 337         path->search_commit_root = 1;
 338         path->reada = 2;
 339
 340         key.objectid = last;
 341         key.offset = 0;
 342         key.type = BTRFS_EXTENT_ITEM_KEY;
 343 again:
 344         mutex_lock(&caching_ctl->mutex);
 345         /* need to make sure the commit_root doesn't disappear */
 346         down_read(&fs_info->extent_commit_sem);
 347
 348         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 349         if (ret < 0)
 350                 goto err;
 351
 352         leaf = path->nodes[0];
 353         nritems = btrfs_header_nritems(leaf);
 354
 355         while (1) {
 356                 smp_mb();
 357                 if (fs_info->closing > 1) {
 358                         last = (u64)-1;
 359                         break;
 360                 }
 361
 362                 if (path->slots[0] < nritems) {
 363                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 364                 } else {
 365                         ret = find_next_key(path, 0, &key);
 366                         if (ret)
 367                                 break;
 368
 369                         caching_ctl->progress = last;
 370                         btrfs_release_path(extent_root, path);
 371                         up_read(&fs_info->extent_commit_sem);
 372                         mutex_unlock(&caching_ctl->mutex);
 373                         if (btrfs_transaction_in_commit(fs_info))
 374                                 schedule_timeout(1);
 375                         else
 376                                 cond_resched();
 377                         goto again;
 378                 }
 379
 380                 if (key.objectid < block_group->key.objectid) {
 381                         path->slots[0]++;
 382                         continue;
 383                 }
 384
 385                 if (key.objectid >= block_group->key.objectid +
 386                     block_group->key.offset)
 387                         break;
 388
 389                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 390                         total_found += add_new_free_space(block_group,
 391                                                           fs_info, last,
 392                                                           key.objectid);
 393                         last = key.objectid + key.offset;
 394
 395                         if (total_found > (1024 * 1024 * 2)) {
 396                                 total_found = 0;
 397                                 wake_up(&caching_ctl->wait);
 398                         }
 399                 }
 400                 path->slots[0]++;
 401         }
 402         ret = 0;
 403
 404         total_found += add_new_free_space(block_group, fs_info, last,
 405                                           block_group->key.objectid +
 406                                           block_group->key.offset);
 407         caching_ctl->progress = (u64)-1;
 408
 409         spin_lock(&block_group->lock);
 410         block_group->caching_ctl = NULL;
 411         block_group->cached = BTRFS_CACHE_FINISHED;
 412         spin_unlock(&block_group->lock);
 413
 414 err:
 415         btrfs_free_path(path);
 416         up_read(&fs_info->extent_commit_sem);
 417
 418         free_excluded_extents(extent_root, block_group);
 419
 420         mutex_unlock(&caching_ctl->mutex);
 421         wake_up(&caching_ctl->wait);
 422
 423         put_caching_control(caching_ctl);
 424         atomic_dec(&block_group->space_info->caching_threads);
 425         btrfs_put_block_group(block_group);
 426
 427         return 0;
 428 }
 429
 430 static int cache_block_group(struct btrfs_block_group_cache *cache,
 431                              struct btrfs_trans_handle *trans,
 432                              int load_cache_only)
 433 {
 434         struct btrfs_fs_info *fs_info = cache->fs_info;
 435         struct btrfs_caching_control *caching_ctl;
 436         struct task_struct *tsk;
 437         int ret = 0;
 438
 439         smp_mb();
 440         if (cache->cached != BTRFS_CACHE_NO)
 441                 return 0;
 442
 443         /*
 444          * We can't do the read from on-disk cache during a commit since we need
 445          * to have the normal tree locking.
 446          */
 447         if (!trans->transaction->in_commit) {
 448                 spin_lock(&cache->lock);
 449                 if (cache->cached != BTRFS_CACHE_NO) {
 450                         spin_unlock(&cache->lock);
 451                         return 0;
 452                 }
 453                 cache->cached = BTRFS_CACHE_STARTED;
 454                 spin_unlock(&cache->lock);
 455
 456                 ret = load_free_space_cache(fs_info, cache);
 457
 458                 spin_lock(&cache->lock);
 459                 if (ret == 1) {
 460                         cache->cached = BTRFS_CACHE_FINISHED;
 461                         cache->last_byte_to_unpin = (u64)-1;
 462                 } else {
 463                         cache->cached = BTRFS_CACHE_NO;
 464                 }
 465                 spin_unlock(&cache->lock);
 466                 if (ret == 1)
 467                         return 0;
 468         }
 469
 470         if (load_cache_only)
 471                 return 0;
 472
 473         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
 474         BUG_ON(!caching_ctl);
 475
 476         INIT_LIST_HEAD(&caching_ctl->list);
 477         mutex_init(&caching_ctl->mutex);
 478         init_waitqueue_head(&caching_ctl->wait);
 479         caching_ctl->block_group = cache;
 480         caching_ctl->progress = cache->key.objectid;
 481         /* one for caching kthread, one for caching block group list */
 482         atomic_set(&caching_ctl->count, 2);
 483
 484         spin_lock(&cache->lock);
 485         if (cache->cached != BTRFS_CACHE_NO) {
 486                 spin_unlock(&cache->lock);
 487                 kfree(caching_ctl);
 488                 return 0;
 489         }
 490         cache->caching_ctl = caching_ctl;
 491         cache->cached = BTRFS_CACHE_STARTED;
 492         spin_unlock(&cache->lock);
 493
 494         down_write(&fs_info->extent_commit_sem);
 495         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 496         up_write(&fs_info->extent_commit_sem);
 497
 498         atomic_inc(&cache->space_info->caching_threads);
 499         btrfs_get_block_group(cache);
 500
 501         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 502                           cache->key.objectid);
 503         if (IS_ERR(tsk)) {
 504                 ret = PTR_ERR(tsk);
 505                 printk(KERN_ERR "error running thread %d\n", ret);
 506                 BUG();
 507         }
 508
 509         return ret;
 510 }
 511
 512 /*
 513  * return the block group that starts at or after bytenr
 514  */
 515 static struct btrfs_block_group_cache *
 516 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 517 {
 518         struct btrfs_block_group_cache *cache;
 519
 520         cache = block_group_cache_tree_search(info, bytenr, 0);
 521
 522         return cache;
 523 }
 524
 525 /*
 526  * return the block group that contains the given bytenr
 527  */
 528 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 529                                                  struct btrfs_fs_info *info,
 530                                                  u64 bytenr)
 531 {
 532         struct btrfs_block_group_cache *cache;
 533
 534         cache = block_group_cache_tree_search(info, bytenr, 1);
 535
 536         return cache;
 537 }
 538
 539 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 540                                                   u64 flags)
 541 {
 542         struct list_head *head = &info->space_info;
 543         struct btrfs_space_info *found;
 544
 545         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 546                  BTRFS_BLOCK_GROUP_METADATA;
 547
 548         rcu_read_lock();
 549         list_for_each_entry_rcu(found, head, list) {
 550                 if (found->flags & flags) {
 551                         rcu_read_unlock();
 552                         return found;
 553                 }
 554         }
 555         rcu_read_unlock();
 556         return NULL;
 557 }
 558
 559 /*
 560  * after adding space to the filesystem, we need to clear the full flags
 561  * on all the space infos.
 562  */
 563 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 564 {
 565         struct list_head *head = &info->space_info;
 566         struct btrfs_space_info *found;
 567
 568         rcu_read_lock();
 569         list_for_each_entry_rcu(found, head, list)
 570                 found->full = 0;
 571         rcu_read_unlock();
 572 }
 573
 574 static u64 div_factor(u64 num, int factor)
 575 {
 576         if (factor == 10)
 577                 return num;
 578         num *= factor;
 579         do_div(num, 10);
 580         return num;
 581 }
 582
 583 static u64 div_factor_fine(u64 num, int factor)
 584 {
 585         if (factor == 100)
 586                 return num;
 587         num *= factor;
 588         do_div(num, 100);
 589         return num;
 590 }
 591
 592 u64 btrfs_find_block_group(struct btrfs_root *root,
 593                            u64 search_start, u64 search_hint, int owner)
 594 {
 595         struct btrfs_block_group_cache *cache;
 596         u64 used;
 597         u64 last = max(search_hint, search_start);
 598         u64 group_start = 0;
 599         int full_search = 0;
 600         int factor = 9;
 601         int wrapped = 0;
 602 again:
 603         while (1) {
 604                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 605                 if (!cache)
 606                         break;
 607
 608                 spin_lock(&cache->lock);
 609                 last = cache->key.objectid + cache->key.offset;
 610                 used = btrfs_block_group_used(&cache->item);
 611
 612                 if ((full_search || !cache->ro) &&
 613                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 614                         if (used + cache->pinned + cache->reserved <
 615                             div_factor(cache->key.offset, factor)) {
 616                                 group_start = cache->key.objectid;
 617                                 spin_unlock(&cache->lock);
 618                                 btrfs_put_block_group(cache);
 619                                 goto found;
 620                         }
 621                 }
 622                 spin_unlock(&cache->lock);
 623                 btrfs_put_block_group(cache);
 624                 cond_resched();
 625         }
 626         if (!wrapped) {
 627                 last = search_start;
 628                 wrapped = 1;
 629                 goto again;
 630         }
 631         if (!full_search && factor < 10) {
 632                 last = search_start;
 633                 full_search = 1;
 634                 factor = 10;
 635                 goto again;
 636         }
 637 found:
 638         return group_start;
 639 }
 640
 641 /* simple helper to search for an existing extent at a given offset */
 642 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 643 {
 644         int ret;
 645         struct btrfs_key key;
 646         struct btrfs_path *path;
 647
 648         path = btrfs_alloc_path();
 649         BUG_ON(!path);
 650         key.objectid = start;
 651         key.offset = len;
 652         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 653         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 654                                 0, 0);
 655         btrfs_free_path(path);
 656         return ret;
 657 }
 658
 659 /*
 660  * helper function to lookup reference count and flags of extent.
 661  *
 662  * the head node for delayed ref is used to store the sum of all the
 663  * reference count modifications queued up in the rbtree. the head
 664  * node may also store the extent flags to set. This way you can check
 665  * to see what the reference count and extent flags would be if all of
 666  * the delayed refs are not processed.
 667  */
 668 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 669                              struct btrfs_root *root, u64 bytenr,
 670                              u64 num_bytes, u64 *refs, u64 *flags)
 671 {
 672         struct btrfs_delayed_ref_head *head;
 673         struct btrfs_delayed_ref_root *delayed_refs;
 674         struct btrfs_path *path;
 675         struct btrfs_extent_item *ei;
 676         struct extent_buffer *leaf;
 677         struct btrfs_key key;
 678         u32 item_size;
 679         u64 num_refs;
 680         u64 extent_flags;
 681         int ret;
 682
 683         path = btrfs_alloc_path();
 684         if (!path)
 685                 return -ENOMEM;
 686
 687         key.objectid = bytenr;
 688         key.type = BTRFS_EXTENT_ITEM_KEY;
 689         key.offset = num_bytes;
 690         if (!trans) {
 691                 path->skip_locking = 1;
 692                 path->search_commit_root = 1;
 693         }
 694 again:
 695         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 696                                 &key, path, 0, 0);
 697         if (ret < 0)
 698                 goto out_free;
 699
 700         if (ret == 0) {
 701                 leaf = path->nodes[0];
 702                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 703                 if (item_size >= sizeof(*ei)) {
 704                         ei = btrfs_item_ptr(leaf, path->slots[0],
 705                                             struct btrfs_extent_item);
 706                         num_refs = btrfs_extent_refs(leaf, ei);
 707                         extent_flags = btrfs_extent_flags(leaf, ei);
 708                 } else {
 709 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 710                         struct btrfs_extent_item_v0 *ei0;
 711                         BUG_ON(item_size != sizeof(*ei0));
 712                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 713                                              struct btrfs_extent_item_v0);
 714                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 715                         /* FIXME: this isn't correct for data */
 716                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 717 #else
 718                         BUG();
 719 #endif
 720                 }
 721                 BUG_ON(num_refs == 0);
 722         } else {
 723                 num_refs = 0;
 724                 extent_flags = 0;
 725                 ret = 0;
 726         }
 727
 728         if (!trans)
 729                 goto out;
 730
 731         delayed_refs = &trans->transaction->delayed_refs;
 732         spin_lock(&delayed_refs->lock);
 733         head = btrfs_find_delayed_ref_head(trans, bytenr);
 734         if (head) {
 735                 if (!mutex_trylock(&head->mutex)) {
 736                         atomic_inc(&head->node.refs);
 737                         spin_unlock(&delayed_refs->lock);
 738
 739                         btrfs_release_path(root->fs_info->extent_root, path);
 740
 741                         mutex_lock(&head->mutex);
 742                         mutex_unlock(&head->mutex);
 743                         btrfs_put_delayed_ref(&head->node);
 744                         goto again;
 745                 }
 746                 if (head->extent_op && head->extent_op->update_flags)
 747                         extent_flags |= head->extent_op->flags_to_set;
 748                 else
 749                         BUG_ON(num_refs == 0);
 750
 751                 num_refs += head->node.ref_mod;
 752                 mutex_unlock(&head->mutex);
 753         }
 754         spin_unlock(&delayed_refs->lock);
 755 out:
 756         WARN_ON(num_refs == 0);
 757         if (refs)
 758                 *refs = num_refs;
 759         if (flags)
 760                 *flags = extent_flags;
 761 out_free:
 762         btrfs_free_path(path);
 763         return ret;
 764 }
 765
 766 /*
 767  * Back reference rules.  Back refs have three main goals:
 768  *
 769  * 1) differentiate between all holders of references to an extent so that
 770  *    when a reference is dropped we can make sure it was a valid reference
 771  *    before freeing the extent.
 772  *
 773  * 2) Provide enough information to quickly find the holders of an extent
 774  *    if we notice a given block is corrupted or bad.
 775  *
 776  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 777  *    maintenance.  This is actually the same as #2, but with a slightly
 778  *    different use case.
 779  *
 780  * There are two kinds of back refs. The implicit back refs is optimized
 781  * for pointers in non-shared tree blocks. For a given pointer in a block,
 782  * back refs of this kind provide information about the block's owner tree
 783  * and the pointer's key. These information allow us to find the block by
 784  * b-tree searching. The full back refs is for pointers in tree blocks not
 785  * referenced by their owner trees. The location of tree block is recorded
 786  * in the back refs. Actually the full back refs is generic, and can be
 787  * used in all cases the implicit back refs is used. The major shortcoming
 788  * of the full back refs is its overhead. Every time a tree block gets
 789  * COWed, we have to update back refs entry for all pointers in it.
 790  *
 791  * For a newly allocated tree block, we use implicit back refs for
 792  * pointers in it. This means most tree related operations only involve
 793  * implicit back refs. For a tree block created in old transaction, the
 794  * only way to drop a reference to it is COW it. So we can detect the
 795  * event that tree block loses its owner tree's reference and do the
 796  * back refs conversion.
 797  *
 798  * When a tree block is COW'd through a tree, there are four cases:
 799  *
 800  * The reference count of the block is one and the tree is the block's
 801  * owner tree. Nothing to do in this case.
 802  *
 803  * The reference count of the block is one and the tree is not the
 804  * block's owner tree. In this case, full back refs is used for pointers
 805  * in the block. Remove these full back refs, add implicit back refs for
 806  * every pointers in the new block.
 807  *
 808  * The reference count of the block is greater than one and the tree is
 809  * the block's owner tree. In this case, implicit back refs is used for
 810  * pointers in the block. Add full back refs for every pointers in the
 811  * block, increase lower level extents' reference counts. The original
 812  * implicit back refs are entailed to the new block.
 813  *
 814  * The reference count of the block is greater than one and the tree is
 815  * not the block's owner tree. Add implicit back refs for every pointer in
 816  * the new block, increase lower level extents' reference count.
 817  *
 818  * Back Reference Key composing:
 819  *
 820  * The key objectid corresponds to the first byte in the extent,
 821  * The key type is used to differentiate between types of back refs.
 822  * There are different meanings of the key offset for different types
 823  * of back refs.
 824  *
 825  * File extents can be referenced by:
 826  *
 827  * - multiple snapshots, subvolumes, or different generations in one subvol
 828  * - different files inside a single subvolume
 829  * - different offsets inside a file (bookend extents in file.c)
 830  *
 831  * The extent ref structure for the implicit back refs has fields for:
 832  *
 833  * - Objectid of the subvolume root
 834  * - objectid of the file holding the reference
 835  * - original offset in the file
 836  * - how many bookend extents
 837  *
 838  * The key offset for the implicit back refs is hash of the first
 839  * three fields.
 840  *
 841  * The extent ref structure for the full back refs has field for:
 842  *
 843  * - number of pointers in the tree leaf
 844  *
 845  * The key offset for the implicit back refs is the first byte of
 846  * the tree leaf
 847  *
 848  * When a file extent is allocated, The implicit back refs is used.
 849  * the fields are filled in:
 850  *
 851  *     (root_key.objectid, inode objectid, offset in file, 1)
 852  *
 853  * When a file extent is removed file truncation, we find the
 854  * corresponding implicit back refs and check the following fields:
 855  *
 856  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 857  *
 858  * Btree extents can be referenced by:
 859  *
 860  * - Different subvolumes
 861  *
 862  * Both the implicit back refs and the full back refs for tree blocks
 863  * only consist of key. The key offset for the implicit back refs is
 864  * objectid of block's owner tree. The key offset for the full back refs
 865  * is the first byte of parent block.
 866  *
 867  * When implicit back refs is used, information about the lowest key and
 868  * level of the tree block are required. These information are stored in
 869  * tree block info structure.
 870  */
 871
 872 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 873 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 874                                   struct btrfs_root *root,
 875                                   struct btrfs_path *path,
 876                                   u64 owner, u32 extra_size)
 877 {
 878         struct btrfs_extent_item *item;
 879         struct btrfs_extent_item_v0 *ei0;
 880         struct btrfs_extent_ref_v0 *ref0;
 881         struct btrfs_tree_block_info *bi;
 882         struct extent_buffer *leaf;
 883         struct btrfs_key key;
 884         struct btrfs_key found_key;
 885         u32 new_size = sizeof(*item);
 886         u64 refs;
 887         int ret;
 888
 889         leaf = path->nodes[0];
 890         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 891
 892         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 893         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 894                              struct btrfs_extent_item_v0);
 895         refs = btrfs_extent_refs_v0(leaf, ei0);
 896
 897         if (owner == (u64)-1) {
 898                 while (1) {
 899                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 900                                 ret = btrfs_next_leaf(root, path);
 901                                 if (ret < 0)
 902                                         return ret;
 903                                 BUG_ON(ret > 0);
 904                                 leaf = path->nodes[0];
 905                         }
 906                         btrfs_item_key_to_cpu(leaf, &found_key,
 907                                               path->slots[0]);
 908                         BUG_ON(key.objectid != found_key.objectid);
 909                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 910                                 path->slots[0]++;
 911                                 continue;
 912                         }
 913                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 914                                               struct btrfs_extent_ref_v0);
 915                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 916                         break;
 917                 }
 918         }
 919         btrfs_release_path(root, path);
 920
 921         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 922                 new_size += sizeof(*bi);
 923
 924         new_size -= sizeof(*ei0);
 925         ret = btrfs_search_slot(trans, root, &key, path,
 926                                 new_size + extra_size, 1);
 927         if (ret < 0)
 928                 return ret;
 929         BUG_ON(ret);
 930
 931         ret = btrfs_extend_item(trans, root, path, new_size);
 932         BUG_ON(ret);
 933
 934         leaf = path->nodes[0];
 935         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 936         btrfs_set_extent_refs(leaf, item, refs);
 937         /* FIXME: get real generation */
 938         btrfs_set_extent_generation(leaf, item, 0);
 939         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 940                 btrfs_set_extent_flags(leaf, item,
 941                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 942                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 943                 bi = (struct btrfs_tree_block_info *)(item + 1);
 944                 /* FIXME: get first key of the block */
 945                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 946                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 947         } else {
 948                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 949         }
 950         btrfs_mark_buffer_dirty(leaf);
 951         return 0;
 952 }
 953 #endif
 954
 955 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 956 {
 957         u32 high_crc = ~(u32)0;
 958         u32 low_crc = ~(u32)0;
 959         __le64 lenum;
 960
 961         lenum = cpu_to_le64(root_objectid);
 962         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 963         lenum = cpu_to_le64(owner);
 964         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 965         lenum = cpu_to_le64(offset);
 966         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 967
 968         return ((u64)high_crc << 31) ^ (u64)low_crc;
 969 }
 970
 971 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 972                                      struct btrfs_extent_data_ref *ref)
 973 {
 974         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 975                                     btrfs_extent_data_ref_objectid(leaf, ref),
 976                                     btrfs_extent_data_ref_offset(leaf, ref));
 977 }
 978
 979 static int match_extent_data_ref(struct extent_buffer *leaf,
 980                                  struct btrfs_extent_data_ref *ref,
 981                                  u64 root_objectid, u64 owner, u64 offset)
 982 {
 983         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 984             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 985             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 986                 return 0;
 987         return 1;
 988 }
 989
 990 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 991                                            struct btrfs_root *root,
 992                                            struct btrfs_path *path,
 993                                            u64 bytenr, u64 parent,
 994                                            u64 root_objectid,
 995                                            u64 owner, u64 offset)
 996 {
 997         struct btrfs_key key;
 998         struct btrfs_extent_data_ref *ref;
 999         struct extent_buffer *leaf;
1000         u32 nritems;
1001         int ret;
1002         int recow;
1003         int err = -ENOENT;
1004
1005         key.objectid = bytenr;
1006         if (parent) {
1007                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1008                 key.offset = parent;
1009         } else {
1010                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1011                 key.offset = hash_extent_data_ref(root_objectid,
1012                                                   owner, offset);
1013         }
1014 again:
1015         recow = 0;
1016         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1017         if (ret < 0) {
1018                 err = ret;
1019                 goto fail;
1020         }
1021
1022         if (parent) {
1023                 if (!ret)
1024                         return 0;
1025 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1026                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1027                 btrfs_release_path(root, path);
1028                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1029                 if (ret < 0) {
1030                         err = ret;
1031                         goto fail;
1032                 }
1033                 if (!ret)
1034                         return 0;
1035 #endif
1036                 goto fail;
1037         }
1038
1039         leaf = path->nodes[0];
1040         nritems = btrfs_header_nritems(leaf);
1041         while (1) {
1042                 if (path->slots[0] >= nritems) {
1043                         ret = btrfs_next_leaf(root, path);
1044                         if (ret < 0)
1045                                 err = ret;
1046                         if (ret)
1047                                 goto fail;
1048
1049                         leaf = path->nodes[0];
1050                         nritems = btrfs_header_nritems(leaf);
1051                         recow = 1;
1052                 }
1053
1054                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1055                 if (key.objectid != bytenr ||
1056                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1057                         goto fail;
1058
1059                 ref = btrfs_item_ptr(leaf, path->slots[0],
1060                                      struct btrfs_extent_data_ref);
1061
1062                 if (match_extent_data_ref(leaf, ref, root_objectid,
1063                                           owner, offset)) {
1064                         if (recow) {
1065                                 btrfs_release_path(root, path);
1066                                 goto again;
1067                         }
1068                         err = 0;
1069                         break;
1070                 }
1071                 path->slots[0]++;
1072         }
1073 fail:
1074         return err;
1075 }
1076
1077 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1078                                            struct btrfs_root *root,
1079                                            struct btrfs_path *path,
1080                                            u64 bytenr, u64 parent,
1081                                            u64 root_objectid, u64 owner,
1082                                            u64 offset, int refs_to_add)
1083 {
1084         struct btrfs_key key;
1085         struct extent_buffer *leaf;
1086         u32 size;
1087         u32 num_refs;
1088         int ret;
1089
1090         key.objectid = bytenr;
1091         if (parent) {
1092                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1093                 key.offset = parent;
1094                 size = sizeof(struct btrfs_shared_data_ref);
1095         } else {
1096                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1097                 key.offset = hash_extent_data_ref(root_objectid,
1098                                                   owner, offset);
1099                 size = sizeof(struct btrfs_extent_data_ref);
1100         }
1101
1102         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1103         if (ret && ret != -EEXIST)
1104                 goto fail;
1105
1106         leaf = path->nodes[0];
1107         if (parent) {
1108                 struct btrfs_shared_data_ref *ref;
1109                 ref = btrfs_item_ptr(leaf, path->slots[0],
1110                                      struct btrfs_shared_data_ref);
1111                 if (ret == 0) {
1112                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1113                 } else {
1114                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1115                         num_refs += refs_to_add;
1116                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1117                 }
1118         } else {
1119                 struct btrfs_extent_data_ref *ref;
1120                 while (ret == -EEXIST) {
1121                         ref = btrfs_item_ptr(leaf, path->slots[0],
1122                                              struct btrfs_extent_data_ref);
1123                         if (match_extent_data_ref(leaf, ref, root_objectid,
1124                                                   owner, offset))
1125                                 break;
1126                         btrfs_release_path(root, path);
1127                         key.offset++;
1128                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1129                                                       size);
1130                         if (ret && ret != -EEXIST)
1131                                 goto fail;
1132
1133                         leaf = path->nodes[0];
1134                 }
1135                 ref = btrfs_item_ptr(leaf, path->slots[0],
1136                                      struct btrfs_extent_data_ref);
1137                 if (ret == 0) {
1138                         btrfs_set_extent_data_ref_root(leaf, ref,
1139                                                        root_objectid);
1140                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1141                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1142                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1143                 } else {
1144                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1145                         num_refs += refs_to_add;
1146                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1147                 }
1148         }
1149         btrfs_mark_buffer_dirty(leaf);
1150         ret = 0;
1151 fail:
1152         btrfs_release_path(root, path);
1153         return ret;
1154 }
1155
1156 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1157                                            struct btrfs_root *root,
1158                                            struct btrfs_path *path,
1159                                            int refs_to_drop)
1160 {
1161         struct btrfs_key key;
1162         struct btrfs_extent_data_ref *ref1 = NULL;
1163         struct btrfs_shared_data_ref *ref2 = NULL;
1164         struct extent_buffer *leaf;
1165         u32 num_refs = 0;
1166         int ret = 0;
1167
1168         leaf = path->nodes[0];
1169         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1170
1171         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1172                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1173                                       struct btrfs_extent_data_ref);
1174                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1175         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1176                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1177                                       struct btrfs_shared_data_ref);
1178                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1179 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1180         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1181                 struct btrfs_extent_ref_v0 *ref0;
1182                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1183                                       struct btrfs_extent_ref_v0);
1184                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1185 #endif
1186         } else {
1187                 BUG();
1188         }
1189
1190         BUG_ON(num_refs < refs_to_drop);
1191         num_refs -= refs_to_drop;
1192
1193         if (num_refs == 0) {
1194                 ret = btrfs_del_item(trans, root, path);
1195         } else {
1196                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1197                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1198                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1199                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1200 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1201                 else {
1202                         struct btrfs_extent_ref_v0 *ref0;
1203                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1204                                         struct btrfs_extent_ref_v0);
1205                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1206                 }
1207 #endif
1208                 btrfs_mark_buffer_dirty(leaf);
1209         }
1210         return ret;
1211 }
1212
1213 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1214                                           struct btrfs_path *path,
1215                                           struct btrfs_extent_inline_ref *iref)
1216 {
1217         struct btrfs_key key;
1218         struct extent_buffer *leaf;
1219         struct btrfs_extent_data_ref *ref1;
1220         struct btrfs_shared_data_ref *ref2;
1221         u32 num_refs = 0;
1222
1223         leaf = path->nodes[0];
1224         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1225         if (iref) {
1226                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1227                     BTRFS_EXTENT_DATA_REF_KEY) {
1228                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1229                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1230                 } else {
1231                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1232                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1233                 }
1234         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1235                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1236                                       struct btrfs_extent_data_ref);
1237                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1238         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1239                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1240                                       struct btrfs_shared_data_ref);
1241                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1242 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1243         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1244                 struct btrfs_extent_ref_v0 *ref0;
1245                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1246                                       struct btrfs_extent_ref_v0);
1247                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1248 #endif
1249         } else {
1250                 WARN_ON(1);
1251         }
1252         return num_refs;
1253 }
1254
1255 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1256                                           struct btrfs_root *root,
1257                                           struct btrfs_path *path,
1258                                           u64 bytenr, u64 parent,
1259                                           u64 root_objectid)
1260 {
1261         struct btrfs_key key;
1262         int ret;
1263
1264         key.objectid = bytenr;
1265         if (parent) {
1266                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1267                 key.offset = parent;
1268         } else {
1269                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1270                 key.offset = root_objectid;
1271         }
1272
1273         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1274         if (ret > 0)
1275                 ret = -ENOENT;
1276 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1277         if (ret == -ENOENT && parent) {
1278                 btrfs_release_path(root, path);
1279                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1280                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1281                 if (ret > 0)
1282                         ret = -ENOENT;
1283         }
1284 #endif
1285         return ret;
1286 }
1287
1288 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1289                                           struct btrfs_root *root,
1290                                           struct btrfs_path *path,
1291                                           u64 bytenr, u64 parent,
1292                                           u64 root_objectid)
1293 {
1294         struct btrfs_key key;
1295         int ret;
1296
1297         key.objectid = bytenr;
1298         if (parent) {
1299                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1300                 key.offset = parent;
1301         } else {
1302                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1303                 key.offset = root_objectid;
1304         }
1305
1306         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1307         btrfs_release_path(root, path);
1308         return ret;
1309 }
1310
1311 static inline int extent_ref_type(u64 parent, u64 owner)
1312 {
1313         int type;
1314         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1315                 if (parent > 0)
1316                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1317                 else
1318                         type = BTRFS_TREE_BLOCK_REF_KEY;
1319         } else {
1320                 if (parent > 0)
1321                         type = BTRFS_SHARED_DATA_REF_KEY;
1322                 else
1323                         type = BTRFS_EXTENT_DATA_REF_KEY;
1324         }
1325         return type;
1326 }
1327
1328 static int find_next_key(struct btrfs_path *path, int level,
1329                          struct btrfs_key *key)
1330
1331 {
1332         for (; level < BTRFS_MAX_LEVEL; level++) {
1333                 if (!path->nodes[level])
1334                         break;
1335                 if (path->slots[level] + 1 >=
1336                     btrfs_header_nritems(path->nodes[level]))
1337                         continue;
1338                 if (level == 0)
1339                         btrfs_item_key_to_cpu(path->nodes[level], key,
1340                                               path->slots[level] + 1);
1341                 else
1342                         btrfs_node_key_to_cpu(path->nodes[level], key,
1343                                               path->slots[level] + 1);
1344                 return 0;
1345         }
1346         return 1;
1347 }
1348
1349 /*
1350  * look for inline back ref. if back ref is found, *ref_ret is set
1351  * to the address of inline back ref, and 0 is returned.
1352  *
1353  * if back ref isn't found, *ref_ret is set to the address where it
1354  * should be inserted, and -ENOENT is returned.
1355  *
1356  * if insert is true and there are too many inline back refs, the path
1357  * points to the extent item, and -EAGAIN is returned.
1358  *
1359  * NOTE: inline back refs are ordered in the same way that back ref
1360  *       items in the tree are ordered.
1361  */
1362 static noinline_for_stack
1363 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1364                                  struct btrfs_root *root,
1365                                  struct btrfs_path *path,
1366                                  struct btrfs_extent_inline_ref **ref_ret,
1367                                  u64 bytenr, u64 num_bytes,
1368                                  u64 parent, u64 root_objectid,
1369                                  u64 owner, u64 offset, int insert)
1370 {
1371         struct btrfs_key key;
1372         struct extent_buffer *leaf;
1373         struct btrfs_extent_item *ei;
1374         struct btrfs_extent_inline_ref *iref;
1375         u64 flags;
1376         u64 item_size;
1377         unsigned long ptr;
1378         unsigned long end;
1379         int extra_size;
1380         int type;
1381         int want;
1382         int ret;
1383         int err = 0;
1384
1385         key.objectid = bytenr;
1386         key.type = BTRFS_EXTENT_ITEM_KEY;
1387         key.offset = num_bytes;
1388
1389         want = extent_ref_type(parent, owner);
1390         if (insert) {
1391                 extra_size = btrfs_extent_inline_ref_size(want);
1392                 path->keep_locks = 1;
1393         } else
1394                 extra_size = -1;
1395         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1396         if (ret < 0) {
1397                 err = ret;
1398                 goto out;
1399         }
1400         BUG_ON(ret);
1401
1402         leaf = path->nodes[0];
1403         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1404 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1405         if (item_size < sizeof(*ei)) {
1406                 if (!insert) {
1407                         err = -ENOENT;
1408                         goto out;
1409                 }
1410                 ret = convert_extent_item_v0(trans, root, path, owner,
1411                                              extra_size);
1412                 if (ret < 0) {
1413                         err = ret;
1414                         goto out;
1415                 }
1416                 leaf = path->nodes[0];
1417                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1418         }
1419 #endif
1420         BUG_ON(item_size < sizeof(*ei));
1421
1422         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1423         flags = btrfs_extent_flags(leaf, ei);
1424
1425         ptr = (unsigned long)(ei + 1);
1426         end = (unsigned long)ei + item_size;
1427
1428         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1429                 ptr += sizeof(struct btrfs_tree_block_info);
1430                 BUG_ON(ptr > end);
1431         } else {
1432                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1433         }
1434
1435         err = -ENOENT;
1436         while (1) {
1437                 if (ptr >= end) {
1438                         WARN_ON(ptr > end);
1439                         break;
1440                 }
1441                 iref = (struct btrfs_extent_inline_ref *)ptr;
1442                 type = btrfs_extent_inline_ref_type(leaf, iref);
1443                 if (want < type)
1444                         break;
1445                 if (want > type) {
1446                         ptr += btrfs_extent_inline_ref_size(type);
1447                         continue;
1448                 }
1449
1450                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1451                         struct btrfs_extent_data_ref *dref;
1452                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1453                         if (match_extent_data_ref(leaf, dref, root_objectid,
1454                                                   owner, offset)) {
1455                                 err = 0;
1456                                 break;
1457                         }
1458                         if (hash_extent_data_ref_item(leaf, dref) <
1459                             hash_extent_data_ref(root_objectid, owner, offset))
1460                                 break;
1461                 } else {
1462                         u64 ref_offset;
1463                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1464                         if (parent > 0) {
1465                                 if (parent == ref_offset) {
1466                                         err = 0;
1467                                         break;
1468                                 }
1469                                 if (ref_offset < parent)
1470                                         break;
1471                         } else {
1472                                 if (root_objectid == ref_offset) {
1473                                         err = 0;
1474                                         break;
1475                                 }
1476                                 if (ref_offset < root_objectid)
1477                                         break;
1478                         }
1479                 }
1480                 ptr += btrfs_extent_inline_ref_size(type);
1481         }
1482         if (err == -ENOENT && insert) {
1483                 if (item_size + extra_size >=
1484                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1485                         err = -EAGAIN;
1486                         goto out;
1487                 }
1488                 /*
1489                  * To add new inline back ref, we have to make sure
1490                  * there is no corresponding back ref item.
1491                  * For simplicity, we just do not add new inline back
1492                  * ref if there is any kind of item for this block
1493                  */
1494                 if (find_next_key(path, 0, &key) == 0 &&
1495                     key.objectid == bytenr &&
1496                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1497                         err = -EAGAIN;
1498                         goto out;
1499                 }
1500         }
1501         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1502 out:
1503         if (insert) {
1504                 path->keep_locks = 0;
1505                 btrfs_unlock_up_safe(path, 1);
1506         }
1507         return err;
1508 }
1509
1510 /*
1511  * helper to add new inline back ref
1512  */
1513 static noinline_for_stack
1514 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1515                                 struct btrfs_root *root,
1516                                 struct btrfs_path *path,
1517                                 struct btrfs_extent_inline_ref *iref,
1518                                 u64 parent, u64 root_objectid,
1519                                 u64 owner, u64 offset, int refs_to_add,
1520                                 struct btrfs_delayed_extent_op *extent_op)
1521 {
1522         struct extent_buffer *leaf;
1523         struct btrfs_extent_item *ei;
1524         unsigned long ptr;
1525         unsigned long end;
1526         unsigned long item_offset;
1527         u64 refs;
1528         int size;
1529         int type;
1530         int ret;
1531
1532         leaf = path->nodes[0];
1533         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1534         item_offset = (unsigned long)iref - (unsigned long)ei;
1535
1536         type = extent_ref_type(parent, owner);
1537         size = btrfs_extent_inline_ref_size(type);
1538
1539         ret = btrfs_extend_item(trans, root, path, size);
1540         BUG_ON(ret);
1541
1542         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1543         refs = btrfs_extent_refs(leaf, ei);
1544         refs += refs_to_add;
1545         btrfs_set_extent_refs(leaf, ei, refs);
1546         if (extent_op)
1547                 __run_delayed_extent_op(extent_op, leaf, ei);
1548
1549         ptr = (unsigned long)ei + item_offset;
1550         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1551         if (ptr < end - size)
1552                 memmove_extent_buffer(leaf, ptr + size, ptr,
1553                                       end - size - ptr);
1554
1555         iref = (struct btrfs_extent_inline_ref *)ptr;
1556         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1557         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1558                 struct btrfs_extent_data_ref *dref;
1559                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1560                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1561                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1562                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1563                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1564         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1565                 struct btrfs_shared_data_ref *sref;
1566                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1567                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1568                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1569         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1570                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1571         } else {
1572                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1573         }
1574         btrfs_mark_buffer_dirty(leaf);
1575         return 0;
1576 }
1577
1578 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1579                                  struct btrfs_root *root,
1580                                  struct btrfs_path *path,
1581                                  struct btrfs_extent_inline_ref **ref_ret,
1582                                  u64 bytenr, u64 num_bytes, u64 parent,
1583                                  u64 root_objectid, u64 owner, u64 offset)
1584 {
1585         int ret;
1586
1587         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1588                                            bytenr, num_bytes, parent,
1589                                            root_objectid, owner, offset, 0);
1590         if (ret != -ENOENT)
1591                 return ret;
1592
1593         btrfs_release_path(root, path);
1594         *ref_ret = NULL;
1595
1596         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1597                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1598                                             root_objectid);
1599         } else {
1600                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1601                                              root_objectid, owner, offset);
1602         }
1603         return ret;
1604 }
1605
1606 /*
1607  * helper to update/remove inline back ref
1608  */
1609 static noinline_for_stack
1610 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1611                                  struct btrfs_root *root,
1612                                  struct btrfs_path *path,
1613                                  struct btrfs_extent_inline_ref *iref,
1614                                  int refs_to_mod,
1615                                  struct btrfs_delayed_extent_op *extent_op)
1616 {
1617         struct extent_buffer *leaf;
1618         struct btrfs_extent_item *ei;
1619         struct btrfs_extent_data_ref *dref = NULL;
1620         struct btrfs_shared_data_ref *sref = NULL;
1621         unsigned long ptr;
1622         unsigned long end;
1623         u32 item_size;
1624         int size;
1625         int type;
1626         int ret;
1627         u64 refs;
1628
1629         leaf = path->nodes[0];
1630         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1631         refs = btrfs_extent_refs(leaf, ei);
1632         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1633         refs += refs_to_mod;
1634         btrfs_set_extent_refs(leaf, ei, refs);
1635         if (extent_op)
1636                 __run_delayed_extent_op(extent_op, leaf, ei);
1637
1638         type = btrfs_extent_inline_ref_type(leaf, iref);
1639
1640         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1641                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1642                 refs = btrfs_extent_data_ref_count(leaf, dref);
1643         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1644                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1645                 refs = btrfs_shared_data_ref_count(leaf, sref);
1646         } else {
1647                 refs = 1;
1648                 BUG_ON(refs_to_mod != -1);
1649         }
1650
1651         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1652         refs += refs_to_mod;
1653
1654         if (refs > 0) {
1655                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1656                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1657                 else
1658                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1659         } else {
1660                 size =  btrfs_extent_inline_ref_size(type);
1661                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1662                 ptr = (unsigned long)iref;
1663                 end = (unsigned long)ei + item_size;
1664                 if (ptr + size < end)
1665                         memmove_extent_buffer(leaf, ptr, ptr + size,
1666                                               end - ptr - size);
1667                 item_size -= size;
1668                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1669                 BUG_ON(ret);
1670         }
1671         btrfs_mark_buffer_dirty(leaf);
1672         return 0;
1673 }
1674
1675 static noinline_for_stack
1676 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1677                                  struct btrfs_root *root,
1678                                  struct btrfs_path *path,
1679                                  u64 bytenr, u64 num_bytes, u64 parent,
1680                                  u64 root_objectid, u64 owner,
1681                                  u64 offset, int refs_to_add,
1682                                  struct btrfs_delayed_extent_op *extent_op)
1683 {
1684         struct btrfs_extent_inline_ref *iref;
1685         int ret;
1686
1687         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1688                                            bytenr, num_bytes, parent,
1689                                            root_objectid, owner, offset, 1);
1690         if (ret == 0) {
1691                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1692                 ret = update_inline_extent_backref(trans, root, path, iref,
1693                                                    refs_to_add, extent_op);
1694         } else if (ret == -ENOENT) {
1695                 ret = setup_inline_extent_backref(trans, root, path, iref,
1696                                                   parent, root_objectid,
1697                                                   owner, offset, refs_to_add,
1698                                                   extent_op);
1699         }
1700         return ret;
1701 }
1702
1703 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1704                                  struct btrfs_root *root,
1705                                  struct btrfs_path *path,
1706                                  u64 bytenr, u64 parent, u64 root_objectid,
1707                                  u64 owner, u64 offset, int refs_to_add)
1708 {
1709         int ret;
1710         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1711                 BUG_ON(refs_to_add != 1);
1712                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1713                                             parent, root_objectid);
1714         } else {
1715                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1716                                              parent, root_objectid,
1717                                              owner, offset, refs_to_add);
1718         }
1719         return ret;
1720 }
1721
1722 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1723                                  struct btrfs_root *root,
1724                                  struct btrfs_path *path,
1725                                  struct btrfs_extent_inline_ref *iref,
1726                                  int refs_to_drop, int is_data)
1727 {
1728         int ret;
1729
1730         BUG_ON(!is_data && refs_to_drop != 1);
1731         if (iref) {
1732                 ret = update_inline_extent_backref(trans, root, path, iref,
1733                                                    -refs_to_drop, NULL);
1734         } else if (is_data) {
1735                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1736         } else {
1737                 ret = btrfs_del_item(trans, root, path);
1738         }
1739         return ret;
1740 }
1741
1742 static void btrfs_issue_discard(struct block_device *bdev,
1743                                 u64 start, u64 len)
1744 {
1745         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1746                         BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
1747 }
1748
1749 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1750                                 u64 num_bytes)
1751 {
1752         int ret;
1753         u64 map_length = num_bytes;
1754         struct btrfs_multi_bio *multi = NULL;
1755
1756         if (!btrfs_test_opt(root, DISCARD))
1757                 return 0;
1758
1759         /* Tell the block device(s) that the sectors can be discarded */
1760         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1761                               bytenr, &map_length, &multi, 0);
1762         if (!ret) {
1763                 struct btrfs_bio_stripe *stripe = multi->stripes;
1764                 int i;
1765
1766                 if (map_length > num_bytes)
1767                         map_length = num_bytes;
1768
1769                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1770                         btrfs_issue_discard(stripe->dev->bdev,
1771                                             stripe->physical,
1772                                             map_length);
1773                 }
1774                 kfree(multi);
1775         }
1776
1777         return ret;
1778 }
1779
1780 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1781                          struct btrfs_root *root,
1782                          u64 bytenr, u64 num_bytes, u64 parent,
1783                          u64 root_objectid, u64 owner, u64 offset)
1784 {
1785         int ret;
1786         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1787                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1788
1789         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1790                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1791                                         parent, root_objectid, (int)owner,
1792                                         BTRFS_ADD_DELAYED_REF, NULL);
1793         } else {
1794                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1795                                         parent, root_objectid, owner, offset,
1796                                         BTRFS_ADD_DELAYED_REF, NULL);
1797         }
1798         return ret;
1799 }
1800
1801 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1802                                   struct btrfs_root *root,
1803                                   u64 bytenr, u64 num_bytes,
1804                                   u64 parent, u64 root_objectid,
1805                                   u64 owner, u64 offset, int refs_to_add,
1806                                   struct btrfs_delayed_extent_op *extent_op)
1807 {
1808         struct btrfs_path *path;
1809         struct extent_buffer *leaf;
1810         struct btrfs_extent_item *item;
1811         u64 refs;
1812         int ret;
1813         int err = 0;
1814
1815         path = btrfs_alloc_path();
1816         if (!path)
1817                 return -ENOMEM;
1818
1819         path->reada = 1;
1820         path->leave_spinning = 1;
1821         /* this will setup the path even if it fails to insert the back ref */
1822         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1823                                            path, bytenr, num_bytes, parent,
1824                                            root_objectid, owner, offset,
1825                                            refs_to_add, extent_op);
1826         if (ret == 0)
1827                 goto out;
1828
1829         if (ret != -EAGAIN) {
1830                 err = ret;
1831                 goto out;
1832         }
1833
1834         leaf = path->nodes[0];
1835         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1836         refs = btrfs_extent_refs(leaf, item);
1837         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1838         if (extent_op)
1839                 __run_delayed_extent_op(extent_op, leaf, item);
1840
1841         btrfs_mark_buffer_dirty(leaf);
1842         btrfs_release_path(root->fs_info->extent_root, path);
1843
1844         path->reada = 1;
1845         path->leave_spinning = 1;
1846
1847         /* now insert the actual backref */
1848         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1849                                     path, bytenr, parent, root_objectid,
1850                                     owner, offset, refs_to_add);
1851         BUG_ON(ret);
1852 out:
1853         btrfs_free_path(path);
1854         return err;
1855 }
1856
1857 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1858                                 struct btrfs_root *root,
1859                                 struct btrfs_delayed_ref_node *node,
1860                                 struct btrfs_delayed_extent_op *extent_op,
1861                                 int insert_reserved)
1862 {
1863         int ret = 0;
1864         struct btrfs_delayed_data_ref *ref;
1865         struct btrfs_key ins;
1866         u64 parent = 0;
1867         u64 ref_root = 0;
1868         u64 flags = 0;
1869
1870         ins.objectid = node->bytenr;
1871         ins.offset = node->num_bytes;
1872         ins.type = BTRFS_EXTENT_ITEM_KEY;
1873
1874         ref = btrfs_delayed_node_to_data_ref(node);
1875         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1876                 parent = ref->parent;
1877         else
1878                 ref_root = ref->root;
1879
1880         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1881                 if (extent_op) {
1882                         BUG_ON(extent_op->update_key);
1883                         flags |= extent_op->flags_to_set;
1884                 }
1885                 ret = alloc_reserved_file_extent(trans, root,
1886                                                  parent, ref_root, flags,
1887                                                  ref->objectid, ref->offset,
1888                                                  &ins, node->ref_mod);
1889         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1890                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1891                                              node->num_bytes, parent,
1892                                              ref_root, ref->objectid,
1893                                              ref->offset, node->ref_mod,
1894                                              extent_op);
1895         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1896                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1897                                           node->num_bytes, parent,
1898                                           ref_root, ref->objectid,
1899                                           ref->offset, node->ref_mod,
1900                                           extent_op);
1901         } else {
1902                 BUG();
1903         }
1904         return ret;
1905 }
1906
1907 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1908                                     struct extent_buffer *leaf,
1909                                     struct btrfs_extent_item *ei)
1910 {
1911         u64 flags = btrfs_extent_flags(leaf, ei);
1912         if (extent_op->update_flags) {
1913                 flags |= extent_op->flags_to_set;
1914                 btrfs_set_extent_flags(leaf, ei, flags);
1915         }
1916
1917         if (extent_op->update_key) {
1918                 struct btrfs_tree_block_info *bi;
1919                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1920                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1921                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1922         }
1923 }
1924
1925 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1926                                  struct btrfs_root *root,
1927                                  struct btrfs_delayed_ref_node *node,
1928                                  struct btrfs_delayed_extent_op *extent_op)
1929 {
1930         struct btrfs_key key;
1931         struct btrfs_path *path;
1932         struct btrfs_extent_item *ei;
1933         struct extent_buffer *leaf;
1934         u32 item_size;
1935         int ret;
1936         int err = 0;
1937
1938         path = btrfs_alloc_path();
1939         if (!path)
1940                 return -ENOMEM;
1941
1942         key.objectid = node->bytenr;
1943         key.type = BTRFS_EXTENT_ITEM_KEY;
1944         key.offset = node->num_bytes;
1945
1946         path->reada = 1;
1947         path->leave_spinning = 1;
1948         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1949                                 path, 0, 1);
1950         if (ret < 0) {
1951                 err = ret;
1952                 goto out;
1953         }
1954         if (ret > 0) {
1955                 err = -EIO;
1956                 goto out;
1957         }
1958
1959         leaf = path->nodes[0];
1960         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1961 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1962         if (item_size < sizeof(*ei)) {
1963                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1964                                              path, (u64)-1, 0);
1965                 if (ret < 0) {
1966                         err = ret;
1967                         goto out;
1968                 }
1969                 leaf = path->nodes[0];
1970                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1971         }
1972 #endif
1973         BUG_ON(item_size < sizeof(*ei));
1974         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1975         __run_delayed_extent_op(extent_op, leaf, ei);
1976
1977         btrfs_mark_buffer_dirty(leaf);
1978 out:
1979         btrfs_free_path(path);
1980         return err;
1981 }
1982
1983 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1984                                 struct btrfs_root *root,
1985                                 struct btrfs_delayed_ref_node *node,
1986                                 struct btrfs_delayed_extent_op *extent_op,
1987                                 int insert_reserved)
1988 {
1989         int ret = 0;
1990         struct btrfs_delayed_tree_ref *ref;
1991         struct btrfs_key ins;
1992         u64 parent = 0;
1993         u64 ref_root = 0;
1994
1995         ins.objectid = node->bytenr;
1996         ins.offset = node->num_bytes;
1997         ins.type = BTRFS_EXTENT_ITEM_KEY;
1998
1999         ref = btrfs_delayed_node_to_tree_ref(node);
2000         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2001                 parent = ref->parent;
2002         else
2003                 ref_root = ref->root;
2004
2005         BUG_ON(node->ref_mod != 1);
2006         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2007                 BUG_ON(!extent_op || !extent_op->update_flags ||
2008                        !extent_op->update_key);
2009                 ret = alloc_reserved_tree_block(trans, root,
2010                                                 parent, ref_root,
2011                                                 extent_op->flags_to_set,
2012                                                 &extent_op->key,
2013                                                 ref->level, &ins);
2014         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2015                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2016                                              node->num_bytes, parent, ref_root,
2017                                              ref->level, 0, 1, extent_op);
2018         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2019                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2020                                           node->num_bytes, parent, ref_root,
2021                                           ref->level, 0, 1, extent_op);
2022         } else {
2023                 BUG();
2024         }
2025         return ret;
2026 }
2027
2028 /* helper function to actually process a single delayed ref entry */
2029 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2030                                struct btrfs_root *root,
2031                                struct btrfs_delayed_ref_node *node,
2032                                struct btrfs_delayed_extent_op *extent_op,
2033                                int insert_reserved)
2034 {
2035         int ret;
2036         if (btrfs_delayed_ref_is_head(node)) {
2037                 struct btrfs_delayed_ref_head *head;
2038                 /*
2039                  * we've hit the end of the chain and we were supposed
2040                  * to insert this extent into the tree.  But, it got
2041                  * deleted before we ever needed to insert it, so all
2042                  * we have to do is clean up the accounting
2043                  */
2044                 BUG_ON(extent_op);
2045                 head = btrfs_delayed_node_to_head(node);
2046                 if (insert_reserved) {
2047                         btrfs_pin_extent(root, node->bytenr,
2048                                          node->num_bytes, 1);
2049                         if (head->is_data) {
2050                                 ret = btrfs_del_csums(trans, root,
2051                                                       node->bytenr,
2052                                                       node->num_bytes);
2053                                 BUG_ON(ret);
2054                         }
2055                 }
2056                 mutex_unlock(&head->mutex);
2057                 return 0;
2058         }
2059
2060         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2061             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2062                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2063                                            insert_reserved);
2064         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2065                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2066                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2067                                            insert_reserved);
2068         else
2069                 BUG();
2070         return ret;
2071 }
2072
2073 static noinline struct btrfs_delayed_ref_node *
2074 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2075 {
2076         struct rb_node *node;
2077         struct btrfs_delayed_ref_node *ref;
2078         int action = BTRFS_ADD_DELAYED_REF;
2079 again:
2080         /*
2081          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2082          * this prevents ref count from going down to zero when
2083          * there still are pending delayed ref.
2084          */
2085         node = rb_prev(&head->node.rb_node);
2086         while (1) {
2087                 if (!node)
2088                         break;
2089                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2090                                 rb_node);
2091                 if (ref->bytenr != head->node.bytenr)
2092                         break;
2093                 if (ref->action == action)
2094                         return ref;
2095                 node = rb_prev(node);
2096         }
2097         if (action == BTRFS_ADD_DELAYED_REF) {
2098                 action = BTRFS_DROP_DELAYED_REF;
2099                 goto again;
2100         }
2101         return NULL;
2102 }
2103
2104 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2105                                        struct btrfs_root *root,
2106                                        struct list_head *cluster)
2107 {
2108         struct btrfs_delayed_ref_root *delayed_refs;
2109         struct btrfs_delayed_ref_node *ref;
2110         struct btrfs_delayed_ref_head *locked_ref = NULL;
2111         struct btrfs_delayed_extent_op *extent_op;
2112         int ret;
2113         int count = 0;
2114         int must_insert_reserved = 0;
2115
2116         delayed_refs = &trans->transaction->delayed_refs;
2117         while (1) {
2118                 if (!locked_ref) {
2119                         /* pick a new head ref from the cluster list */
2120                         if (list_empty(cluster))
2121                                 break;
2122
2123                         locked_ref = list_entry(cluster->next,
2124                                      struct btrfs_delayed_ref_head, cluster);
2125
2126                         /* grab the lock that says we are going to process
2127                          * all the refs for this head */
2128                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2129
2130                         /*
2131                          * we may have dropped the spin lock to get the head
2132                          * mutex lock, and that might have given someone else
2133                          * time to free the head.  If that's true, it has been
2134                          * removed from our list and we can move on.
2135                          */
2136                         if (ret == -EAGAIN) {
2137                                 locked_ref = NULL;
2138                                 count++;
2139                                 continue;
2140                         }
2141                 }
2142
2143                 /*
2144                  * record the must insert reserved flag before we
2145                  * drop the spin lock.
2146                  */
2147                 must_insert_reserved = locked_ref->must_insert_reserved;
2148                 locked_ref->must_insert_reserved = 0;
2149
2150                 extent_op = locked_ref->extent_op;
2151                 locked_ref->extent_op = NULL;
2152
2153                 /*
2154                  * locked_ref is the head node, so we have to go one
2155                  * node back for any delayed ref updates
2156                  */
2157                 ref = select_delayed_ref(locked_ref);
2158                 if (!ref) {
2159                         /* All delayed refs have been processed, Go ahead
2160                          * and send the head node to run_one_delayed_ref,
2161                          * so that any accounting fixes can happen
2162                          */
2163                         ref = &locked_ref->node;
2164
2165                         if (extent_op && must_insert_reserved) {
2166                                 kfree(extent_op);
2167                                 extent_op = NULL;
2168                         }
2169
2170                         if (extent_op) {
2171                                 spin_unlock(&delayed_refs->lock);
2172
2173                                 ret = run_delayed_extent_op(trans, root,
2174                                                             ref, extent_op);
2175                                 BUG_ON(ret);
2176                                 kfree(extent_op);
2177
2178                                 cond_resched();
2179                                 spin_lock(&delayed_refs->lock);
2180                                 continue;
2181                         }
2182
2183                         list_del_init(&locked_ref->cluster);
2184                         locked_ref = NULL;
2185                 }
2186
2187                 ref->in_tree = 0;
2188                 rb_erase(&ref->rb_node, &delayed_refs->root);
2189                 delayed_refs->num_entries--;
2190
2191                 spin_unlock(&delayed_refs->lock);
2192
2193                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2194                                           must_insert_reserved);
2195                 BUG_ON(ret);
2196
2197                 btrfs_put_delayed_ref(ref);
2198                 kfree(extent_op);
2199                 count++;
2200
2201                 cond_resched();
2202                 spin_lock(&delayed_refs->lock);
2203         }
2204         return count;
2205 }
2206
2207 /*
2208  * this starts processing the delayed reference count updates and
2209  * extent insertions we have queued up so far.  count can be
2210  * 0, which means to process everything in the tree at the start
2211  * of the run (but not newly added entries), or it can be some target
2212  * number you'd like to process.
2213  */
2214 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2215                            struct btrfs_root *root, unsigned long count)
2216 {
2217         struct rb_node *node;
2218         struct btrfs_delayed_ref_root *delayed_refs;
2219         struct btrfs_delayed_ref_node *ref;
2220         struct list_head cluster;
2221         int ret;
2222         int run_all = count == (unsigned long)-1;
2223         int run_most = 0;
2224
2225         if (root == root->fs_info->extent_root)
2226                 root = root->fs_info->tree_root;
2227
2228         delayed_refs = &trans->transaction->delayed_refs;
2229         INIT_LIST_HEAD(&cluster);
2230 again:
2231         spin_lock(&delayed_refs->lock);
2232         if (count == 0) {
2233                 count = delayed_refs->num_entries * 2;
2234                 run_most = 1;
2235         }
2236         while (1) {
2237                 if (!(run_all || run_most) &&
2238                     delayed_refs->num_heads_ready < 64)
2239                         break;
2240
2241                 /*
2242                  * go find something we can process in the rbtree.  We start at
2243                  * the beginning of the tree, and then build a cluster
2244                  * of refs to process starting at the first one we are able to
2245                  * lock
2246                  */
2247                 ret = btrfs_find_ref_cluster(trans, &cluster,
2248                                              delayed_refs->run_delayed_start);
2249                 if (ret)
2250                         break;
2251
2252                 ret = run_clustered_refs(trans, root, &cluster);
2253                 BUG_ON(ret < 0);
2254
2255                 count -= min_t(unsigned long, ret, count);
2256
2257                 if (count == 0)
2258                         break;
2259         }
2260
2261         if (run_all) {
2262                 node = rb_first(&delayed_refs->root);
2263                 if (!node)
2264                         goto out;
2265                 count = (unsigned long)-1;
2266
2267                 while (node) {
2268                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2269                                        rb_node);
2270                         if (btrfs_delayed_ref_is_head(ref)) {
2271                                 struct btrfs_delayed_ref_head *head;
2272
2273                                 head = btrfs_delayed_node_to_head(ref);
2274                                 atomic_inc(&ref->refs);
2275
2276                                 spin_unlock(&delayed_refs->lock);
2277                                 mutex_lock(&head->mutex);
2278                                 mutex_unlock(&head->mutex);
2279
2280                                 btrfs_put_delayed_ref(ref);
2281                                 cond_resched();
2282                                 goto again;
2283                         }
2284                         node = rb_next(node);
2285                 }
2286                 spin_unlock(&delayed_refs->lock);
2287                 schedule_timeout(1);
2288                 goto again;
2289         }
2290 out:
2291         spin_unlock(&delayed_refs->lock);
2292         return 0;
2293 }
2294
2295 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2296                                 struct btrfs_root *root,
2297                                 u64 bytenr, u64 num_bytes, u64 flags,
2298                                 int is_data)
2299 {
2300         struct btrfs_delayed_extent_op *extent_op;
2301         int ret;
2302
2303         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2304         if (!extent_op)
2305                 return -ENOMEM;
2306
2307         extent_op->flags_to_set = flags;
2308         extent_op->update_flags = 1;
2309         extent_op->update_key = 0;
2310         extent_op->is_data = is_data ? 1 : 0;
2311
2312         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2313         if (ret)
2314                 kfree(extent_op);
2315         return ret;
2316 }
2317
2318 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2319                                       struct btrfs_root *root,
2320                                       struct btrfs_path *path,
2321                                       u64 objectid, u64 offset, u64 bytenr)
2322 {
2323         struct btrfs_delayed_ref_head *head;
2324         struct btrfs_delayed_ref_node *ref;
2325         struct btrfs_delayed_data_ref *data_ref;
2326         struct btrfs_delayed_ref_root *delayed_refs;
2327         struct rb_node *node;
2328         int ret = 0;
2329
2330         ret = -ENOENT;
2331         delayed_refs = &trans->transaction->delayed_refs;
2332         spin_lock(&delayed_refs->lock);
2333         head = btrfs_find_delayed_ref_head(trans, bytenr);
2334         if (!head)
2335                 goto out;
2336
2337         if (!mutex_trylock(&head->mutex)) {
2338                 atomic_inc(&head->node.refs);
2339                 spin_unlock(&delayed_refs->lock);
2340
2341                 btrfs_release_path(root->fs_info->extent_root, path);
2342
2343                 mutex_lock(&head->mutex);
2344                 mutex_unlock(&head->mutex);
2345                 btrfs_put_delayed_ref(&head->node);
2346                 return -EAGAIN;
2347         }
2348
2349         node = rb_prev(&head->node.rb_node);
2350         if (!node)
2351                 goto out_unlock;
2352
2353         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2354
2355         if (ref->bytenr != bytenr)
2356                 goto out_unlock;
2357
2358         ret = 1;
2359         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2360                 goto out_unlock;
2361
2362         data_ref = btrfs_delayed_node_to_data_ref(ref);
2363
2364         node = rb_prev(node);
2365         if (node) {
2366                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2367                 if (ref->bytenr == bytenr)
2368                         goto out_unlock;
2369         }
2370
2371         if (data_ref->root != root->root_key.objectid ||
2372             data_ref->objectid != objectid || data_ref->offset != offset)
2373                 goto out_unlock;
2374
2375         ret = 0;
2376 out_unlock:
2377         mutex_unlock(&head->mutex);
2378 out:
2379         spin_unlock(&delayed_refs->lock);
2380         return ret;
2381 }
2382
2383 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2384                                         struct btrfs_root *root,
2385                                         struct btrfs_path *path,
2386                                         u64 objectid, u64 offset, u64 bytenr)
2387 {
2388         struct btrfs_root *extent_root = root->fs_info->extent_root;
2389         struct extent_buffer *leaf;
2390         struct btrfs_extent_data_ref *ref;
2391         struct btrfs_extent_inline_ref *iref;
2392         struct btrfs_extent_item *ei;
2393         struct btrfs_key key;
2394         u32 item_size;
2395         int ret;
2396
2397         key.objectid = bytenr;
2398         key.offset = (u64)-1;
2399         key.type = BTRFS_EXTENT_ITEM_KEY;
2400
2401         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2402         if (ret < 0)
2403                 goto out;
2404         BUG_ON(ret == 0);
2405
2406         ret = -ENOENT;
2407         if (path->slots[0] == 0)
2408                 goto out;
2409
2410         path->slots[0]--;
2411         leaf = path->nodes[0];
2412         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2413
2414         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2415                 goto out;
2416
2417         ret = 1;
2418         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2419 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2420         if (item_size < sizeof(*ei)) {
2421                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2422                 goto out;
2423         }
2424 #endif
2425         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2426
2427         if (item_size != sizeof(*ei) +
2428             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2429                 goto out;
2430
2431         if (btrfs_extent_generation(leaf, ei) <=
2432             btrfs_root_last_snapshot(&root->root_item))
2433                 goto out;
2434
2435         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2436         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2437             BTRFS_EXTENT_DATA_REF_KEY)
2438                 goto out;
2439
2440         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2441         if (btrfs_extent_refs(leaf, ei) !=
2442             btrfs_extent_data_ref_count(leaf, ref) ||
2443             btrfs_extent_data_ref_root(leaf, ref) !=
2444             root->root_key.objectid ||
2445             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2446             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2447                 goto out;
2448
2449         ret = 0;
2450 out:
2451         return ret;
2452 }
2453
2454 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2455                           struct btrfs_root *root,
2456                           u64 objectid, u64 offset, u64 bytenr)
2457 {
2458         struct btrfs_path *path;
2459         int ret;
2460         int ret2;
2461
2462         path = btrfs_alloc_path();
2463         if (!path)
2464                 return -ENOENT;
2465
2466         do {
2467                 ret = check_committed_ref(trans, root, path, objectid,
2468                                           offset, bytenr);
2469                 if (ret && ret != -ENOENT)
2470                         goto out;
2471
2472                 ret2 = check_delayed_ref(trans, root, path, objectid,
2473                                          offset, bytenr);
2474         } while (ret2 == -EAGAIN);
2475
2476         if (ret2 && ret2 != -ENOENT) {
2477                 ret = ret2;
2478                 goto out;
2479         }
2480
2481         if (ret != -ENOENT || ret2 != -ENOENT)
2482                 ret = 0;
2483 out:
2484         btrfs_free_path(path);
2485         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2486                 WARN_ON(ret > 0);
2487         return ret;
2488 }
2489
2490 #if 0
2491 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2492                     struct extent_buffer *buf, u32 nr_extents)
2493 {
2494         struct btrfs_key key;
2495         struct btrfs_file_extent_item *fi;
2496         u64 root_gen;
2497         u32 nritems;
2498         int i;
2499         int level;
2500         int ret = 0;
2501         int shared = 0;
2502
2503         if (!root->ref_cows)
2504                 return 0;
2505
2506         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2507                 shared = 0;
2508                 root_gen = root->root_key.offset;
2509         } else {
2510                 shared = 1;
2511                 root_gen = trans->transid - 1;
2512         }
2513
2514         level = btrfs_header_level(buf);
2515         nritems = btrfs_header_nritems(buf);
2516
2517         if (level == 0) {
2518                 struct btrfs_leaf_ref *ref;
2519                 struct btrfs_extent_info *info;
2520
2521                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2522                 if (!ref) {
2523                         ret = -ENOMEM;
2524                         goto out;
2525                 }
2526
2527                 ref->root_gen = root_gen;
2528                 ref->bytenr = buf->start;
2529                 ref->owner = btrfs_header_owner(buf);
2530                 ref->generation = btrfs_header_generation(buf);
2531                 ref->nritems = nr_extents;
2532                 info = ref->extents;
2533
2534                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2535                         u64 disk_bytenr;
2536                         btrfs_item_key_to_cpu(buf, &key, i);
2537                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2538                                 continue;
2539                         fi = btrfs_item_ptr(buf, i,
2540                                             struct btrfs_file_extent_item);
2541                         if (btrfs_file_extent_type(buf, fi) ==
2542                             BTRFS_FILE_EXTENT_INLINE)
2543                                 continue;
2544                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2545                         if (disk_bytenr == 0)
2546                                 continue;
2547
2548                         info->bytenr = disk_bytenr;
2549                         info->num_bytes =
2550                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2551                         info->objectid = key.objectid;
2552                         info->offset = key.offset;
2553                         info++;
2554                 }
2555
2556                 ret = btrfs_add_leaf_ref(root, ref, shared);
2557                 if (ret == -EEXIST && shared) {
2558                         struct btrfs_leaf_ref *old;
2559                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2560                         BUG_ON(!old);
2561                         btrfs_remove_leaf_ref(root, old);
2562                         btrfs_free_leaf_ref(root, old);
2563                         ret = btrfs_add_leaf_ref(root, ref, shared);
2564                 }
2565                 WARN_ON(ret);
2566                 btrfs_free_leaf_ref(root, ref);
2567         }
2568 out:
2569         return ret;
2570 }
2571
2572 /* when a block goes through cow, we update the reference counts of
2573  * everything that block points to.  The internal pointers of the block
2574  * can be in just about any order, and it is likely to have clusters of
2575  * things that are close together and clusters of things that are not.
2576  *
2577  * To help reduce the seeks that come with updating all of these reference
2578  * counts, sort them by byte number before actual updates are done.
2579  *
2580  * struct refsort is used to match byte number to slot in the btree block.
2581  * we sort based on the byte number and then use the slot to actually
2582  * find the item.
2583  *
2584  * struct refsort is smaller than strcut btrfs_item and smaller than
2585  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2586  * for a btree block, there's no way for a kmalloc of refsorts for a
2587  * single node to be bigger than a page.
2588  */
2589 struct refsort {
2590         u64 bytenr;
2591         u32 slot;
2592 };
2593
2594 /*
2595  * for passing into sort()
2596  */
2597 static int refsort_cmp(const void *a_void, const void *b_void)
2598 {
2599         const struct refsort *a = a_void;
2600         const struct refsort *b = b_void;
2601
2602         if (a->bytenr < b->bytenr)
2603                 return -1;
2604         if (a->bytenr > b->bytenr)
2605                 return 1;
2606         return 0;
2607 }
2608 #endif
2609
2610 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2611                            struct btrfs_root *root,
2612                            struct extent_buffer *buf,
2613                            int full_backref, int inc)
2614 {
2615         u64 bytenr;
2616         u64 num_bytes;
2617         u64 parent;
2618         u64 ref_root;
2619         u32 nritems;
2620         struct btrfs_key key;
2621         struct btrfs_file_extent_item *fi;
2622         int i;
2623         int level;
2624         int ret = 0;
2625         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2626                             u64, u64, u64, u64, u64, u64);
2627
2628         ref_root = btrfs_header_owner(buf);
2629         nritems = btrfs_header_nritems(buf);
2630         level = btrfs_header_level(buf);
2631
2632         if (!root->ref_cows && level == 0)
2633                 return 0;
2634
2635         if (inc)
2636                 process_func = btrfs_inc_extent_ref;
2637         else
2638                 process_func = btrfs_free_extent;
2639
2640         if (full_backref)
2641                 parent = buf->start;
2642         else
2643                 parent = 0;
2644
2645         for (i = 0; i < nritems; i++) {
2646                 if (level == 0) {
2647                         btrfs_item_key_to_cpu(buf, &key, i);
2648                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2649                                 continue;
2650                         fi = btrfs_item_ptr(buf, i,
2651                                             struct btrfs_file_extent_item);
2652                         if (btrfs_file_extent_type(buf, fi) ==
2653                             BTRFS_FILE_EXTENT_INLINE)
2654                                 continue;
2655                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2656                         if (bytenr == 0)
2657                                 continue;
2658
2659                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2660                         key.offset -= btrfs_file_extent_offset(buf, fi);
2661                         ret = process_func(trans, root, bytenr, num_bytes,
2662                                            parent, ref_root, key.objectid,
2663                                            key.offset);
2664                         if (ret)
2665                                 goto fail;
2666                 } else {
2667                         bytenr = btrfs_node_blockptr(buf, i);
2668                         num_bytes = btrfs_level_size(root, level - 1);
2669                         ret = process_func(trans, root, bytenr, num_bytes,
2670                                            parent, ref_root, level - 1, 0);
2671                         if (ret)
2672                                 goto fail;
2673                 }
2674         }
2675         return 0;
2676 fail:
2677         BUG();
2678         return ret;
2679 }
2680
2681 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2682                   struct extent_buffer *buf, int full_backref)
2683 {
2684         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2685 }
2686
2687 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2688                   struct extent_buffer *buf, int full_backref)
2689 {
2690         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2691 }
2692
2693 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2694                                  struct btrfs_root *root,
2695                                  struct btrfs_path *path,
2696                                  struct btrfs_block_group_cache *cache)
2697 {
2698         int ret;
2699         struct btrfs_root *extent_root = root->fs_info->extent_root;
2700         unsigned long bi;
2701         struct extent_buffer *leaf;
2702
2703         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2704         if (ret < 0)
2705                 goto fail;
2706         BUG_ON(ret);
2707
2708         leaf = path->nodes[0];
2709         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2710         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2711         btrfs_mark_buffer_dirty(leaf);
2712         btrfs_release_path(extent_root, path);
2713 fail:
2714         if (ret)
2715                 return ret;
2716         return 0;
2717
2718 }
2719
2720 static struct btrfs_block_group_cache *
2721 next_block_group(struct btrfs_root *root,
2722                  struct btrfs_block_group_cache *cache)
2723 {
2724         struct rb_node *node;
2725         spin_lock(&root->fs_info->block_group_cache_lock);
2726         node = rb_next(&cache->cache_node);
2727         btrfs_put_block_group(cache);
2728         if (node) {
2729                 cache = rb_entry(node, struct btrfs_block_group_cache,
2730                                  cache_node);
2731                 btrfs_get_block_group(cache);
2732         } else
2733                 cache = NULL;
2734         spin_unlock(&root->fs_info->block_group_cache_lock);
2735         return cache;
2736 }
2737
2738 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2739                             struct btrfs_trans_handle *trans,
2740                             struct btrfs_path *path)
2741 {
2742         struct btrfs_root *root = block_group->fs_info->tree_root;
2743         struct inode *inode = NULL;
2744         u64 alloc_hint = 0;
2745         int dcs = BTRFS_DC_ERROR;
2746         int num_pages = 0;
2747         int retries = 0;
2748         int ret = 0;
2749
2750         /*
2751          * If this block group is smaller than 100 megs don't bother caching the
2752          * block group.
2753          */
2754         if (block_group->key.offset < (100 * 1024 * 1024)) {
2755                 spin_lock(&block_group->lock);
2756                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2757                 spin_unlock(&block_group->lock);
2758                 return 0;
2759         }
2760
2761 again:
2762         inode = lookup_free_space_inode(root, block_group, path);
2763         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2764                 ret = PTR_ERR(inode);
2765                 btrfs_release_path(root, path);
2766                 goto out;
2767         }
2768
2769         if (IS_ERR(inode)) {
2770                 BUG_ON(retries);
2771                 retries++;
2772
2773                 if (block_group->ro)
2774                         goto out_free;
2775
2776                 ret = create_free_space_inode(root, trans, block_group, path);
2777                 if (ret)
2778                         goto out_free;
2779                 goto again;
2780         }
2781
2782         /*
2783          * We want to set the generation to 0, that way if anything goes wrong
2784          * from here on out we know not to trust this cache when we load up next
2785          * time.
2786          */
2787         BTRFS_I(inode)->generation = 0;
2788         ret = btrfs_update_inode(trans, root, inode);
2789         WARN_ON(ret);
2790
2791         if (i_size_read(inode) > 0) {
2792                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2793                                                       inode);
2794                 if (ret)
2795                         goto out_put;
2796         }
2797
2798         spin_lock(&block_group->lock);
2799         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2800                 /* We're not cached, don't bother trying to write stuff out */
2801                 dcs = BTRFS_DC_WRITTEN;
2802                 spin_unlock(&block_group->lock);
2803                 goto out_put;
2804         }
2805         spin_unlock(&block_group->lock);
2806
2807         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2808         if (!num_pages)
2809                 num_pages = 1;
2810
2811         /*
2812          * Just to make absolutely sure we have enough space, we're going to
2813          * preallocate 12 pages worth of space for each block group.  In
2814          * practice we ought to use at most 8, but we need extra space so we can
2815          * add our header and have a terminator between the extents and the
2816          * bitmaps.
2817          */
2818         num_pages *= 16;
2819         num_pages *= PAGE_CACHE_SIZE;
2820
2821         ret = btrfs_check_data_free_space(inode, num_pages);
2822         if (ret)
2823                 goto out_put;
2824
2825         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2826                                               num_pages, num_pages,
2827                                               &alloc_hint);
2828         if (!ret)
2829                 dcs = BTRFS_DC_SETUP;
2830         btrfs_free_reserved_data_space(inode, num_pages);
2831 out_put:
2832         iput(inode);
2833 out_free:
2834         btrfs_release_path(root, path);
2835 out:
2836         spin_lock(&block_group->lock);
2837         block_group->disk_cache_state = dcs;
2838         spin_unlock(&block_group->lock);
2839
2840         return ret;
2841 }
2842
2843 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2844                                    struct btrfs_root *root)
2845 {
2846         struct btrfs_block_group_cache *cache;
2847         int err = 0;
2848         struct btrfs_path *path;
2849         u64 last = 0;
2850
2851         path = btrfs_alloc_path();
2852         if (!path)
2853                 return -ENOMEM;
2854
2855 again:
2856         while (1) {
2857                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2858                 while (cache) {
2859                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2860                                 break;
2861                         cache = next_block_group(root, cache);
2862                 }
2863                 if (!cache) {
2864                         if (last == 0)
2865                                 break;
2866                         last = 0;
2867                         continue;
2868                 }
2869                 err = cache_save_setup(cache, trans, path);
2870                 last = cache->key.objectid + cache->key.offset;
2871                 btrfs_put_block_group(cache);
2872         }
2873
2874         while (1) {
2875                 if (last == 0) {
2876                         err = btrfs_run_delayed_refs(trans, root,
2877                                                      (unsigned long)-1);
2878                         BUG_ON(err);
2879                 }
2880
2881                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2882                 while (cache) {
2883                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2884                                 btrfs_put_block_group(cache);
2885                                 goto again;
2886                         }
2887
2888                         if (cache->dirty)
2889                                 break;
2890                         cache = next_block_group(root, cache);
2891                 }
2892                 if (!cache) {
2893                         if (last == 0)
2894                                 break;
2895                         last = 0;
2896                         continue;
2897                 }
2898
2899                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2900                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2901                 cache->dirty = 0;
2902                 last = cache->key.objectid + cache->key.offset;
2903
2904                 err = write_one_cache_group(trans, root, path, cache);
2905                 BUG_ON(err);
2906                 btrfs_put_block_group(cache);
2907         }
2908
2909         while (1) {
2910                 /*
2911                  * I don't think this is needed since we're just marking our
2912                  * preallocated extent as written, but just in case it can't
2913                  * hurt.
2914                  */
2915                 if (last == 0) {
2916                         err = btrfs_run_delayed_refs(trans, root,
2917                                                      (unsigned long)-1);
2918                         BUG_ON(err);
2919                 }
2920
2921                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2922                 while (cache) {
2923                         /*
2924                          * Really this shouldn't happen, but it could if we
2925                          * couldn't write the entire preallocated extent and
2926                          * splitting the extent resulted in a new block.
2927                          */
2928                         if (cache->dirty) {
2929                                 btrfs_put_block_group(cache);
2930                                 goto again;
2931                         }
2932                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2933                                 break;
2934                         cache = next_block_group(root, cache);
2935                 }
2936                 if (!cache) {
2937                         if (last == 0)
2938                                 break;
2939                         last = 0;
2940                         continue;
2941                 }
2942
2943                 btrfs_write_out_cache(root, trans, cache, path);
2944
2945                 /*
2946                  * If we didn't have an error then the cache state is still
2947                  * NEED_WRITE, so we can set it to WRITTEN.
2948                  */
2949                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2950                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2951                 last = cache->key.objectid + cache->key.offset;
2952                 btrfs_put_block_group(cache);
2953         }
2954
2955         btrfs_free_path(path);
2956         return 0;
2957 }
2958
2959 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2960 {
2961         struct btrfs_block_group_cache *block_group;
2962         int readonly = 0;
2963
2964         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2965         if (!block_group || block_group->ro)
2966                 readonly = 1;
2967         if (block_group)
2968                 btrfs_put_block_group(block_group);
2969         return readonly;
2970 }
2971
2972 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2973                              u64 total_bytes, u64 bytes_used,
2974                              struct btrfs_space_info **space_info)
2975 {
2976         struct btrfs_space_info *found;
2977         int i;
2978         int factor;
2979
2980         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2981                      BTRFS_BLOCK_GROUP_RAID10))
2982                 factor = 2;
2983         else
2984                 factor = 1;
2985
2986         found = __find_space_info(info, flags);
2987         if (found) {
2988                 spin_lock(&found->lock);
2989                 found->total_bytes += total_bytes;
2990                 found->disk_total += total_bytes * factor;
2991                 found->bytes_used += bytes_used;
2992                 found->disk_used += bytes_used * factor;
2993                 found->full = 0;
2994                 spin_unlock(&found->lock);
2995                 *space_info = found;
2996                 return 0;
2997         }
2998         found = kzalloc(sizeof(*found), GFP_NOFS);
2999         if (!found)
3000                 return -ENOMEM;
3001
3002         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3003                 INIT_LIST_HEAD(&found->block_groups[i]);
3004         init_rwsem(&found->groups_sem);
3005         spin_lock_init(&found->lock);
3006         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3007                                 BTRFS_BLOCK_GROUP_SYSTEM |
3008                                 BTRFS_BLOCK_GROUP_METADATA);
3009         found->total_bytes = total_bytes;
3010         found->disk_total = total_bytes * factor;
3011         found->bytes_used = bytes_used;
3012         found->disk_used = bytes_used * factor;
3013         found->bytes_pinned = 0;
3014         found->bytes_reserved = 0;
3015         found->bytes_readonly = 0;
3016         found->bytes_may_use = 0;
3017         found->full = 0;
3018         found->force_alloc = 0;
3019         *space_info = found;
3020         list_add_rcu(&found->list, &info->space_info);
3021         atomic_set(&found->caching_threads, 0);
3022         return 0;
3023 }
3024
3025 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3026 {
3027         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3028                                    BTRFS_BLOCK_GROUP_RAID1 |
3029                                    BTRFS_BLOCK_GROUP_RAID10 |
3030                                    BTRFS_BLOCK_GROUP_DUP);
3031         if (extra_flags) {
3032                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3033                         fs_info->avail_data_alloc_bits |= extra_flags;
3034                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3035                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3036                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3037                         fs_info->avail_system_alloc_bits |= extra_flags;
3038         }
3039 }
3040
3041 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3042 {
3043         u64 num_devices = root->fs_info->fs_devices->rw_devices;
3044
3045         if (num_devices == 1)
3046                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3047         if (num_devices < 4)
3048                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3049
3050         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3051             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3052                       BTRFS_BLOCK_GROUP_RAID10))) {
3053                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3054         }
3055
3056         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3057             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3058                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3059         }
3060
3061         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3062             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3063              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3064              (flags & BTRFS_BLOCK_GROUP_DUP)))
3065                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3066         return flags;
3067 }
3068
3069 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3070 {
3071         if (flags & BTRFS_BLOCK_GROUP_DATA)
3072                 flags |= root->fs_info->avail_data_alloc_bits &
3073                          root->fs_info->data_alloc_profile;
3074         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3075                 flags |= root->fs_info->avail_system_alloc_bits &
3076                          root->fs_info->system_alloc_profile;
3077         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3078                 flags |= root->fs_info->avail_metadata_alloc_bits &
3079                          root->fs_info->metadata_alloc_profile;
3080         return btrfs_reduce_alloc_profile(root, flags);
3081 }
3082
3083 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3084 {
3085         u64 flags;
3086
3087         if (data)
3088                 flags = BTRFS_BLOCK_GROUP_DATA;
3089         else if (root == root->fs_info->chunk_root)
3090                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3091         else
3092                 flags = BTRFS_BLOCK_GROUP_METADATA;
3093
3094         return get_alloc_profile(root, flags);
3095 }
3096
3097 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3098 {
3099         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3100                                                        BTRFS_BLOCK_GROUP_DATA);
3101 }
3102
3103 /*
3104  * This will check the space that the inode allocates from to make sure we have
3105  * enough space for bytes.
3106  */
3107 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3108 {
3109         struct btrfs_space_info *data_sinfo;
3110         struct btrfs_root *root = BTRFS_I(inode)->root;
3111         u64 used;
3112         int ret = 0, committed = 0, alloc_chunk = 1;
3113
3114         /* make sure bytes are sectorsize aligned */
3115         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3116
3117         if (root == root->fs_info->tree_root) {
3118                 alloc_chunk = 0;
3119                 committed = 1;
3120         }
3121
3122         data_sinfo = BTRFS_I(inode)->space_info;
3123         if (!data_sinfo)
3124                 goto alloc;
3125
3126 again:
3127         /* make sure we have enough space to handle the data first */
3128         spin_lock(&data_sinfo->lock);
3129         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3130                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3131                 data_sinfo->bytes_may_use;
3132
3133         if (used + bytes > data_sinfo->total_bytes) {
3134                 struct btrfs_trans_handle *trans;
3135
3136                 /*
3137                  * if we don't have enough free bytes in this space then we need
3138                  * to alloc a new chunk.
3139                  */
3140                 if (!data_sinfo->full && alloc_chunk) {
3141                         u64 alloc_target;
3142
3143                         data_sinfo->force_alloc = 1;
3144                         spin_unlock(&data_sinfo->lock);
3145 alloc:
3146                         alloc_target = btrfs_get_alloc_profile(root, 1);
3147                         trans = btrfs_join_transaction(root, 1);
3148                         if (IS_ERR(trans))
3149                                 return PTR_ERR(trans);
3150
3151                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3152                                              bytes + 2 * 1024 * 1024,
3153                                              alloc_target, 0);
3154                         btrfs_end_transaction(trans, root);
3155                         if (ret < 0)
3156                                 return ret;
3157
3158                         if (!data_sinfo) {
3159                                 btrfs_set_inode_space_info(root, inode);
3160                                 data_sinfo = BTRFS_I(inode)->space_info;
3161                         }
3162                         goto again;
3163                 }
3164                 spin_unlock(&data_sinfo->lock);
3165
3166                 /* commit the current transaction and try again */
3167                 if (!committed && !root->fs_info->open_ioctl_trans) {
3168                         committed = 1;
3169                         trans = btrfs_join_transaction(root, 1);
3170                         if (IS_ERR(trans))
3171                                 return PTR_ERR(trans);
3172                         ret = btrfs_commit_transaction(trans, root);
3173                         if (ret)
3174                                 return ret;
3175                         goto again;
3176                 }
3177
3178 #if 0 /* I hope we never need this code again, just in case */
3179                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3180                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3181                        "%llu bytes_readonly, %llu may use %llu total\n",
3182                        (unsigned long long)bytes,
3183                        (unsigned long long)data_sinfo->bytes_used,
3184                        (unsigned long long)data_sinfo->bytes_reserved,
3185                        (unsigned long long)data_sinfo->bytes_pinned,
3186                        (unsigned long long)data_sinfo->bytes_readonly,
3187                        (unsigned long long)data_sinfo->bytes_may_use,
3188                        (unsigned long long)data_sinfo->total_bytes);
3189 #endif
3190                 return -ENOSPC;
3191         }
3192         data_sinfo->bytes_may_use += bytes;
3193         BTRFS_I(inode)->reserved_bytes += bytes;
3194         spin_unlock(&data_sinfo->lock);
3195
3196         return 0;
3197 }
3198
3199 /*
3200  * called when we are clearing an delalloc extent from the
3201  * inode's io_tree or there was an error for whatever reason
3202  * after calling btrfs_check_data_free_space
3203  */
3204 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3205 {
3206         struct btrfs_root *root = BTRFS_I(inode)->root;
3207         struct btrfs_space_info *data_sinfo;
3208
3209         /* make sure bytes are sectorsize aligned */
3210         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3211
3212         data_sinfo = BTRFS_I(inode)->space_info;
3213         spin_lock(&data_sinfo->lock);
3214         data_sinfo->bytes_may_use -= bytes;
3215         BTRFS_I(inode)->reserved_bytes -= bytes;
3216         spin_unlock(&data_sinfo->lock);
3217 }
3218
3219 static void force_metadata_allocation(struct btrfs_fs_info *info)
3220 {
3221         struct list_head *head = &info->space_info;
3222         struct btrfs_space_info *found;
3223
3224         rcu_read_lock();
3225         list_for_each_entry_rcu(found, head, list) {
3226                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3227                         found->force_alloc = 1;
3228         }
3229         rcu_read_unlock();
3230 }
3231
3232 static int should_alloc_chunk(struct btrfs_root *root,
3233                               struct btrfs_space_info *sinfo, u64 alloc_bytes)
3234 {
3235         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3236         u64 thresh;
3237
3238         if (sinfo->bytes_used + sinfo->bytes_reserved +
3239             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3240                 return 0;
3241
3242         if (sinfo->bytes_used + sinfo->bytes_reserved +
3243             alloc_bytes < div_factor(num_bytes, 8))
3244                 return 0;
3245
3246         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3247         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3248
3249         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3250                 return 0;
3251
3252         return 1;
3253 }
3254
3255 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3256                           struct btrfs_root *extent_root, u64 alloc_bytes,
3257                           u64 flags, int force)
3258 {
3259         struct btrfs_space_info *space_info;
3260         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3261         int ret = 0;
3262
3263         mutex_lock(&fs_info->chunk_mutex);
3264
3265         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3266
3267         space_info = __find_space_info(extent_root->fs_info, flags);
3268         if (!space_info) {
3269                 ret = update_space_info(extent_root->fs_info, flags,
3270                                         0, 0, &space_info);
3271                 BUG_ON(ret);
3272         }
3273         BUG_ON(!space_info);
3274
3275         spin_lock(&space_info->lock);
3276         if (space_info->force_alloc)
3277                 force = 1;
3278         if (space_info->full) {
3279                 spin_unlock(&space_info->lock);
3280                 goto out;
3281         }
3282
3283         if (!force && !should_alloc_chunk(extent_root, space_info,
3284                                           alloc_bytes)) {
3285                 spin_unlock(&space_info->lock);
3286                 goto out;
3287         }
3288         spin_unlock(&space_info->lock);
3289
3290         /*
3291          * If we have mixed data/metadata chunks we want to make sure we keep
3292          * allocating mixed chunks instead of individual chunks.
3293          */
3294         if (btrfs_mixed_space_info(space_info))
3295                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3296
3297         /*
3298          * if we're doing a data chunk, go ahead and make sure that
3299          * we keep a reasonable number of metadata chunks allocated in the
3300          * FS as well.
3301          */
3302         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3303                 fs_info->data_chunk_allocations++;
3304                 if (!(fs_info->data_chunk_allocations %
3305                       fs_info->metadata_ratio))
3306                         force_metadata_allocation(fs_info);
3307         }
3308
3309         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3310         spin_lock(&space_info->lock);
3311         if (ret)
3312                 space_info->full = 1;
3313         else
3314                 ret = 1;
3315         space_info->force_alloc = 0;
3316         spin_unlock(&space_info->lock);
3317 out:
3318         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3319         return ret;
3320 }
3321
3322 /*
3323  * shrink metadata reservation for delalloc
3324  */
3325 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3326                            struct btrfs_root *root, u64 to_reclaim, int sync)
3327 {
3328         struct btrfs_block_rsv *block_rsv;
3329         struct btrfs_space_info *space_info;
3330         u64 reserved;
3331         u64 max_reclaim;
3332         u64 reclaimed = 0;
3333         int pause = 1;
3334         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3335
3336         block_rsv = &root->fs_info->delalloc_block_rsv;
3337         space_info = block_rsv->space_info;
3338
3339         smp_mb();
3340         reserved = space_info->bytes_reserved;
3341
3342         if (reserved == 0)
3343                 return 0;
3344
3345         max_reclaim = min(reserved, to_reclaim);
3346
3347         while (1) {
3348                 /* have the flusher threads jump in and do some IO */
3349                 smp_mb();
3350                 nr_pages = min_t(unsigned long, nr_pages,
3351                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3352                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3353
3354                 spin_lock(&space_info->lock);
3355                 if (reserved > space_info->bytes_reserved)
3356                         reclaimed += reserved - space_info->bytes_reserved;
3357                 reserved = space_info->bytes_reserved;
3358                 spin_unlock(&space_info->lock);
3359
3360                 if (reserved == 0 || reclaimed >= max_reclaim)
3361                         break;
3362
3363                 if (trans && trans->transaction->blocked)
3364                         return -EAGAIN;
3365
3366                 __set_current_state(TASK_INTERRUPTIBLE);
3367                 schedule_timeout(pause);
3368                 pause <<= 1;
3369                 if (pause > HZ / 10)
3370                         pause = HZ / 10;
3371
3372         }
3373         return reclaimed >= to_reclaim;
3374 }
3375
3376 /*
3377  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3378  * idea is if this is the first call (retries == 0) then we will add to our
3379  * reserved count if we can't make the allocation in order to hold our place
3380  * while we go and try and free up space.  That way for retries > 1 we don't try
3381  * and add space, we just check to see if the amount of unused space is >= the
3382  * total space, meaning that our reservation is valid.
3383  *
3384  * However if we don't intend to retry this reservation, pass -1 as retries so
3385  * that it short circuits this logic.
3386  */
3387 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3388                                   struct btrfs_root *root,
3389                                   struct btrfs_block_rsv *block_rsv,
3390                                   u64 orig_bytes, int flush)
3391 {
3392         struct btrfs_space_info *space_info = block_rsv->space_info;
3393         u64 unused;
3394         u64 num_bytes = orig_bytes;
3395         int retries = 0;
3396         int ret = 0;
3397         bool reserved = false;
3398         bool committed = false;
3399
3400 again:
3401         ret = -ENOSPC;
3402         if (reserved)
3403                 num_bytes = 0;
3404
3405         spin_lock(&space_info->lock);
3406         unused = space_info->bytes_used + space_info->bytes_reserved +
3407                  space_info->bytes_pinned + space_info->bytes_readonly +
3408                  space_info->bytes_may_use;
3409
3410         /*
3411          * The idea here is that we've not already over-reserved the block group
3412          * then we can go ahead and save our reservation first and then start
3413          * flushing if we need to.  Otherwise if we've already overcommitted
3414          * lets start flushing stuff first and then come back and try to make
3415          * our reservation.
3416          */
3417         if (unused <= space_info->total_bytes) {
3418                 unused = space_info->total_bytes - unused;
3419                 if (unused >= num_bytes) {
3420                         if (!reserved)
3421                                 space_info->bytes_reserved += orig_bytes;
3422                         ret = 0;
3423                 } else {
3424                         /*
3425                          * Ok set num_bytes to orig_bytes since we aren't
3426                          * overocmmitted, this way we only try and reclaim what
3427                          * we need.
3428                          */
3429                         num_bytes = orig_bytes;
3430                 }
3431         } else {
3432                 /*
3433                  * Ok we're over committed, set num_bytes to the overcommitted
3434                  * amount plus the amount of bytes that we need for this
3435                  * reservation.
3436                  */
3437                 num_bytes = unused - space_info->total_bytes +
3438                         (orig_bytes * (retries + 1));
3439         }
3440
3441         /*
3442          * Couldn't make our reservation, save our place so while we're trying
3443          * to reclaim space we can actually use it instead of somebody else
3444          * stealing it from us.
3445          */
3446         if (ret && !reserved) {
3447                 space_info->bytes_reserved += orig_bytes;
3448                 reserved = true;
3449         }
3450
3451         spin_unlock(&space_info->lock);
3452
3453         if (!ret)
3454                 return 0;
3455
3456         if (!flush)
3457                 goto out;
3458
3459         /*
3460          * We do synchronous shrinking since we don't actually unreserve
3461          * metadata until after the IO is completed.
3462          */
3463         ret = shrink_delalloc(trans, root, num_bytes, 1);
3464         if (ret > 0)
3465                 return 0;
3466         else if (ret < 0)
3467                 goto out;
3468
3469         /*
3470          * So if we were overcommitted it's possible that somebody else flushed
3471          * out enough space and we simply didn't have enough space to reclaim,
3472          * so go back around and try again.
3473          */
3474         if (retries < 2) {
3475                 retries++;
3476                 goto again;
3477         }
3478
3479         spin_lock(&space_info->lock);
3480         /*
3481          * Not enough space to be reclaimed, don't bother committing the
3482          * transaction.
3483          */
3484         if (space_info->bytes_pinned < orig_bytes)
3485                 ret = -ENOSPC;
3486         spin_unlock(&space_info->lock);
3487         if (ret)
3488                 goto out;
3489
3490         ret = -EAGAIN;
3491         if (trans || committed)
3492                 goto out;
3493
3494         ret = -ENOSPC;
3495         trans = btrfs_join_transaction(root, 1);
3496         if (IS_ERR(trans))
3497                 goto out;
3498         ret = btrfs_commit_transaction(trans, root);
3499         if (!ret) {
3500                 trans = NULL;
3501                 committed = true;
3502                 goto again;
3503         }
3504
3505 out:
3506         if (reserved) {
3507                 spin_lock(&space_info->lock);
3508                 space_info->bytes_reserved -= orig_bytes;
3509                 spin_unlock(&space_info->lock);
3510         }
3511
3512         return ret;
3513 }
3514
3515 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3516                                              struct btrfs_root *root)
3517 {
3518         struct btrfs_block_rsv *block_rsv;
3519         if (root->ref_cows)
3520                 block_rsv = trans->block_rsv;
3521         else
3522                 block_rsv = root->block_rsv;
3523
3524         if (!block_rsv)
3525                 block_rsv = &root->fs_info->empty_block_rsv;
3526
3527         return block_rsv;
3528 }
3529
3530 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3531                                u64 num_bytes)
3532 {
3533         int ret = -ENOSPC;
3534         spin_lock(&block_rsv->lock);
3535         if (block_rsv->reserved >= num_bytes) {
3536                 block_rsv->reserved -= num_bytes;
3537                 if (block_rsv->reserved < block_rsv->size)
3538                         block_rsv->full = 0;
3539                 ret = 0;
3540         }
3541         spin_unlock(&block_rsv->lock);
3542         return ret;
3543 }
3544
3545 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3546                                 u64 num_bytes, int update_size)
3547 {
3548         spin_lock(&block_rsv->lock);
3549         block_rsv->reserved += num_bytes;
3550         if (update_size)
3551                 block_rsv->size += num_bytes;
3552         else if (block_rsv->reserved >= block_rsv->size)
3553                 block_rsv->full = 1;
3554         spin_unlock(&block_rsv->lock);
3555 }
3556
3557 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3558                              struct btrfs_block_rsv *dest, u64 num_bytes)
3559 {
3560         struct btrfs_space_info *space_info = block_rsv->space_info;
3561
3562         spin_lock(&block_rsv->lock);
3563         if (num_bytes == (u64)-1)
3564                 num_bytes = block_rsv->size;
3565         block_rsv->size -= num_bytes;
3566         if (block_rsv->reserved >= block_rsv->size) {
3567                 num_bytes = block_rsv->reserved - block_rsv->size;
3568                 block_rsv->reserved = block_rsv->size;
3569                 block_rsv->full = 1;
3570         } else {
3571                 num_bytes = 0;
3572         }
3573         spin_unlock(&block_rsv->lock);
3574
3575         if (num_bytes > 0) {
3576                 if (dest) {
3577                         block_rsv_add_bytes(dest, num_bytes, 0);
3578                 } else {
3579                         spin_lock(&space_info->lock);
3580                         space_info->bytes_reserved -= num_bytes;
3581                         spin_unlock(&space_info->lock);
3582                 }
3583         }
3584 }
3585
3586 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3587                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3588 {
3589         int ret;
3590
3591         ret = block_rsv_use_bytes(src, num_bytes);
3592         if (ret)
3593                 return ret;
3594
3595         block_rsv_add_bytes(dst, num_bytes, 1);
3596         return 0;
3597 }
3598
3599 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3600 {
3601         memset(rsv, 0, sizeof(*rsv));
3602         spin_lock_init(&rsv->lock);
3603         atomic_set(&rsv->usage, 1);
3604         rsv->priority = 6;
3605         INIT_LIST_HEAD(&rsv->list);
3606 }
3607
3608 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3609 {
3610         struct btrfs_block_rsv *block_rsv;
3611         struct btrfs_fs_info *fs_info = root->fs_info;
3612
3613         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3614         if (!block_rsv)
3615                 return NULL;
3616
3617         btrfs_init_block_rsv(block_rsv);
3618         block_rsv->space_info = __find_space_info(fs_info,
3619                                                   BTRFS_BLOCK_GROUP_METADATA);
3620         return block_rsv;
3621 }
3622
3623 void btrfs_free_block_rsv(struct btrfs_root *root,
3624                           struct btrfs_block_rsv *rsv)
3625 {
3626         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3627                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3628                 if (!rsv->durable)
3629                         kfree(rsv);
3630         }
3631 }
3632
3633 /*
3634  * make the block_rsv struct be able to capture freed space.
3635  * the captured space will re-add to the the block_rsv struct
3636  * after transaction commit
3637  */
3638 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3639                                  struct btrfs_block_rsv *block_rsv)
3640 {
3641         block_rsv->durable = 1;
3642         mutex_lock(&fs_info->durable_block_rsv_mutex);
3643         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3644         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3645 }
3646
3647 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3648                         struct btrfs_root *root,
3649                         struct btrfs_block_rsv *block_rsv,
3650                         u64 num_bytes)
3651 {
3652         int ret;
3653
3654         if (num_bytes == 0)
3655                 return 0;
3656
3657         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3658         if (!ret) {
3659                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3660                 return 0;
3661         }
3662
3663         return ret;
3664 }
3665
3666 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3667                           struct btrfs_root *root,
3668                           struct btrfs_block_rsv *block_rsv,
3669                           u64 min_reserved, int min_factor)
3670 {
3671         u64 num_bytes = 0;
3672         int commit_trans = 0;
3673         int ret = -ENOSPC;
3674
3675         if (!block_rsv)
3676                 return 0;
3677
3678         spin_lock(&block_rsv->lock);
3679         if (min_factor > 0)
3680                 num_bytes = div_factor(block_rsv->size, min_factor);
3681         if (min_reserved > num_bytes)
3682                 num_bytes = min_reserved;
3683
3684         if (block_rsv->reserved >= num_bytes) {
3685                 ret = 0;
3686         } else {
3687                 num_bytes -= block_rsv->reserved;
3688                 if (block_rsv->durable &&
3689                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3690                         commit_trans = 1;
3691         }
3692         spin_unlock(&block_rsv->lock);
3693         if (!ret)
3694                 return 0;
3695
3696         if (block_rsv->refill_used) {
3697                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3698                                              num_bytes, 0);
3699                 if (!ret) {
3700                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3701                         return 0;
3702                 }
3703         }
3704
3705         if (commit_trans) {
3706                 if (trans)
3707                         return -EAGAIN;
3708
3709                 trans = btrfs_join_transaction(root, 1);
3710                 BUG_ON(IS_ERR(trans));
3711                 ret = btrfs_commit_transaction(trans, root);
3712                 return 0;
3713         }
3714
3715         WARN_ON(1);
3716         printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
3717                 block_rsv->size, block_rsv->reserved,
3718                 block_rsv->freed[0], block_rsv->freed[1]);
3719
3720         return -ENOSPC;
3721 }
3722
3723 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3724                             struct btrfs_block_rsv *dst_rsv,
3725                             u64 num_bytes)
3726 {
3727         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3728 }
3729
3730 void btrfs_block_rsv_release(struct btrfs_root *root,
3731                              struct btrfs_block_rsv *block_rsv,
3732                              u64 num_bytes)
3733 {
3734         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3735         if (global_rsv->full || global_rsv == block_rsv ||
3736             block_rsv->space_info != global_rsv->space_info)
3737                 global_rsv = NULL;
3738         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3739 }
3740
3741 /*
3742  * helper to calculate size of global block reservation.
3743  * the desired value is sum of space used by extent tree,
3744  * checksum tree and root tree
3745  */
3746 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3747 {
3748         struct btrfs_space_info *sinfo;
3749         u64 num_bytes;
3750         u64 meta_used;
3751         u64 data_used;
3752         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3753 #if 0
3754         /*
3755          * per tree used space accounting can be inaccuracy, so we
3756          * can't rely on it.
3757          */
3758         spin_lock(&fs_info->extent_root->accounting_lock);
3759         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3760         spin_unlock(&fs_info->extent_root->accounting_lock);
3761
3762         spin_lock(&fs_info->csum_root->accounting_lock);
3763         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3764         spin_unlock(&fs_info->csum_root->accounting_lock);
3765
3766         spin_lock(&fs_info->tree_root->accounting_lock);
3767         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3768         spin_unlock(&fs_info->tree_root->accounting_lock);
3769 #endif
3770         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3771         spin_lock(&sinfo->lock);
3772         data_used = sinfo->bytes_used;
3773         spin_unlock(&sinfo->lock);
3774
3775         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3776         spin_lock(&sinfo->lock);
3777         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3778                 data_used = 0;
3779         meta_used = sinfo->bytes_used;
3780         spin_unlock(&sinfo->lock);
3781
3782         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3783                     csum_size * 2;
3784         num_bytes += div64_u64(data_used + meta_used, 50);
3785
3786         if (num_bytes * 3 > meta_used)
3787                 num_bytes = div64_u64(meta_used, 3);
3788
3789         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3790 }
3791
3792 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3793 {
3794         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3795         struct btrfs_space_info *sinfo = block_rsv->space_info;
3796         u64 num_bytes;
3797
3798         num_bytes = calc_global_metadata_size(fs_info);
3799
3800         spin_lock(&block_rsv->lock);
3801         spin_lock(&sinfo->lock);
3802
3803         block_rsv->size = num_bytes;
3804
3805         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3806                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3807                     sinfo->bytes_may_use;
3808
3809         if (sinfo->total_bytes > num_bytes) {
3810                 num_bytes = sinfo->total_bytes - num_bytes;
3811                 block_rsv->reserved += num_bytes;
3812                 sinfo->bytes_reserved += num_bytes;
3813         }
3814
3815         if (block_rsv->reserved >= block_rsv->size) {
3816                 num_bytes = block_rsv->reserved - block_rsv->size;
3817                 sinfo->bytes_reserved -= num_bytes;
3818                 block_rsv->reserved = block_rsv->size;
3819                 block_rsv->full = 1;
3820         }
3821 #if 0
3822         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3823                 block_rsv->size, block_rsv->reserved);
3824 #endif
3825         spin_unlock(&sinfo->lock);
3826         spin_unlock(&block_rsv->lock);
3827 }
3828
3829 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3830 {
3831         struct btrfs_space_info *space_info;
3832
3833         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3834         fs_info->chunk_block_rsv.space_info = space_info;
3835         fs_info->chunk_block_rsv.priority = 10;
3836
3837         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3838         fs_info->global_block_rsv.space_info = space_info;
3839         fs_info->global_block_rsv.priority = 10;
3840         fs_info->global_block_rsv.refill_used = 1;
3841         fs_info->delalloc_block_rsv.space_info = space_info;
3842         fs_info->trans_block_rsv.space_info = space_info;
3843         fs_info->empty_block_rsv.space_info = space_info;
3844         fs_info->empty_block_rsv.priority = 10;
3845
3846         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3847         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3848         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3849         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3850         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3851
3852         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3853
3854         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3855
3856         update_global_block_rsv(fs_info);
3857 }
3858
3859 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3860 {
3861         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3862         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3863         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3864         WARN_ON(fs_info->trans_block_rsv.size > 0);
3865         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3866         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3867         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3868 }
3869
3870 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3871 {
3872         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3873                 3 * num_items;
3874 }
3875
3876 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3877                                  struct btrfs_root *root,
3878                                  int num_items)
3879 {
3880         u64 num_bytes;
3881         int ret;
3882
3883         if (num_items == 0 || root->fs_info->chunk_root == root)
3884                 return 0;
3885
3886         num_bytes = calc_trans_metadata_size(root, num_items);
3887         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3888                                   num_bytes);
3889         if (!ret) {
3890                 trans->bytes_reserved += num_bytes;
3891                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3892         }
3893         return ret;
3894 }
3895
3896 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3897                                   struct btrfs_root *root)
3898 {
3899         if (!trans->bytes_reserved)
3900                 return;
3901
3902         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3903         btrfs_block_rsv_release(root, trans->block_rsv,
3904                                 trans->bytes_reserved);
3905         trans->bytes_reserved = 0;
3906 }
3907
3908 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3909                                   struct inode *inode)
3910 {
3911         struct btrfs_root *root = BTRFS_I(inode)->root;
3912         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3913         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3914
3915         /*
3916          * one for deleting orphan item, one for updating inode and
3917          * two for calling btrfs_truncate_inode_items.
3918          *
3919          * btrfs_truncate_inode_items is a delete operation, it frees
3920          * more space than it uses in most cases. So two units of
3921          * metadata space should be enough for calling it many times.
3922          * If all of the metadata space is used, we can commit
3923          * transaction and use space it freed.
3924          */
3925         u64 num_bytes = calc_trans_metadata_size(root, 4);
3926         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3927 }
3928
3929 void btrfs_orphan_release_metadata(struct inode *inode)
3930 {
3931         struct btrfs_root *root = BTRFS_I(inode)->root;
3932         u64 num_bytes = calc_trans_metadata_size(root, 4);
3933         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3934 }
3935
3936 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3937                                 struct btrfs_pending_snapshot *pending)
3938 {
3939         struct btrfs_root *root = pending->root;
3940         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3941         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3942         /*
3943          * two for root back/forward refs, two for directory entries
3944          * and one for root of the snapshot.
3945          */
3946         u64 num_bytes = calc_trans_metadata_size(root, 5);
3947         dst_rsv->space_info = src_rsv->space_info;
3948         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3949 }
3950
3951 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3952 {
3953         return num_bytes >>= 3;
3954 }
3955
3956 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3957 {
3958         struct btrfs_root *root = BTRFS_I(inode)->root;
3959         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3960         u64 to_reserve;
3961         int nr_extents;
3962         int ret;
3963
3964         if (btrfs_transaction_in_commit(root->fs_info))
3965                 schedule_timeout(1);
3966
3967         num_bytes = ALIGN(num_bytes, root->sectorsize);
3968
3969         spin_lock(&BTRFS_I(inode)->accounting_lock);
3970         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
3971         if (nr_extents > BTRFS_I(inode)->reserved_extents) {
3972                 nr_extents -= BTRFS_I(inode)->reserved_extents;
3973                 to_reserve = calc_trans_metadata_size(root, nr_extents);
3974         } else {
3975                 nr_extents = 0;
3976                 to_reserve = 0;
3977         }
3978         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3979
3980         to_reserve += calc_csum_metadata_size(inode, num_bytes);
3981         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
3982         if (ret)
3983                 return ret;
3984
3985         spin_lock(&BTRFS_I(inode)->accounting_lock);
3986         BTRFS_I(inode)->reserved_extents += nr_extents;
3987         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
3988         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3989
3990         block_rsv_add_bytes(block_rsv, to_reserve, 1);
3991
3992         if (block_rsv->size > 512 * 1024 * 1024)
3993                 shrink_delalloc(NULL, root, to_reserve, 0);
3994
3995         return 0;
3996 }
3997
3998 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
3999 {
4000         struct btrfs_root *root = BTRFS_I(inode)->root;
4001         u64 to_free;
4002         int nr_extents;
4003
4004         num_bytes = ALIGN(num_bytes, root->sectorsize);
4005         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4006
4007         spin_lock(&BTRFS_I(inode)->accounting_lock);
4008         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4009         if (nr_extents < BTRFS_I(inode)->reserved_extents) {
4010                 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
4011                 BTRFS_I(inode)->reserved_extents -= nr_extents;
4012         } else {
4013                 nr_extents = 0;
4014         }
4015         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4016
4017         to_free = calc_csum_metadata_size(inode, num_bytes);
4018         if (nr_extents > 0)
4019                 to_free += calc_trans_metadata_size(root, nr_extents);
4020
4021         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4022                                 to_free);
4023 }
4024
4025 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4026 {
4027         int ret;
4028
4029         ret = btrfs_check_data_free_space(inode, num_bytes);
4030         if (ret)
4031                 return ret;
4032
4033         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4034         if (ret) {
4035                 btrfs_free_reserved_data_space(inode, num_bytes);
4036                 return ret;
4037         }
4038
4039         return 0;
4040 }
4041
4042 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4043 {
4044         btrfs_delalloc_release_metadata(inode, num_bytes);
4045         btrfs_free_reserved_data_space(inode, num_bytes);
4046 }
4047
4048 static int update_block_group(struct btrfs_trans_handle *trans,
4049                               struct btrfs_root *root,
4050                               u64 bytenr, u64 num_bytes, int alloc)
4051 {
4052         struct btrfs_block_group_cache *cache = NULL;
4053         struct btrfs_fs_info *info = root->fs_info;
4054         u64 total = num_bytes;
4055         u64 old_val;
4056         u64 byte_in_group;
4057         int factor;
4058
4059         /* block accounting for super block */
4060         spin_lock(&info->delalloc_lock);
4061         old_val = btrfs_super_bytes_used(&info->super_copy);
4062         if (alloc)
4063                 old_val += num_bytes;
4064         else
4065                 old_val -= num_bytes;
4066         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4067         spin_unlock(&info->delalloc_lock);
4068
4069         while (total) {
4070                 cache = btrfs_lookup_block_group(info, bytenr);
4071                 if (!cache)
4072                         return -1;
4073                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4074                                     BTRFS_BLOCK_GROUP_RAID1 |
4075                                     BTRFS_BLOCK_GROUP_RAID10))
4076                         factor = 2;
4077                 else
4078                         factor = 1;
4079                 /*
4080                  * If this block group has free space cache written out, we
4081                  * need to make sure to load it if we are removing space.  This
4082                  * is because we need the unpinning stage to actually add the
4083                  * space back to the block group, otherwise we will leak space.
4084                  */
4085                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4086                         cache_block_group(cache, trans, 1);
4087
4088                 byte_in_group = bytenr - cache->key.objectid;
4089                 WARN_ON(byte_in_group > cache->key.offset);
4090
4091                 spin_lock(&cache->space_info->lock);
4092                 spin_lock(&cache->lock);
4093
4094                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4095                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4096                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4097
4098                 cache->dirty = 1;
4099                 old_val = btrfs_block_group_used(&cache->item);
4100                 num_bytes = min(total, cache->key.offset - byte_in_group);
4101                 if (alloc) {
4102                         old_val += num_bytes;
4103                         btrfs_set_block_group_used(&cache->item, old_val);
4104                         cache->reserved -= num_bytes;
4105                         cache->space_info->bytes_reserved -= num_bytes;
4106                         cache->space_info->bytes_used += num_bytes;
4107                         cache->space_info->disk_used += num_bytes * factor;
4108                         spin_unlock(&cache->lock);
4109                         spin_unlock(&cache->space_info->lock);
4110                 } else {
4111                         old_val -= num_bytes;
4112                         btrfs_set_block_group_used(&cache->item, old_val);
4113                         cache->pinned += num_bytes;
4114                         cache->space_info->bytes_pinned += num_bytes;
4115                         cache->space_info->bytes_used -= num_bytes;
4116                         cache->space_info->disk_used -= num_bytes * factor;
4117                         spin_unlock(&cache->lock);
4118                         spin_unlock(&cache->space_info->lock);
4119
4120                         set_extent_dirty(info->pinned_extents,
4121                                          bytenr, bytenr + num_bytes - 1,
4122                                          GFP_NOFS | __GFP_NOFAIL);
4123                 }
4124                 btrfs_put_block_group(cache);
4125                 total -= num_bytes;
4126                 bytenr += num_bytes;
4127         }
4128         return 0;
4129 }
4130
4131 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4132 {
4133         struct btrfs_block_group_cache *cache;
4134         u64 bytenr;
4135
4136         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4137         if (!cache)
4138                 return 0;
4139
4140         bytenr = cache->key.objectid;
4141         btrfs_put_block_group(cache);
4142
4143         return bytenr;
4144 }
4145
4146 static int pin_down_extent(struct btrfs_root *root,
4147                            struct btrfs_block_group_cache *cache,
4148                            u64 bytenr, u64 num_bytes, int reserved)
4149 {
4150         spin_lock(&cache->space_info->lock);
4151         spin_lock(&cache->lock);
4152         cache->pinned += num_bytes;
4153         cache->space_info->bytes_pinned += num_bytes;
4154         if (reserved) {
4155                 cache->reserved -= num_bytes;
4156                 cache->space_info->bytes_reserved -= num_bytes;
4157         }
4158         spin_unlock(&cache->lock);
4159         spin_unlock(&cache->space_info->lock);
4160
4161         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4162                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4163         return 0;
4164 }
4165
4166 /*
4167  * this function must be called within transaction
4168  */
4169 int btrfs_pin_extent(struct btrfs_root *root,
4170                      u64 bytenr, u64 num_bytes, int reserved)
4171 {
4172         struct btrfs_block_group_cache *cache;
4173
4174         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4175         BUG_ON(!cache);
4176
4177         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4178
4179         btrfs_put_block_group(cache);
4180         return 0;
4181 }
4182
4183 /*
4184  * update size of reserved extents. this function may return -EAGAIN
4185  * if 'reserve' is true or 'sinfo' is false.
4186  */
4187 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4188                                  u64 num_bytes, int reserve, int sinfo)
4189 {
4190         int ret = 0;
4191         if (sinfo) {
4192                 struct btrfs_space_info *space_info = cache->space_info;
4193                 spin_lock(&space_info->lock);
4194                 spin_lock(&cache->lock);
4195                 if (reserve) {
4196                         if (cache->ro) {
4197                                 ret = -EAGAIN;
4198                         } else {
4199                                 cache->reserved += num_bytes;
4200                                 space_info->bytes_reserved += num_bytes;
4201                         }
4202                 } else {
4203                         if (cache->ro)
4204                                 space_info->bytes_readonly += num_bytes;
4205                         cache->reserved -= num_bytes;
4206                         space_info->bytes_reserved -= num_bytes;
4207                 }
4208                 spin_unlock(&cache->lock);
4209                 spin_unlock(&space_info->lock);
4210         } else {
4211                 spin_lock(&cache->lock);
4212                 if (cache->ro) {
4213                         ret = -EAGAIN;
4214                 } else {
4215                         if (reserve)
4216                                 cache->reserved += num_bytes;
4217                         else
4218                                 cache->reserved -= num_bytes;
4219                 }
4220                 spin_unlock(&cache->lock);
4221         }
4222         return ret;
4223 }
4224
4225 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4226                                 struct btrfs_root *root)
4227 {
4228         struct btrfs_fs_info *fs_info = root->fs_info;
4229         struct btrfs_caching_control *next;
4230         struct btrfs_caching_control *caching_ctl;
4231         struct btrfs_block_group_cache *cache;
4232
4233         down_write(&fs_info->extent_commit_sem);
4234
4235         list_for_each_entry_safe(caching_ctl, next,
4236                                  &fs_info->caching_block_groups, list) {
4237                 cache = caching_ctl->block_group;
4238                 if (block_group_cache_done(cache)) {
4239                         cache->last_byte_to_unpin = (u64)-1;
4240                         list_del_init(&caching_ctl->list);
4241                         put_caching_control(caching_ctl);
4242                 } else {
4243                         cache->last_byte_to_unpin = caching_ctl->progress;
4244                 }
4245         }
4246
4247         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4248                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4249         else
4250                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4251
4252         up_write(&fs_info->extent_commit_sem);
4253
4254         update_global_block_rsv(fs_info);
4255         return 0;
4256 }
4257
4258 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4259 {
4260         struct btrfs_fs_info *fs_info = root->fs_info;
4261         struct btrfs_block_group_cache *cache = NULL;
4262         u64 len;
4263
4264         while (start <= end) {
4265                 if (!cache ||
4266                     start >= cache->key.objectid + cache->key.offset) {
4267                         if (cache)
4268                                 btrfs_put_block_group(cache);
4269                         cache = btrfs_lookup_block_group(fs_info, start);
4270                         BUG_ON(!cache);
4271                 }
4272
4273                 len = cache->key.objectid + cache->key.offset - start;
4274                 len = min(len, end + 1 - start);
4275
4276                 if (start < cache->last_byte_to_unpin) {
4277                         len = min(len, cache->last_byte_to_unpin - start);
4278                         btrfs_add_free_space(cache, start, len);
4279                 }
4280
4281                 start += len;
4282
4283                 spin_lock(&cache->space_info->lock);
4284                 spin_lock(&cache->lock);
4285                 cache->pinned -= len;
4286                 cache->space_info->bytes_pinned -= len;
4287                 if (cache->ro) {
4288                         cache->space_info->bytes_readonly += len;
4289                 } else if (cache->reserved_pinned > 0) {
4290                         len = min(len, cache->reserved_pinned);
4291                         cache->reserved_pinned -= len;
4292                         cache->space_info->bytes_reserved += len;
4293                 }
4294                 spin_unlock(&cache->lock);
4295                 spin_unlock(&cache->space_info->lock);
4296         }
4297
4298         if (cache)
4299                 btrfs_put_block_group(cache);
4300         return 0;
4301 }
4302
4303 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4304                                struct btrfs_root *root)
4305 {
4306         struct btrfs_fs_info *fs_info = root->fs_info;
4307         struct extent_io_tree *unpin;
4308         struct btrfs_block_rsv *block_rsv;
4309         struct btrfs_block_rsv *next_rsv;
4310         u64 start;
4311         u64 end;
4312         int idx;
4313         int ret;
4314
4315         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4316                 unpin = &fs_info->freed_extents[1];
4317         else
4318                 unpin = &fs_info->freed_extents[0];
4319
4320         while (1) {
4321                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4322                                             EXTENT_DIRTY);
4323                 if (ret)
4324                         break;
4325
4326                 ret = btrfs_discard_extent(root, start, end + 1 - start);
4327
4328                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4329                 unpin_extent_range(root, start, end);
4330                 cond_resched();
4331         }
4332
4333         mutex_lock(&fs_info->durable_block_rsv_mutex);
4334         list_for_each_entry_safe(block_rsv, next_rsv,
4335                                  &fs_info->durable_block_rsv_list, list) {
4336
4337                 idx = trans->transid & 0x1;
4338                 if (block_rsv->freed[idx] > 0) {
4339                         block_rsv_add_bytes(block_rsv,
4340                                             block_rsv->freed[idx], 0);
4341                         block_rsv->freed[idx] = 0;
4342                 }
4343                 if (atomic_read(&block_rsv->usage) == 0) {
4344                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4345
4346                         if (block_rsv->freed[0] == 0 &&
4347                             block_rsv->freed[1] == 0) {
4348                                 list_del_init(&block_rsv->list);
4349                                 kfree(block_rsv);
4350                         }
4351                 } else {
4352                         btrfs_block_rsv_release(root, block_rsv, 0);
4353                 }
4354         }
4355         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4356
4357         return 0;
4358 }
4359
4360 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4361                                 struct btrfs_root *root,
4362                                 u64 bytenr, u64 num_bytes, u64 parent,
4363                                 u64 root_objectid, u64 owner_objectid,
4364                                 u64 owner_offset, int refs_to_drop,
4365                                 struct btrfs_delayed_extent_op *extent_op)
4366 {
4367         struct btrfs_key key;
4368         struct btrfs_path *path;
4369         struct btrfs_fs_info *info = root->fs_info;
4370         struct btrfs_root *extent_root = info->extent_root;
4371         struct extent_buffer *leaf;
4372         struct btrfs_extent_item *ei;
4373         struct btrfs_extent_inline_ref *iref;
4374         int ret;
4375         int is_data;
4376         int extent_slot = 0;
4377         int found_extent = 0;
4378         int num_to_del = 1;
4379         u32 item_size;
4380         u64 refs;
4381
4382         path = btrfs_alloc_path();
4383         if (!path)
4384                 return -ENOMEM;
4385
4386         path->reada = 1;
4387         path->leave_spinning = 1;
4388
4389         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4390         BUG_ON(!is_data && refs_to_drop != 1);
4391
4392         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4393                                     bytenr, num_bytes, parent,
4394                                     root_objectid, owner_objectid,
4395                                     owner_offset);
4396         if (ret == 0) {
4397                 extent_slot = path->slots[0];
4398                 while (extent_slot >= 0) {
4399                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4400                                               extent_slot);
4401                         if (key.objectid != bytenr)
4402                                 break;
4403                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4404                             key.offset == num_bytes) {
4405                                 found_extent = 1;
4406                                 break;
4407                         }
4408                         if (path->slots[0] - extent_slot > 5)
4409                                 break;
4410                         extent_slot--;
4411                 }
4412 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4413                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4414                 if (found_extent && item_size < sizeof(*ei))
4415                         found_extent = 0;
4416 #endif
4417                 if (!found_extent) {
4418                         BUG_ON(iref);
4419                         ret = remove_extent_backref(trans, extent_root, path,
4420                                                     NULL, refs_to_drop,
4421                                                     is_data);
4422                         BUG_ON(ret);
4423                         btrfs_release_path(extent_root, path);
4424                         path->leave_spinning = 1;
4425
4426                         key.objectid = bytenr;
4427                         key.type = BTRFS_EXTENT_ITEM_KEY;
4428                         key.offset = num_bytes;
4429
4430                         ret = btrfs_search_slot(trans, extent_root,
4431                                                 &key, path, -1, 1);
4432                         if (ret) {
4433                                 printk(KERN_ERR "umm, got %d back from search"
4434                                        ", was looking for %llu\n", ret,
4435                                        (unsigned long long)bytenr);
4436                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4437                         }
4438                         BUG_ON(ret);
4439                         extent_slot = path->slots[0];
4440                 }
4441         } else {
4442                 btrfs_print_leaf(extent_root, path->nodes[0]);
4443                 WARN_ON(1);
4444                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4445                        "parent %llu root %llu  owner %llu offset %llu\n",
4446                        (unsigned long long)bytenr,
4447                        (unsigned long long)parent,
4448                        (unsigned long long)root_objectid,
4449                        (unsigned long long)owner_objectid,
4450                        (unsigned long long)owner_offset);
4451         }
4452
4453         leaf = path->nodes[0];
4454         item_size = btrfs_item_size_nr(leaf, extent_slot);
4455 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4456         if (item_size < sizeof(*ei)) {
4457                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4458                 ret = convert_extent_item_v0(trans, extent_root, path,
4459                                              owner_objectid, 0);
4460                 BUG_ON(ret < 0);
4461
4462                 btrfs_release_path(extent_root, path);
4463                 path->leave_spinning = 1;
4464
4465                 key.objectid = bytenr;
4466                 key.type = BTRFS_EXTENT_ITEM_KEY;
4467                 key.offset = num_bytes;
4468
4469                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4470                                         -1, 1);
4471                 if (ret) {
4472                         printk(KERN_ERR "umm, got %d back from search"
4473                                ", was looking for %llu\n", ret,
4474                                (unsigned long long)bytenr);
4475                         btrfs_print_leaf(extent_root, path->nodes[0]);
4476                 }
4477                 BUG_ON(ret);
4478                 extent_slot = path->slots[0];
4479                 leaf = path->nodes[0];
4480                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4481         }
4482 #endif
4483         BUG_ON(item_size < sizeof(*ei));
4484         ei = btrfs_item_ptr(leaf, extent_slot,
4485                             struct btrfs_extent_item);
4486         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4487                 struct btrfs_tree_block_info *bi;
4488                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4489                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4490                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4491         }
4492
4493         refs = btrfs_extent_refs(leaf, ei);
4494         BUG_ON(refs < refs_to_drop);
4495         refs -= refs_to_drop;
4496
4497         if (refs > 0) {
4498                 if (extent_op)
4499                         __run_delayed_extent_op(extent_op, leaf, ei);
4500                 /*
4501                  * In the case of inline back ref, reference count will
4502                  * be updated by remove_extent_backref
4503                  */
4504                 if (iref) {
4505                         BUG_ON(!found_extent);
4506                 } else {
4507                         btrfs_set_extent_refs(leaf, ei, refs);
4508                         btrfs_mark_buffer_dirty(leaf);
4509                 }
4510                 if (found_extent) {
4511                         ret = remove_extent_backref(trans, extent_root, path,
4512                                                     iref, refs_to_drop,
4513                                                     is_data);
4514                         BUG_ON(ret);
4515                 }
4516         } else {
4517                 if (found_extent) {
4518                         BUG_ON(is_data && refs_to_drop !=
4519                                extent_data_ref_count(root, path, iref));
4520                         if (iref) {
4521                                 BUG_ON(path->slots[0] != extent_slot);
4522                         } else {
4523                                 BUG_ON(path->slots[0] != extent_slot + 1);
4524                                 path->slots[0] = extent_slot;
4525                                 num_to_del = 2;
4526                         }
4527                 }
4528
4529                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4530                                       num_to_del);
4531                 BUG_ON(ret);
4532                 btrfs_release_path(extent_root, path);
4533
4534                 if (is_data) {
4535                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4536                         BUG_ON(ret);
4537                 } else {
4538                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4539                              bytenr >> PAGE_CACHE_SHIFT,
4540                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4541                 }
4542
4543                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4544                 BUG_ON(ret);
4545         }
4546         btrfs_free_path(path);
4547         return ret;
4548 }
4549
4550 /*
4551  * when we free an block, it is possible (and likely) that we free the last
4552  * delayed ref for that extent as well.  This searches the delayed ref tree for
4553  * a given extent, and if there are no other delayed refs to be processed, it
4554  * removes it from the tree.
4555  */
4556 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4557                                       struct btrfs_root *root, u64 bytenr)
4558 {
4559         struct btrfs_delayed_ref_head *head;
4560         struct btrfs_delayed_ref_root *delayed_refs;
4561         struct btrfs_delayed_ref_node *ref;
4562         struct rb_node *node;
4563         int ret = 0;
4564
4565         delayed_refs = &trans->transaction->delayed_refs;
4566         spin_lock(&delayed_refs->lock);
4567         head = btrfs_find_delayed_ref_head(trans, bytenr);
4568         if (!head)
4569                 goto out;
4570
4571         node = rb_prev(&head->node.rb_node);
4572         if (!node)
4573                 goto out;
4574
4575         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4576
4577         /* there are still entries for this ref, we can't drop it */
4578         if (ref->bytenr == bytenr)
4579                 goto out;
4580
4581         if (head->extent_op) {
4582                 if (!head->must_insert_reserved)
4583                         goto out;
4584                 kfree(head->extent_op);
4585                 head->extent_op = NULL;
4586         }
4587
4588         /*
4589          * waiting for the lock here would deadlock.  If someone else has it
4590          * locked they are already in the process of dropping it anyway
4591          */
4592         if (!mutex_trylock(&head->mutex))
4593                 goto out;
4594
4595         /*
4596          * at this point we have a head with no other entries.  Go
4597          * ahead and process it.
4598          */
4599         head->node.in_tree = 0;
4600         rb_erase(&head->node.rb_node, &delayed_refs->root);
4601
4602         delayed_refs->num_entries--;
4603
4604         /*
4605          * we don't take a ref on the node because we're removing it from the
4606          * tree, so we just steal the ref the tree was holding.
4607          */
4608         delayed_refs->num_heads--;
4609         if (list_empty(&head->cluster))
4610                 delayed_refs->num_heads_ready--;
4611
4612         list_del_init(&head->cluster);
4613         spin_unlock(&delayed_refs->lock);
4614
4615         BUG_ON(head->extent_op);
4616         if (head->must_insert_reserved)
4617                 ret = 1;
4618
4619         mutex_unlock(&head->mutex);
4620         btrfs_put_delayed_ref(&head->node);
4621         return ret;
4622 out:
4623         spin_unlock(&delayed_refs->lock);
4624         return 0;
4625 }
4626
4627 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4628                            struct btrfs_root *root,
4629                            struct extent_buffer *buf,
4630                            u64 parent, int last_ref)
4631 {
4632         struct btrfs_block_rsv *block_rsv;
4633         struct btrfs_block_group_cache *cache = NULL;
4634         int ret;
4635
4636         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4637                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4638                                                 parent, root->root_key.objectid,
4639                                                 btrfs_header_level(buf),
4640                                                 BTRFS_DROP_DELAYED_REF, NULL);
4641                 BUG_ON(ret);
4642         }
4643
4644         if (!last_ref)
4645                 return;
4646
4647         block_rsv = get_block_rsv(trans, root);
4648         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4649         if (block_rsv->space_info != cache->space_info)
4650                 goto out;
4651
4652         if (btrfs_header_generation(buf) == trans->transid) {
4653                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4654                         ret = check_ref_cleanup(trans, root, buf->start);
4655                         if (!ret)
4656                                 goto pin;
4657                 }
4658
4659                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4660                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4661                         goto pin;
4662                 }
4663
4664                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4665
4666                 btrfs_add_free_space(cache, buf->start, buf->len);
4667                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4668                 if (ret == -EAGAIN) {
4669                         /* block group became read-only */
4670                         update_reserved_bytes(cache, buf->len, 0, 1);
4671                         goto out;
4672                 }
4673
4674                 ret = 1;
4675                 spin_lock(&block_rsv->lock);
4676                 if (block_rsv->reserved < block_rsv->size) {
4677                         block_rsv->reserved += buf->len;
4678                         ret = 0;
4679                 }
4680                 spin_unlock(&block_rsv->lock);
4681
4682                 if (ret) {
4683                         spin_lock(&cache->space_info->lock);
4684                         cache->space_info->bytes_reserved -= buf->len;
4685                         spin_unlock(&cache->space_info->lock);
4686                 }
4687                 goto out;
4688         }
4689 pin:
4690         if (block_rsv->durable && !cache->ro) {
4691                 ret = 0;
4692                 spin_lock(&cache->lock);
4693                 if (!cache->ro) {
4694                         cache->reserved_pinned += buf->len;
4695                         ret = 1;
4696                 }
4697                 spin_unlock(&cache->lock);
4698
4699                 if (ret) {
4700                         spin_lock(&block_rsv->lock);
4701                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4702                         spin_unlock(&block_rsv->lock);
4703                 }
4704         }
4705 out:
4706         btrfs_put_block_group(cache);
4707 }
4708
4709 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4710                       struct btrfs_root *root,
4711                       u64 bytenr, u64 num_bytes, u64 parent,
4712                       u64 root_objectid, u64 owner, u64 offset)
4713 {
4714         int ret;
4715
4716         /*
4717          * tree log blocks never actually go into the extent allocation
4718          * tree, just update pinning info and exit early.
4719          */
4720         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4721                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4722                 /* unlocks the pinned mutex */
4723                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4724                 ret = 0;
4725         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4726                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4727                                         parent, root_objectid, (int)owner,
4728                                         BTRFS_DROP_DELAYED_REF, NULL);
4729                 BUG_ON(ret);
4730         } else {
4731                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4732                                         parent, root_objectid, owner,
4733                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4734                 BUG_ON(ret);
4735         }
4736         return ret;
4737 }
4738
4739 static u64 stripe_align(struct btrfs_root *root, u64 val)
4740 {
4741         u64 mask = ((u64)root->stripesize - 1);
4742         u64 ret = (val + mask) & ~mask;
4743         return ret;
4744 }
4745
4746 /*
4747  * when we wait for progress in the block group caching, its because
4748  * our allocation attempt failed at least once.  So, we must sleep
4749  * and let some progress happen before we try again.
4750  *
4751  * This function will sleep at least once waiting for new free space to
4752  * show up, and then it will check the block group free space numbers
4753  * for our min num_bytes.  Another option is to have it go ahead
4754  * and look in the rbtree for a free extent of a given size, but this
4755  * is a good start.
4756  */
4757 static noinline int
4758 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4759                                 u64 num_bytes)
4760 {
4761         struct btrfs_caching_control *caching_ctl;
4762         DEFINE_WAIT(wait);
4763
4764         caching_ctl = get_caching_control(cache);
4765         if (!caching_ctl)
4766                 return 0;
4767
4768         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4769                    (cache->free_space >= num_bytes));
4770
4771         put_caching_control(caching_ctl);
4772         return 0;
4773 }
4774
4775 static noinline int
4776 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4777 {
4778         struct btrfs_caching_control *caching_ctl;
4779         DEFINE_WAIT(wait);
4780
4781         caching_ctl = get_caching_control(cache);
4782         if (!caching_ctl)
4783                 return 0;
4784
4785         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4786
4787         put_caching_control(caching_ctl);
4788         return 0;
4789 }
4790
4791 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4792 {
4793         int index;
4794         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4795                 index = 0;
4796         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4797                 index = 1;
4798         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4799                 index = 2;
4800         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4801                 index = 3;
4802         else
4803                 index = 4;
4804         return index;
4805 }
4806
4807 enum btrfs_loop_type {
4808         LOOP_FIND_IDEAL = 0,
4809         LOOP_CACHING_NOWAIT = 1,
4810         LOOP_CACHING_WAIT = 2,
4811         LOOP_ALLOC_CHUNK = 3,
4812         LOOP_NO_EMPTY_SIZE = 4,
4813 };
4814
4815 /*
4816  * walks the btree of allocated extents and find a hole of a given size.
4817  * The key ins is changed to record the hole:
4818  * ins->objectid == block start
4819  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4820  * ins->offset == number of blocks
4821  * Any available blocks before search_start are skipped.
4822  */
4823 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4824                                      struct btrfs_root *orig_root,
4825                                      u64 num_bytes, u64 empty_size,
4826                                      u64 search_start, u64 search_end,
4827                                      u64 hint_byte, struct btrfs_key *ins,
4828                                      int data)
4829 {
4830         int ret = 0;
4831         struct btrfs_root *root = orig_root->fs_info->extent_root;
4832         struct btrfs_free_cluster *last_ptr = NULL;
4833         struct btrfs_block_group_cache *block_group = NULL;
4834         int empty_cluster = 2 * 1024 * 1024;
4835         int allowed_chunk_alloc = 0;
4836         int done_chunk_alloc = 0;
4837         struct btrfs_space_info *space_info;
4838         int last_ptr_loop = 0;
4839         int loop = 0;
4840         int index = 0;
4841         bool found_uncached_bg = false;
4842         bool failed_cluster_refill = false;
4843         bool failed_alloc = false;
4844         bool use_cluster = true;
4845         u64 ideal_cache_percent = 0;
4846         u64 ideal_cache_offset = 0;
4847
4848         WARN_ON(num_bytes < root->sectorsize);
4849         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4850         ins->objectid = 0;
4851         ins->offset = 0;
4852
4853         space_info = __find_space_info(root->fs_info, data);
4854         if (!space_info) {
4855                 printk(KERN_ERR "No space info for %d\n", data);
4856                 return -ENOSPC;
4857         }
4858
4859         /*
4860          * If the space info is for both data and metadata it means we have a
4861          * small filesystem and we can't use the clustering stuff.
4862          */
4863         if (btrfs_mixed_space_info(space_info))
4864                 use_cluster = false;
4865
4866         if (orig_root->ref_cows || empty_size)
4867                 allowed_chunk_alloc = 1;
4868
4869         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4870                 last_ptr = &root->fs_info->meta_alloc_cluster;
4871                 if (!btrfs_test_opt(root, SSD))
4872                         empty_cluster = 64 * 1024;
4873         }
4874
4875         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4876             btrfs_test_opt(root, SSD)) {
4877                 last_ptr = &root->fs_info->data_alloc_cluster;
4878         }
4879
4880         if (last_ptr) {
4881                 spin_lock(&last_ptr->lock);
4882                 if (last_ptr->block_group)
4883                         hint_byte = last_ptr->window_start;
4884                 spin_unlock(&last_ptr->lock);
4885         }
4886
4887         search_start = max(search_start, first_logical_byte(root, 0));
4888         search_start = max(search_start, hint_byte);
4889
4890         if (!last_ptr)
4891                 empty_cluster = 0;
4892
4893         if (search_start == hint_byte) {
4894 ideal_cache:
4895                 block_group = btrfs_lookup_block_group(root->fs_info,
4896                                                        search_start);
4897                 /*
4898                  * we don't want to use the block group if it doesn't match our
4899                  * allocation bits, or if its not cached.
4900                  *
4901                  * However if we are re-searching with an ideal block group
4902                  * picked out then we don't care that the block group is cached.
4903                  */
4904                 if (block_group && block_group_bits(block_group, data) &&
4905                     (block_group->cached != BTRFS_CACHE_NO ||
4906                      search_start == ideal_cache_offset)) {
4907                         down_read(&space_info->groups_sem);
4908                         if (list_empty(&block_group->list) ||
4909                             block_group->ro) {
4910                                 /*
4911                                  * someone is removing this block group,
4912                                  * we can't jump into the have_block_group
4913                                  * target because our list pointers are not
4914                                  * valid
4915                                  */
4916                                 btrfs_put_block_group(block_group);
4917                                 up_read(&space_info->groups_sem);
4918                         } else {
4919                                 index = get_block_group_index(block_group);
4920                                 goto have_block_group;
4921                         }
4922                 } else if (block_group) {
4923                         btrfs_put_block_group(block_group);
4924                 }
4925         }
4926 search:
4927         down_read(&space_info->groups_sem);
4928         list_for_each_entry(block_group, &space_info->block_groups[index],
4929                             list) {
4930                 u64 offset;
4931                 int cached;
4932
4933                 btrfs_get_block_group(block_group);
4934                 search_start = block_group->key.objectid;
4935
4936 have_block_group:
4937                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4938                         u64 free_percent;
4939
4940                         ret = cache_block_group(block_group, trans, 1);
4941                         if (block_group->cached == BTRFS_CACHE_FINISHED)
4942                                 goto have_block_group;
4943
4944                         free_percent = btrfs_block_group_used(&block_group->item);
4945                         free_percent *= 100;
4946                         free_percent = div64_u64(free_percent,
4947                                                  block_group->key.offset);
4948                         free_percent = 100 - free_percent;
4949                         if (free_percent > ideal_cache_percent &&
4950                             likely(!block_group->ro)) {
4951                                 ideal_cache_offset = block_group->key.objectid;
4952                                 ideal_cache_percent = free_percent;
4953                         }
4954
4955                         /*
4956                          * We only want to start kthread caching if we are at
4957                          * the point where we will wait for caching to make
4958                          * progress, or if our ideal search is over and we've
4959                          * found somebody to start caching.
4960                          */
4961                         if (loop > LOOP_CACHING_NOWAIT ||
4962                             (loop > LOOP_FIND_IDEAL &&
4963                              atomic_read(&space_info->caching_threads) < 2)) {
4964                                 ret = cache_block_group(block_group, trans, 0);
4965                                 BUG_ON(ret);
4966                         }
4967                         found_uncached_bg = true;
4968
4969                         /*
4970                          * If loop is set for cached only, try the next block
4971                          * group.
4972                          */
4973                         if (loop == LOOP_FIND_IDEAL)
4974                                 goto loop;
4975                 }
4976
4977                 cached = block_group_cache_done(block_group);
4978                 if (unlikely(!cached))
4979                         found_uncached_bg = true;
4980
4981                 if (unlikely(block_group->ro))
4982                         goto loop;
4983
4984                 /*
4985                  * Ok we want to try and use the cluster allocator, so lets look
4986                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4987                  * have tried the cluster allocator plenty of times at this
4988                  * point and not have found anything, so we are likely way too
4989                  * fragmented for the clustering stuff to find anything, so lets
4990                  * just skip it and let the allocator find whatever block it can
4991                  * find
4992                  */
4993                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4994                         /*
4995                          * the refill lock keeps out other
4996                          * people trying to start a new cluster
4997                          */
4998                         spin_lock(&last_ptr->refill_lock);
4999                         if (last_ptr->block_group &&
5000                             (last_ptr->block_group->ro ||
5001                             !block_group_bits(last_ptr->block_group, data))) {
5002                                 offset = 0;
5003                                 goto refill_cluster;
5004                         }
5005
5006                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5007                                                  num_bytes, search_start);
5008                         if (offset) {
5009                                 /* we have a block, we're done */
5010                                 spin_unlock(&last_ptr->refill_lock);
5011                                 goto checks;
5012                         }
5013
5014                         spin_lock(&last_ptr->lock);
5015                         /*
5016                          * whoops, this cluster doesn't actually point to
5017                          * this block group.  Get a ref on the block
5018                          * group is does point to and try again
5019                          */
5020                         if (!last_ptr_loop && last_ptr->block_group &&
5021                             last_ptr->block_group != block_group) {
5022
5023                                 btrfs_put_block_group(block_group);
5024                                 block_group = last_ptr->block_group;
5025                                 btrfs_get_block_group(block_group);
5026                                 spin_unlock(&last_ptr->lock);
5027                                 spin_unlock(&last_ptr->refill_lock);
5028
5029                                 last_ptr_loop = 1;
5030                                 search_start = block_group->key.objectid;
5031                                 /*
5032                                  * we know this block group is properly
5033                                  * in the list because
5034                                  * btrfs_remove_block_group, drops the
5035                                  * cluster before it removes the block
5036                                  * group from the list
5037                                  */
5038                                 goto have_block_group;
5039                         }
5040                         spin_unlock(&last_ptr->lock);
5041 refill_cluster:
5042                         /*
5043                          * this cluster didn't work out, free it and
5044                          * start over
5045                          */
5046                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5047
5048                         last_ptr_loop = 0;
5049
5050                         /* allocate a cluster in this block group */
5051                         ret = btrfs_find_space_cluster(trans, root,
5052                                                block_group, last_ptr,
5053                                                offset, num_bytes,
5054                                                empty_cluster + empty_size);
5055                         if (ret == 0) {
5056                                 /*
5057                                  * now pull our allocation out of this
5058                                  * cluster
5059                                  */
5060                                 offset = btrfs_alloc_from_cluster(block_group,
5061                                                   last_ptr, num_bytes,
5062                                                   search_start);
5063                                 if (offset) {
5064                                         /* we found one, proceed */
5065                                         spin_unlock(&last_ptr->refill_lock);
5066                                         goto checks;
5067                                 }
5068                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5069                                    && !failed_cluster_refill) {
5070                                 spin_unlock(&last_ptr->refill_lock);
5071
5072                                 failed_cluster_refill = true;
5073                                 wait_block_group_cache_progress(block_group,
5074                                        num_bytes + empty_cluster + empty_size);
5075                                 goto have_block_group;
5076                         }
5077
5078                         /*
5079                          * at this point we either didn't find a cluster
5080                          * or we weren't able to allocate a block from our
5081                          * cluster.  Free the cluster we've been trying
5082                          * to use, and go to the next block group
5083                          */
5084                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5085                         spin_unlock(&last_ptr->refill_lock);
5086                         goto loop;
5087                 }
5088
5089                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5090                                                     num_bytes, empty_size);
5091                 /*
5092                  * If we didn't find a chunk, and we haven't failed on this
5093                  * block group before, and this block group is in the middle of
5094                  * caching and we are ok with waiting, then go ahead and wait
5095                  * for progress to be made, and set failed_alloc to true.
5096                  *
5097                  * If failed_alloc is true then we've already waited on this
5098                  * block group once and should move on to the next block group.
5099                  */
5100                 if (!offset && !failed_alloc && !cached &&
5101                     loop > LOOP_CACHING_NOWAIT) {
5102                         wait_block_group_cache_progress(block_group,
5103                                                 num_bytes + empty_size);
5104                         failed_alloc = true;
5105                         goto have_block_group;
5106                 } else if (!offset) {
5107                         goto loop;
5108                 }
5109 checks:
5110                 search_start = stripe_align(root, offset);
5111                 /* move on to the next group */
5112                 if (search_start + num_bytes >= search_end) {
5113                         btrfs_add_free_space(block_group, offset, num_bytes);
5114                         goto loop;
5115                 }
5116
5117                 /* move on to the next group */
5118                 if (search_start + num_bytes >
5119                     block_group->key.objectid + block_group->key.offset) {
5120                         btrfs_add_free_space(block_group, offset, num_bytes);
5121                         goto loop;
5122                 }
5123
5124                 ins->objectid = search_start;
5125                 ins->offset = num_bytes;
5126
5127                 if (offset < search_start)
5128                         btrfs_add_free_space(block_group, offset,
5129                                              search_start - offset);
5130                 BUG_ON(offset > search_start);
5131
5132                 ret = update_reserved_bytes(block_group, num_bytes, 1,
5133                                             (data & BTRFS_BLOCK_GROUP_DATA));
5134                 if (ret == -EAGAIN) {
5135                         btrfs_add_free_space(block_group, offset, num_bytes);
5136                         goto loop;
5137                 }
5138
5139                 /* we are all good, lets return */
5140                 ins->objectid = search_start;
5141                 ins->offset = num_bytes;
5142
5143                 if (offset < search_start)
5144                         btrfs_add_free_space(block_group, offset,
5145                                              search_start - offset);
5146                 BUG_ON(offset > search_start);
5147                 break;
5148 loop:
5149                 failed_cluster_refill = false;
5150                 failed_alloc = false;
5151                 BUG_ON(index != get_block_group_index(block_group));
5152                 btrfs_put_block_group(block_group);
5153         }
5154         up_read(&space_info->groups_sem);
5155
5156         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5157                 goto search;
5158
5159         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5160          *                      for them to make caching progress.  Also
5161          *                      determine the best possible bg to cache
5162          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5163          *                      caching kthreads as we move along
5164          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5165          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5166          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5167          *                      again
5168          */
5169         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5170             (found_uncached_bg || empty_size || empty_cluster ||
5171              allowed_chunk_alloc)) {
5172                 index = 0;
5173                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5174                         found_uncached_bg = false;
5175                         loop++;
5176                         if (!ideal_cache_percent &&
5177                             atomic_read(&space_info->caching_threads))
5178                                 goto search;
5179
5180                         /*
5181                          * 1 of the following 2 things have happened so far
5182                          *
5183                          * 1) We found an ideal block group for caching that
5184                          * is mostly full and will cache quickly, so we might
5185                          * as well wait for it.
5186                          *
5187                          * 2) We searched for cached only and we didn't find
5188                          * anything, and we didn't start any caching kthreads
5189                          * either, so chances are we will loop through and
5190                          * start a couple caching kthreads, and then come back
5191                          * around and just wait for them.  This will be slower
5192                          * because we will have 2 caching kthreads reading at
5193                          * the same time when we could have just started one
5194                          * and waited for it to get far enough to give us an
5195                          * allocation, so go ahead and go to the wait caching
5196                          * loop.
5197                          */
5198                         loop = LOOP_CACHING_WAIT;
5199                         search_start = ideal_cache_offset;
5200                         ideal_cache_percent = 0;
5201                         goto ideal_cache;
5202                 } else if (loop == LOOP_FIND_IDEAL) {
5203                         /*
5204                          * Didn't find a uncached bg, wait on anything we find
5205                          * next.
5206                          */
5207                         loop = LOOP_CACHING_WAIT;
5208                         goto search;
5209                 }
5210
5211                 if (loop < LOOP_CACHING_WAIT) {
5212                         loop++;
5213                         goto search;
5214                 }
5215
5216                 if (loop == LOOP_ALLOC_CHUNK) {
5217                         empty_size = 0;
5218                         empty_cluster = 0;
5219                 }
5220
5221                 if (allowed_chunk_alloc) {
5222                         ret = do_chunk_alloc(trans, root, num_bytes +
5223                                              2 * 1024 * 1024, data, 1);
5224                         allowed_chunk_alloc = 0;
5225                         done_chunk_alloc = 1;
5226                 } else if (!done_chunk_alloc) {
5227                         space_info->force_alloc = 1;
5228                 }
5229
5230                 if (loop < LOOP_NO_EMPTY_SIZE) {
5231                         loop++;
5232                         goto search;
5233                 }
5234                 ret = -ENOSPC;
5235         } else if (!ins->objectid) {
5236                 ret = -ENOSPC;
5237         }
5238
5239         /* we found what we needed */
5240         if (ins->objectid) {
5241                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5242                         trans->block_group = block_group->key.objectid;
5243
5244                 btrfs_put_block_group(block_group);
5245                 ret = 0;
5246         }
5247
5248         return ret;
5249 }
5250
5251 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5252                             int dump_block_groups)
5253 {
5254         struct btrfs_block_group_cache *cache;
5255         int index = 0;
5256
5257         spin_lock(&info->lock);
5258         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5259                (unsigned long long)(info->total_bytes - info->bytes_used -
5260                                     info->bytes_pinned - info->bytes_reserved -
5261                                     info->bytes_readonly),
5262                (info->full) ? "" : "not ");
5263         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5264                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5265                (unsigned long long)info->total_bytes,
5266                (unsigned long long)info->bytes_used,
5267                (unsigned long long)info->bytes_pinned,
5268                (unsigned long long)info->bytes_reserved,
5269                (unsigned long long)info->bytes_may_use,
5270                (unsigned long long)info->bytes_readonly);
5271         spin_unlock(&info->lock);
5272
5273         if (!dump_block_groups)
5274                 return;
5275
5276         down_read(&info->groups_sem);
5277 again:
5278         list_for_each_entry(cache, &info->block_groups[index], list) {
5279                 spin_lock(&cache->lock);
5280                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5281                        "%llu pinned %llu reserved\n",
5282                        (unsigned long long)cache->key.objectid,
5283                        (unsigned long long)cache->key.offset,
5284                        (unsigned long long)btrfs_block_group_used(&cache->item),
5285                        (unsigned long long)cache->pinned,
5286                        (unsigned long long)cache->reserved);
5287                 btrfs_dump_free_space(cache, bytes);
5288                 spin_unlock(&cache->lock);
5289         }
5290         if (++index < BTRFS_NR_RAID_TYPES)
5291                 goto again;
5292         up_read(&info->groups_sem);
5293 }
5294
5295 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5296                          struct btrfs_root *root,
5297                          u64 num_bytes, u64 min_alloc_size,
5298                          u64 empty_size, u64 hint_byte,
5299                          u64 search_end, struct btrfs_key *ins,
5300                          u64 data)
5301 {
5302         int ret;
5303         u64 search_start = 0;
5304
5305         data = btrfs_get_alloc_profile(root, data);
5306 again:
5307         /*
5308          * the only place that sets empty_size is btrfs_realloc_node, which
5309          * is not called recursively on allocations
5310          */
5311         if (empty_size || root->ref_cows)
5312                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5313                                      num_bytes + 2 * 1024 * 1024, data, 0);
5314
5315         WARN_ON(num_bytes < root->sectorsize);
5316         ret = find_free_extent(trans, root, num_bytes, empty_size,
5317                                search_start, search_end, hint_byte,
5318                                ins, data);
5319
5320         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5321                 num_bytes = num_bytes >> 1;
5322                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5323                 num_bytes = max(num_bytes, min_alloc_size);
5324                 do_chunk_alloc(trans, root->fs_info->extent_root,
5325                                num_bytes, data, 1);
5326                 goto again;
5327         }
5328         if (ret == -ENOSPC) {
5329                 struct btrfs_space_info *sinfo;
5330
5331                 sinfo = __find_space_info(root->fs_info, data);
5332                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5333                        "wanted %llu\n", (unsigned long long)data,
5334                        (unsigned long long)num_bytes);
5335                 dump_space_info(sinfo, num_bytes, 1);
5336         }
5337
5338         return ret;
5339 }
5340
5341 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5342 {
5343         struct btrfs_block_group_cache *cache;
5344         int ret = 0;
5345
5346         cache = btrfs_lookup_block_group(root->fs_info, start);
5347         if (!cache) {
5348                 printk(KERN_ERR "Unable to find block group for %llu\n",
5349                        (unsigned long long)start);
5350                 return -ENOSPC;
5351         }
5352
5353         ret = btrfs_discard_extent(root, start, len);
5354
5355         btrfs_add_free_space(cache, start, len);
5356         update_reserved_bytes(cache, len, 0, 1);
5357         btrfs_put_block_group(cache);
5358
5359         return ret;
5360 }
5361
5362 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5363                                       struct btrfs_root *root,
5364                                       u64 parent, u64 root_objectid,
5365                                       u64 flags, u64 owner, u64 offset,
5366                                       struct btrfs_key *ins, int ref_mod)
5367 {
5368         int ret;
5369         struct btrfs_fs_info *fs_info = root->fs_info;
5370         struct btrfs_extent_item *extent_item;
5371         struct btrfs_extent_inline_ref *iref;
5372         struct btrfs_path *path;
5373         struct extent_buffer *leaf;
5374         int type;
5375         u32 size;
5376
5377         if (parent > 0)
5378                 type = BTRFS_SHARED_DATA_REF_KEY;
5379         else
5380                 type = BTRFS_EXTENT_DATA_REF_KEY;
5381
5382         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5383
5384         path = btrfs_alloc_path();
5385         BUG_ON(!path);
5386
5387         path->leave_spinning = 1;
5388         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5389                                       ins, size);
5390         BUG_ON(ret);
5391
5392         leaf = path->nodes[0];
5393         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5394                                      struct btrfs_extent_item);
5395         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5396         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5397         btrfs_set_extent_flags(leaf, extent_item,
5398                                flags | BTRFS_EXTENT_FLAG_DATA);
5399
5400         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5401         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5402         if (parent > 0) {
5403                 struct btrfs_shared_data_ref *ref;
5404                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5405                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5406                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5407         } else {
5408                 struct btrfs_extent_data_ref *ref;
5409                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5410                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5411                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5412                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5413                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5414         }
5415
5416         btrfs_mark_buffer_dirty(path->nodes[0]);
5417         btrfs_free_path(path);
5418
5419         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5420         if (ret) {
5421                 printk(KERN_ERR "btrfs update block group failed for %llu "
5422                        "%llu\n", (unsigned long long)ins->objectid,
5423                        (unsigned long long)ins->offset);
5424                 BUG();
5425         }
5426         return ret;
5427 }
5428
5429 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5430                                      struct btrfs_root *root,
5431                                      u64 parent, u64 root_objectid,
5432                                      u64 flags, struct btrfs_disk_key *key,
5433                                      int level, struct btrfs_key *ins)
5434 {
5435         int ret;
5436         struct btrfs_fs_info *fs_info = root->fs_info;
5437         struct btrfs_extent_item *extent_item;
5438         struct btrfs_tree_block_info *block_info;
5439         struct btrfs_extent_inline_ref *iref;
5440         struct btrfs_path *path;
5441         struct extent_buffer *leaf;
5442         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5443
5444         path = btrfs_alloc_path();
5445         BUG_ON(!path);
5446
5447         path->leave_spinning = 1;
5448         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5449                                       ins, size);
5450         BUG_ON(ret);
5451
5452         leaf = path->nodes[0];
5453         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5454                                      struct btrfs_extent_item);
5455         btrfs_set_extent_refs(leaf, extent_item, 1);
5456         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5457         btrfs_set_extent_flags(leaf, extent_item,
5458                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5459         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5460
5461         btrfs_set_tree_block_key(leaf, block_info, key);
5462         btrfs_set_tree_block_level(leaf, block_info, level);
5463
5464         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5465         if (parent > 0) {
5466                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5467                 btrfs_set_extent_inline_ref_type(leaf, iref,
5468                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5469                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5470         } else {
5471                 btrfs_set_extent_inline_ref_type(leaf, iref,
5472                                                  BTRFS_TREE_BLOCK_REF_KEY);
5473                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5474         }
5475
5476         btrfs_mark_buffer_dirty(leaf);
5477         btrfs_free_path(path);
5478
5479         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5480         if (ret) {
5481                 printk(KERN_ERR "btrfs update block group failed for %llu "
5482                        "%llu\n", (unsigned long long)ins->objectid,
5483                        (unsigned long long)ins->offset);
5484                 BUG();
5485         }
5486         return ret;
5487 }
5488
5489 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5490                                      struct btrfs_root *root,
5491                                      u64 root_objectid, u64 owner,
5492                                      u64 offset, struct btrfs_key *ins)
5493 {
5494         int ret;
5495
5496         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5497
5498         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5499                                          0, root_objectid, owner, offset,
5500                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5501         return ret;
5502 }
5503
5504 /*
5505  * this is used by the tree logging recovery code.  It records that
5506  * an extent has been allocated and makes sure to clear the free
5507  * space cache bits as well
5508  */
5509 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5510                                    struct btrfs_root *root,
5511                                    u64 root_objectid, u64 owner, u64 offset,
5512                                    struct btrfs_key *ins)
5513 {
5514         int ret;
5515         struct btrfs_block_group_cache *block_group;
5516         struct btrfs_caching_control *caching_ctl;
5517         u64 start = ins->objectid;
5518         u64 num_bytes = ins->offset;
5519
5520         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5521         cache_block_group(block_group, trans, 0);
5522         caching_ctl = get_caching_control(block_group);
5523
5524         if (!caching_ctl) {
5525                 BUG_ON(!block_group_cache_done(block_group));
5526                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5527                 BUG_ON(ret);
5528         } else {
5529                 mutex_lock(&caching_ctl->mutex);
5530
5531                 if (start >= caching_ctl->progress) {
5532                         ret = add_excluded_extent(root, start, num_bytes);
5533                         BUG_ON(ret);
5534                 } else if (start + num_bytes <= caching_ctl->progress) {
5535                         ret = btrfs_remove_free_space(block_group,
5536                                                       start, num_bytes);
5537                         BUG_ON(ret);
5538                 } else {
5539                         num_bytes = caching_ctl->progress - start;
5540                         ret = btrfs_remove_free_space(block_group,
5541                                                       start, num_bytes);
5542                         BUG_ON(ret);
5543
5544                         start = caching_ctl->progress;
5545                         num_bytes = ins->objectid + ins->offset -
5546                                     caching_ctl->progress;
5547                         ret = add_excluded_extent(root, start, num_bytes);
5548                         BUG_ON(ret);
5549                 }
5550
5551                 mutex_unlock(&caching_ctl->mutex);
5552                 put_caching_control(caching_ctl);
5553         }
5554
5555         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5556         BUG_ON(ret);
5557         btrfs_put_block_group(block_group);
5558         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5559                                          0, owner, offset, ins, 1);
5560         return ret;
5561 }
5562
5563 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5564                                             struct btrfs_root *root,
5565                                             u64 bytenr, u32 blocksize,
5566                                             int level)
5567 {
5568         struct extent_buffer *buf;
5569
5570         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5571         if (!buf)
5572                 return ERR_PTR(-ENOMEM);
5573         btrfs_set_header_generation(buf, trans->transid);
5574         btrfs_set_buffer_lockdep_class(buf, level);
5575         btrfs_tree_lock(buf);
5576         clean_tree_block(trans, root, buf);
5577
5578         btrfs_set_lock_blocking(buf);
5579         btrfs_set_buffer_uptodate(buf);
5580
5581         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5582                 /*
5583                  * we allow two log transactions at a time, use different
5584                  * EXENT bit to differentiate dirty pages.
5585                  */
5586                 if (root->log_transid % 2 == 0)
5587                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5588                                         buf->start + buf->len - 1, GFP_NOFS);
5589                 else
5590                         set_extent_new(&root->dirty_log_pages, buf->start,
5591                                         buf->start + buf->len - 1, GFP_NOFS);
5592         } else {
5593                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5594                          buf->start + buf->len - 1, GFP_NOFS);
5595         }
5596         trans->blocks_used++;
5597         /* this returns a buffer locked for blocking */
5598         return buf;
5599 }
5600
5601 static struct btrfs_block_rsv *
5602 use_block_rsv(struct btrfs_trans_handle *trans,
5603               struct btrfs_root *root, u32 blocksize)
5604 {
5605         struct btrfs_block_rsv *block_rsv;
5606         int ret;
5607
5608         block_rsv = get_block_rsv(trans, root);
5609
5610         if (block_rsv->size == 0) {
5611                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5612                                              blocksize, 0);
5613                 if (ret)
5614                         return ERR_PTR(ret);
5615                 return block_rsv;
5616         }
5617
5618         ret = block_rsv_use_bytes(block_rsv, blocksize);
5619         if (!ret)
5620                 return block_rsv;
5621
5622         return ERR_PTR(-ENOSPC);
5623 }
5624
5625 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5626 {
5627         block_rsv_add_bytes(block_rsv, blocksize, 0);
5628         block_rsv_release_bytes(block_rsv, NULL, 0);
5629 }
5630
5631 /*
5632  * finds a free extent and does all the dirty work required for allocation
5633  * returns the key for the extent through ins, and a tree buffer for
5634  * the first block of the extent through buf.
5635  *
5636  * returns the tree buffer or NULL.
5637  */
5638 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5639                                         struct btrfs_root *root, u32 blocksize,
5640                                         u64 parent, u64 root_objectid,
5641                                         struct btrfs_disk_key *key, int level,
5642                                         u64 hint, u64 empty_size)
5643 {
5644         struct btrfs_key ins;
5645         struct btrfs_block_rsv *block_rsv;
5646         struct extent_buffer *buf;
5647         u64 flags = 0;
5648         int ret;
5649
5650
5651         block_rsv = use_block_rsv(trans, root, blocksize);
5652         if (IS_ERR(block_rsv))
5653                 return ERR_CAST(block_rsv);
5654
5655         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5656                                    empty_size, hint, (u64)-1, &ins, 0);
5657         if (ret) {
5658                 unuse_block_rsv(block_rsv, blocksize);
5659                 return ERR_PTR(ret);
5660         }
5661
5662         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5663                                     blocksize, level);
5664         BUG_ON(IS_ERR(buf));
5665
5666         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5667                 if (parent == 0)
5668                         parent = ins.objectid;
5669                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5670         } else
5671                 BUG_ON(parent > 0);
5672
5673         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5674                 struct btrfs_delayed_extent_op *extent_op;
5675                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5676                 BUG_ON(!extent_op);
5677                 if (key)
5678                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5679                 else
5680                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5681                 extent_op->flags_to_set = flags;
5682                 extent_op->update_key = 1;
5683                 extent_op->update_flags = 1;
5684                 extent_op->is_data = 0;
5685
5686                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5687                                         ins.offset, parent, root_objectid,
5688                                         level, BTRFS_ADD_DELAYED_EXTENT,
5689                                         extent_op);
5690                 BUG_ON(ret);
5691         }
5692         return buf;
5693 }
5694
5695 struct walk_control {
5696         u64 refs[BTRFS_MAX_LEVEL];
5697         u64 flags[BTRFS_MAX_LEVEL];
5698         struct btrfs_key update_progress;
5699         int stage;
5700         int level;
5701         int shared_level;
5702         int update_ref;
5703         int keep_locks;
5704         int reada_slot;
5705         int reada_count;
5706 };
5707
5708 #define DROP_REFERENCE  1
5709 #define UPDATE_BACKREF  2
5710
5711 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5712                                      struct btrfs_root *root,
5713                                      struct walk_control *wc,
5714                                      struct btrfs_path *path)
5715 {
5716         u64 bytenr;
5717         u64 generation;
5718         u64 refs;
5719         u64 flags;
5720         u32 nritems;
5721         u32 blocksize;
5722         struct btrfs_key key;
5723         struct extent_buffer *eb;
5724         int ret;
5725         int slot;
5726         int nread = 0;
5727
5728         if (path->slots[wc->level] < wc->reada_slot) {
5729                 wc->reada_count = wc->reada_count * 2 / 3;
5730                 wc->reada_count = max(wc->reada_count, 2);
5731         } else {
5732                 wc->reada_count = wc->reada_count * 3 / 2;
5733                 wc->reada_count = min_t(int, wc->reada_count,
5734                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5735         }
5736
5737         eb = path->nodes[wc->level];
5738         nritems = btrfs_header_nritems(eb);
5739         blocksize = btrfs_level_size(root, wc->level - 1);
5740
5741         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5742                 if (nread >= wc->reada_count)
5743                         break;
5744
5745                 cond_resched();
5746                 bytenr = btrfs_node_blockptr(eb, slot);
5747                 generation = btrfs_node_ptr_generation(eb, slot);
5748
5749                 if (slot == path->slots[wc->level])
5750                         goto reada;
5751
5752                 if (wc->stage == UPDATE_BACKREF &&
5753                     generation <= root->root_key.offset)
5754                         continue;
5755
5756                 /* We don't lock the tree block, it's OK to be racy here */
5757                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5758                                                &refs, &flags);
5759                 BUG_ON(ret);
5760                 BUG_ON(refs == 0);
5761
5762                 if (wc->stage == DROP_REFERENCE) {
5763                         if (refs == 1)
5764                                 goto reada;
5765
5766                         if (wc->level == 1 &&
5767                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5768                                 continue;
5769                         if (!wc->update_ref ||
5770                             generation <= root->root_key.offset)
5771                                 continue;
5772                         btrfs_node_key_to_cpu(eb, &key, slot);
5773                         ret = btrfs_comp_cpu_keys(&key,
5774                                                   &wc->update_progress);
5775                         if (ret < 0)
5776                                 continue;
5777                 } else {
5778                         if (wc->level == 1 &&
5779                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5780                                 continue;
5781                 }
5782 reada:
5783                 ret = readahead_tree_block(root, bytenr, blocksize,
5784                                            generation);
5785                 if (ret)
5786                         break;
5787                 nread++;
5788         }
5789         wc->reada_slot = slot;
5790 }
5791
5792 /*
5793  * hepler to process tree block while walking down the tree.
5794  *
5795  * when wc->stage == UPDATE_BACKREF, this function updates
5796  * back refs for pointers in the block.
5797  *
5798  * NOTE: return value 1 means we should stop walking down.
5799  */
5800 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5801                                    struct btrfs_root *root,
5802                                    struct btrfs_path *path,
5803                                    struct walk_control *wc, int lookup_info)
5804 {
5805         int level = wc->level;
5806         struct extent_buffer *eb = path->nodes[level];
5807         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5808         int ret;
5809
5810         if (wc->stage == UPDATE_BACKREF &&
5811             btrfs_header_owner(eb) != root->root_key.objectid)
5812                 return 1;
5813
5814         /*
5815          * when reference count of tree block is 1, it won't increase
5816          * again. once full backref flag is set, we never clear it.
5817          */
5818         if (lookup_info &&
5819             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5820              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5821                 BUG_ON(!path->locks[level]);
5822                 ret = btrfs_lookup_extent_info(trans, root,
5823                                                eb->start, eb->len,
5824                                                &wc->refs[level],
5825                                                &wc->flags[level]);
5826                 BUG_ON(ret);
5827                 BUG_ON(wc->refs[level] == 0);
5828         }
5829
5830         if (wc->stage == DROP_REFERENCE) {
5831                 if (wc->refs[level] > 1)
5832                         return 1;
5833
5834                 if (path->locks[level] && !wc->keep_locks) {
5835                         btrfs_tree_unlock(eb);
5836                         path->locks[level] = 0;
5837                 }
5838                 return 0;
5839         }
5840
5841         /* wc->stage == UPDATE_BACKREF */
5842         if (!(wc->flags[level] & flag)) {
5843                 BUG_ON(!path->locks[level]);
5844                 ret = btrfs_inc_ref(trans, root, eb, 1);
5845                 BUG_ON(ret);
5846                 ret = btrfs_dec_ref(trans, root, eb, 0);
5847                 BUG_ON(ret);
5848                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5849                                                   eb->len, flag, 0);
5850                 BUG_ON(ret);
5851                 wc->flags[level] |= flag;
5852         }
5853
5854         /*
5855          * the block is shared by multiple trees, so it's not good to
5856          * keep the tree lock
5857          */
5858         if (path->locks[level] && level > 0) {
5859                 btrfs_tree_unlock(eb);
5860                 path->locks[level] = 0;
5861         }
5862         return 0;
5863 }
5864
5865 /*
5866  * hepler to process tree block pointer.
5867  *
5868  * when wc->stage == DROP_REFERENCE, this function checks
5869  * reference count of the block pointed to. if the block
5870  * is shared and we need update back refs for the subtree
5871  * rooted at the block, this function changes wc->stage to
5872  * UPDATE_BACKREF. if the block is shared and there is no
5873  * need to update back, this function drops the reference
5874  * to the block.
5875  *
5876  * NOTE: return value 1 means we should stop walking down.
5877  */
5878 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5879                                  struct btrfs_root *root,
5880                                  struct btrfs_path *path,
5881                                  struct walk_control *wc, int *lookup_info)
5882 {
5883         u64 bytenr;
5884         u64 generation;
5885         u64 parent;
5886         u32 blocksize;
5887         struct btrfs_key key;
5888         struct extent_buffer *next;
5889         int level = wc->level;
5890         int reada = 0;
5891         int ret = 0;
5892
5893         generation = btrfs_node_ptr_generation(path->nodes[level],
5894                                                path->slots[level]);
5895         /*
5896          * if the lower level block was created before the snapshot
5897          * was created, we know there is no need to update back refs
5898          * for the subtree
5899          */
5900         if (wc->stage == UPDATE_BACKREF &&
5901             generation <= root->root_key.offset) {
5902                 *lookup_info = 1;
5903                 return 1;
5904         }
5905
5906         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5907         blocksize = btrfs_level_size(root, level - 1);
5908
5909         next = btrfs_find_tree_block(root, bytenr, blocksize);
5910         if (!next) {
5911                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5912                 if (!next)
5913                         return -ENOMEM;
5914                 reada = 1;
5915         }
5916         btrfs_tree_lock(next);
5917         btrfs_set_lock_blocking(next);
5918
5919         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5920                                        &wc->refs[level - 1],
5921                                        &wc->flags[level - 1]);
5922         BUG_ON(ret);
5923         BUG_ON(wc->refs[level - 1] == 0);
5924         *lookup_info = 0;
5925
5926         if (wc->stage == DROP_REFERENCE) {
5927                 if (wc->refs[level - 1] > 1) {
5928                         if (level == 1 &&
5929                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5930                                 goto skip;
5931
5932                         if (!wc->update_ref ||
5933                             generation <= root->root_key.offset)
5934                                 goto skip;
5935
5936                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5937                                               path->slots[level]);
5938                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5939                         if (ret < 0)
5940                                 goto skip;
5941
5942                         wc->stage = UPDATE_BACKREF;
5943                         wc->shared_level = level - 1;
5944                 }
5945         } else {
5946                 if (level == 1 &&
5947                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5948                         goto skip;
5949         }
5950
5951         if (!btrfs_buffer_uptodate(next, generation)) {
5952                 btrfs_tree_unlock(next);
5953                 free_extent_buffer(next);
5954                 next = NULL;
5955                 *lookup_info = 1;
5956         }
5957
5958         if (!next) {
5959                 if (reada && level == 1)
5960                         reada_walk_down(trans, root, wc, path);
5961                 next = read_tree_block(root, bytenr, blocksize, generation);
5962                 btrfs_tree_lock(next);
5963                 btrfs_set_lock_blocking(next);
5964         }
5965
5966         level--;
5967         BUG_ON(level != btrfs_header_level(next));
5968         path->nodes[level] = next;
5969         path->slots[level] = 0;
5970         path->locks[level] = 1;
5971         wc->level = level;
5972         if (wc->level == 1)
5973                 wc->reada_slot = 0;
5974         return 0;
5975 skip:
5976         wc->refs[level - 1] = 0;
5977         wc->flags[level - 1] = 0;
5978         if (wc->stage == DROP_REFERENCE) {
5979                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5980                         parent = path->nodes[level]->start;
5981                 } else {
5982                         BUG_ON(root->root_key.objectid !=
5983                                btrfs_header_owner(path->nodes[level]));
5984                         parent = 0;
5985                 }
5986
5987                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5988                                         root->root_key.objectid, level - 1, 0);
5989                 BUG_ON(ret);
5990         }
5991         btrfs_tree_unlock(next);
5992         free_extent_buffer(next);
5993         *lookup_info = 1;
5994         return 1;
5995 }
5996
5997 /*
5998  * hepler to process tree block while walking up the tree.
5999  *
6000  * when wc->stage == DROP_REFERENCE, this function drops
6001  * reference count on the block.
6002  *
6003  * when wc->stage == UPDATE_BACKREF, this function changes
6004  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6005  * to UPDATE_BACKREF previously while processing the block.
6006  *
6007  * NOTE: return value 1 means we should stop walking up.
6008  */
6009 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6010                                  struct btrfs_root *root,
6011                                  struct btrfs_path *path,
6012                                  struct walk_control *wc)
6013 {
6014         int ret;
6015         int level = wc->level;
6016         struct extent_buffer *eb = path->nodes[level];
6017         u64 parent = 0;
6018
6019         if (wc->stage == UPDATE_BACKREF) {
6020                 BUG_ON(wc->shared_level < level);
6021                 if (level < wc->shared_level)
6022                         goto out;
6023
6024                 ret = find_next_key(path, level + 1, &wc->update_progress);
6025                 if (ret > 0)
6026                         wc->update_ref = 0;
6027
6028                 wc->stage = DROP_REFERENCE;
6029                 wc->shared_level = -1;
6030                 path->slots[level] = 0;
6031
6032                 /*
6033                  * check reference count again if the block isn't locked.
6034                  * we should start walking down the tree again if reference
6035                  * count is one.
6036                  */
6037                 if (!path->locks[level]) {
6038                         BUG_ON(level == 0);
6039                         btrfs_tree_lock(eb);
6040                         btrfs_set_lock_blocking(eb);
6041                         path->locks[level] = 1;
6042
6043                         ret = btrfs_lookup_extent_info(trans, root,
6044                                                        eb->start, eb->len,
6045                                                        &wc->refs[level],
6046                                                        &wc->flags[level]);
6047                         BUG_ON(ret);
6048                         BUG_ON(wc->refs[level] == 0);
6049                         if (wc->refs[level] == 1) {
6050                                 btrfs_tree_unlock(eb);
6051                                 path->locks[level] = 0;
6052                                 return 1;
6053                         }
6054                 }
6055         }
6056
6057         /* wc->stage == DROP_REFERENCE */
6058         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6059
6060         if (wc->refs[level] == 1) {
6061                 if (level == 0) {
6062                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6063                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6064                         else
6065                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6066                         BUG_ON(ret);
6067                 }
6068                 /* make block locked assertion in clean_tree_block happy */
6069                 if (!path->locks[level] &&
6070                     btrfs_header_generation(eb) == trans->transid) {
6071                         btrfs_tree_lock(eb);
6072                         btrfs_set_lock_blocking(eb);
6073                         path->locks[level] = 1;
6074                 }
6075                 clean_tree_block(trans, root, eb);
6076         }
6077
6078         if (eb == root->node) {
6079                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6080                         parent = eb->start;
6081                 else
6082                         BUG_ON(root->root_key.objectid !=
6083                                btrfs_header_owner(eb));
6084         } else {
6085                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6086                         parent = path->nodes[level + 1]->start;
6087                 else
6088                         BUG_ON(root->root_key.objectid !=
6089                                btrfs_header_owner(path->nodes[level + 1]));
6090         }
6091
6092         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6093 out:
6094         wc->refs[level] = 0;
6095         wc->flags[level] = 0;
6096         return 0;
6097 }
6098
6099 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6100                                    struct btrfs_root *root,
6101                                    struct btrfs_path *path,
6102                                    struct walk_control *wc)
6103 {
6104         int level = wc->level;
6105         int lookup_info = 1;
6106         int ret;
6107
6108         while (level >= 0) {
6109                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6110                 if (ret > 0)
6111                         break;
6112
6113                 if (level == 0)
6114                         break;
6115
6116                 if (path->slots[level] >=
6117                     btrfs_header_nritems(path->nodes[level]))
6118                         break;
6119
6120                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6121                 if (ret > 0) {
6122                         path->slots[level]++;
6123                         continue;
6124                 } else if (ret < 0)
6125                         return ret;
6126                 level = wc->level;
6127         }
6128         return 0;
6129 }
6130
6131 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6132                                  struct btrfs_root *root,
6133                                  struct btrfs_path *path,
6134                                  struct walk_control *wc, int max_level)
6135 {
6136         int level = wc->level;
6137         int ret;
6138
6139         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6140         while (level < max_level && path->nodes[level]) {
6141                 wc->level = level;
6142                 if (path->slots[level] + 1 <
6143                     btrfs_header_nritems(path->nodes[level])) {
6144                         path->slots[level]++;
6145                         return 0;
6146                 } else {
6147                         ret = walk_up_proc(trans, root, path, wc);
6148                         if (ret > 0)
6149                                 return 0;
6150
6151                         if (path->locks[level]) {
6152                                 btrfs_tree_unlock(path->nodes[level]);
6153                                 path->locks[level] = 0;
6154                         }
6155                         free_extent_buffer(path->nodes[level]);
6156                         path->nodes[level] = NULL;
6157                         level++;
6158                 }
6159         }
6160         return 1;
6161 }
6162
6163 /*
6164  * drop a subvolume tree.
6165  *
6166  * this function traverses the tree freeing any blocks that only
6167  * referenced by the tree.
6168  *
6169  * when a shared tree block is found. this function decreases its
6170  * reference count by one. if update_ref is true, this function
6171  * also make sure backrefs for the shared block and all lower level
6172  * blocks are properly updated.
6173  */
6174 int btrfs_drop_snapshot(struct btrfs_root *root,
6175                         struct btrfs_block_rsv *block_rsv, int update_ref)
6176 {
6177         struct btrfs_path *path;
6178         struct btrfs_trans_handle *trans;
6179         struct btrfs_root *tree_root = root->fs_info->tree_root;
6180         struct btrfs_root_item *root_item = &root->root_item;
6181         struct walk_control *wc;
6182         struct btrfs_key key;
6183         int err = 0;
6184         int ret;
6185         int level;
6186
6187         path = btrfs_alloc_path();
6188         BUG_ON(!path);
6189
6190         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6191         BUG_ON(!wc);
6192
6193         trans = btrfs_start_transaction(tree_root, 0);
6194         if (block_rsv)
6195                 trans->block_rsv = block_rsv;
6196
6197         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6198                 level = btrfs_header_level(root->node);
6199                 path->nodes[level] = btrfs_lock_root_node(root);
6200                 btrfs_set_lock_blocking(path->nodes[level]);
6201                 path->slots[level] = 0;
6202                 path->locks[level] = 1;
6203                 memset(&wc->update_progress, 0,
6204                        sizeof(wc->update_progress));
6205         } else {
6206                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6207                 memcpy(&wc->update_progress, &key,
6208                        sizeof(wc->update_progress));
6209
6210                 level = root_item->drop_level;
6211                 BUG_ON(level == 0);
6212                 path->lowest_level = level;
6213                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6214                 path->lowest_level = 0;
6215                 if (ret < 0) {
6216                         err = ret;
6217                         goto out;
6218                 }
6219                 WARN_ON(ret > 0);
6220
6221                 /*
6222                  * unlock our path, this is safe because only this
6223                  * function is allowed to delete this snapshot
6224                  */
6225                 btrfs_unlock_up_safe(path, 0);
6226
6227                 level = btrfs_header_level(root->node);
6228                 while (1) {
6229                         btrfs_tree_lock(path->nodes[level]);
6230                         btrfs_set_lock_blocking(path->nodes[level]);
6231
6232                         ret = btrfs_lookup_extent_info(trans, root,
6233                                                 path->nodes[level]->start,
6234                                                 path->nodes[level]->len,
6235                                                 &wc->refs[level],
6236                                                 &wc->flags[level]);
6237                         BUG_ON(ret);
6238                         BUG_ON(wc->refs[level] == 0);
6239
6240                         if (level == root_item->drop_level)
6241                                 break;
6242
6243                         btrfs_tree_unlock(path->nodes[level]);
6244                         WARN_ON(wc->refs[level] != 1);
6245                         level--;
6246                 }
6247         }
6248
6249         wc->level = level;
6250         wc->shared_level = -1;
6251         wc->stage = DROP_REFERENCE;
6252         wc->update_ref = update_ref;
6253         wc->keep_locks = 0;
6254         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6255
6256         while (1) {
6257                 ret = walk_down_tree(trans, root, path, wc);
6258                 if (ret < 0) {
6259                         err = ret;
6260                         break;
6261                 }
6262
6263                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6264                 if (ret < 0) {
6265                         err = ret;
6266                         break;
6267                 }
6268
6269                 if (ret > 0) {
6270                         BUG_ON(wc->stage != DROP_REFERENCE);
6271                         break;
6272                 }
6273
6274                 if (wc->stage == DROP_REFERENCE) {
6275                         level = wc->level;
6276                         btrfs_node_key(path->nodes[level],
6277                                        &root_item->drop_progress,
6278                                        path->slots[level]);
6279                         root_item->drop_level = level;
6280                 }
6281
6282                 BUG_ON(wc->level == 0);
6283                 if (btrfs_should_end_transaction(trans, tree_root)) {
6284                         ret = btrfs_update_root(trans, tree_root,
6285                                                 &root->root_key,
6286                                                 root_item);
6287                         BUG_ON(ret);
6288
6289                         btrfs_end_transaction_throttle(trans, tree_root);
6290                         trans = btrfs_start_transaction(tree_root, 0);
6291                         if (block_rsv)
6292                                 trans->block_rsv = block_rsv;
6293                 }
6294         }
6295         btrfs_release_path(root, path);
6296         BUG_ON(err);
6297
6298         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6299         BUG_ON(ret);
6300
6301         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6302                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6303                                            NULL, NULL);
6304                 BUG_ON(ret < 0);
6305                 if (ret > 0) {
6306                         ret = btrfs_del_orphan_item(trans, tree_root,
6307                                                     root->root_key.objectid);
6308                         BUG_ON(ret);
6309                 }
6310         }
6311
6312         if (root->in_radix) {
6313                 btrfs_free_fs_root(tree_root->fs_info, root);
6314         } else {
6315                 free_extent_buffer(root->node);
6316                 free_extent_buffer(root->commit_root);
6317                 kfree(root);
6318         }
6319 out:
6320         btrfs_end_transaction_throttle(trans, tree_root);
6321         kfree(wc);
6322         btrfs_free_path(path);
6323         return err;
6324 }
6325
6326 /*
6327  * drop subtree rooted at tree block 'node'.
6328  *
6329  * NOTE: this function will unlock and release tree block 'node'
6330  */
6331 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6332                         struct btrfs_root *root,
6333                         struct extent_buffer *node,
6334                         struct extent_buffer *parent)
6335 {
6336         struct btrfs_path *path;
6337         struct walk_control *wc;
6338         int level;
6339         int parent_level;
6340         int ret = 0;
6341         int wret;
6342
6343         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6344
6345         path = btrfs_alloc_path();
6346         BUG_ON(!path);
6347
6348         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6349         BUG_ON(!wc);
6350
6351         btrfs_assert_tree_locked(parent);
6352         parent_level = btrfs_header_level(parent);
6353         extent_buffer_get(parent);
6354         path->nodes[parent_level] = parent;
6355         path->slots[parent_level] = btrfs_header_nritems(parent);
6356
6357         btrfs_assert_tree_locked(node);
6358         level = btrfs_header_level(node);
6359         path->nodes[level] = node;
6360         path->slots[level] = 0;
6361         path->locks[level] = 1;
6362
6363         wc->refs[parent_level] = 1;
6364         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6365         wc->level = level;
6366         wc->shared_level = -1;
6367         wc->stage = DROP_REFERENCE;
6368         wc->update_ref = 0;
6369         wc->keep_locks = 1;
6370         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6371
6372         while (1) {
6373                 wret = walk_down_tree(trans, root, path, wc);
6374                 if (wret < 0) {
6375                         ret = wret;
6376                         break;
6377                 }
6378
6379                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6380                 if (wret < 0)
6381                         ret = wret;
6382                 if (wret != 0)
6383                         break;
6384         }
6385
6386         kfree(wc);
6387         btrfs_free_path(path);
6388         return ret;
6389 }
6390
6391 #if 0
6392 static unsigned long calc_ra(unsigned long start, unsigned long last,
6393                              unsigned long nr)
6394 {
6395         return min(last, start + nr - 1);
6396 }
6397
6398 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6399                                          u64 len)
6400 {
6401         u64 page_start;
6402         u64 page_end;
6403         unsigned long first_index;
6404         unsigned long last_index;
6405         unsigned long i;
6406         struct page *page;
6407         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6408         struct file_ra_state *ra;
6409         struct btrfs_ordered_extent *ordered;
6410         unsigned int total_read = 0;
6411         unsigned int total_dirty = 0;
6412         int ret = 0;
6413
6414         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6415
6416         mutex_lock(&inode->i_mutex);
6417         first_index = start >> PAGE_CACHE_SHIFT;
6418         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6419
6420         /* make sure the dirty trick played by the caller work */
6421         ret = invalidate_inode_pages2_range(inode->i_mapping,
6422                                             first_index, last_index);
6423         if (ret)
6424                 goto out_unlock;
6425
6426         file_ra_state_init(ra, inode->i_mapping);
6427
6428         for (i = first_index ; i <= last_index; i++) {
6429                 if (total_read % ra->ra_pages == 0) {
6430                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6431                                        calc_ra(i, last_index, ra->ra_pages));
6432                 }
6433                 total_read++;
6434 again:
6435                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6436                         BUG_ON(1);
6437                 page = grab_cache_page(inode->i_mapping, i);
6438                 if (!page) {
6439                         ret = -ENOMEM;
6440                         goto out_unlock;
6441                 }
6442                 if (!PageUptodate(page)) {
6443                         btrfs_readpage(NULL, page);
6444                         lock_page(page);
6445                         if (!PageUptodate(page)) {
6446                                 unlock_page(page);
6447                                 page_cache_release(page);
6448                                 ret = -EIO;
6449                                 goto out_unlock;
6450                         }
6451                 }
6452                 wait_on_page_writeback(page);
6453
6454                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6455                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6456                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6457
6458                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6459                 if (ordered) {
6460                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6461                         unlock_page(page);
6462                         page_cache_release(page);
6463                         btrfs_start_ordered_extent(inode, ordered, 1);
6464                         btrfs_put_ordered_extent(ordered);
6465                         goto again;
6466                 }
6467                 set_page_extent_mapped(page);
6468
6469                 if (i == first_index)
6470                         set_extent_bits(io_tree, page_start, page_end,
6471                                         EXTENT_BOUNDARY, GFP_NOFS);
6472                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6473
6474                 set_page_dirty(page);
6475                 total_dirty++;
6476
6477                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6478                 unlock_page(page);
6479                 page_cache_release(page);
6480         }
6481
6482 out_unlock:
6483         kfree(ra);
6484         mutex_unlock(&inode->i_mutex);
6485         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6486         return ret;
6487 }
6488
6489 static noinline int relocate_data_extent(struct inode *reloc_inode,
6490                                          struct btrfs_key *extent_key,
6491                                          u64 offset)
6492 {
6493         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6494         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6495         struct extent_map *em;
6496         u64 start = extent_key->objectid - offset;
6497         u64 end = start + extent_key->offset - 1;
6498
6499         em = alloc_extent_map(GFP_NOFS);
6500         BUG_ON(!em || IS_ERR(em));
6501
6502         em->start = start;
6503         em->len = extent_key->offset;
6504         em->block_len = extent_key->offset;
6505         em->block_start = extent_key->objectid;
6506         em->bdev = root->fs_info->fs_devices->latest_bdev;
6507         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6508
6509         /* setup extent map to cheat btrfs_readpage */
6510         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6511         while (1) {
6512                 int ret;
6513                 write_lock(&em_tree->lock);
6514                 ret = add_extent_mapping(em_tree, em);
6515                 write_unlock(&em_tree->lock);
6516                 if (ret != -EEXIST) {
6517                         free_extent_map(em);
6518                         break;
6519                 }
6520                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6521         }
6522         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6523
6524         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6525 }
6526
6527 struct btrfs_ref_path {
6528         u64 extent_start;
6529         u64 nodes[BTRFS_MAX_LEVEL];
6530         u64 root_objectid;
6531         u64 root_generation;
6532         u64 owner_objectid;
6533         u32 num_refs;
6534         int lowest_level;
6535         int current_level;
6536         int shared_level;
6537
6538         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6539         u64 new_nodes[BTRFS_MAX_LEVEL];
6540 };
6541
6542 struct disk_extent {
6543         u64 ram_bytes;
6544         u64 disk_bytenr;
6545         u64 disk_num_bytes;
6546         u64 offset;
6547         u64 num_bytes;
6548         u8 compression;
6549         u8 encryption;
6550         u16 other_encoding;
6551 };
6552
6553 static int is_cowonly_root(u64 root_objectid)
6554 {
6555         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6556             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6557             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6558             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6559             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6560             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6561                 return 1;
6562         return 0;
6563 }
6564
6565 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6566                                     struct btrfs_root *extent_root,
6567                                     struct btrfs_ref_path *ref_path,
6568                                     int first_time)
6569 {
6570         struct extent_buffer *leaf;
6571         struct btrfs_path *path;
6572         struct btrfs_extent_ref *ref;
6573         struct btrfs_key key;
6574         struct btrfs_key found_key;
6575         u64 bytenr;
6576         u32 nritems;
6577         int level;
6578         int ret = 1;
6579
6580         path = btrfs_alloc_path();
6581         if (!path)
6582                 return -ENOMEM;
6583
6584         if (first_time) {
6585                 ref_path->lowest_level = -1;
6586                 ref_path->current_level = -1;
6587                 ref_path->shared_level = -1;
6588                 goto walk_up;
6589         }
6590 walk_down:
6591         level = ref_path->current_level - 1;
6592         while (level >= -1) {
6593                 u64 parent;
6594                 if (level < ref_path->lowest_level)
6595                         break;
6596
6597                 if (level >= 0)
6598                         bytenr = ref_path->nodes[level];
6599                 else
6600                         bytenr = ref_path->extent_start;
6601                 BUG_ON(bytenr == 0);
6602
6603                 parent = ref_path->nodes[level + 1];
6604                 ref_path->nodes[level + 1] = 0;
6605                 ref_path->current_level = level;
6606                 BUG_ON(parent == 0);
6607
6608                 key.objectid = bytenr;
6609                 key.offset = parent + 1;
6610                 key.type = BTRFS_EXTENT_REF_KEY;
6611
6612                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6613                 if (ret < 0)
6614                         goto out;
6615                 BUG_ON(ret == 0);
6616
6617                 leaf = path->nodes[0];
6618                 nritems = btrfs_header_nritems(leaf);
6619                 if (path->slots[0] >= nritems) {
6620                         ret = btrfs_next_leaf(extent_root, path);
6621                         if (ret < 0)
6622                                 goto out;
6623                         if (ret > 0)
6624                                 goto next;
6625                         leaf = path->nodes[0];
6626                 }
6627
6628                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6629                 if (found_key.objectid == bytenr &&
6630                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6631                         if (level < ref_path->shared_level)
6632                                 ref_path->shared_level = level;
6633                         goto found;
6634                 }
6635 next:
6636                 level--;
6637                 btrfs_release_path(extent_root, path);
6638                 cond_resched();
6639         }
6640         /* reached lowest level */
6641         ret = 1;
6642         goto out;
6643 walk_up:
6644         level = ref_path->current_level;
6645         while (level < BTRFS_MAX_LEVEL - 1) {
6646                 u64 ref_objectid;
6647
6648                 if (level >= 0)
6649                         bytenr = ref_path->nodes[level];
6650                 else
6651                         bytenr = ref_path->extent_start;
6652
6653                 BUG_ON(bytenr == 0);
6654
6655                 key.objectid = bytenr;
6656                 key.offset = 0;
6657                 key.type = BTRFS_EXTENT_REF_KEY;
6658
6659                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6660                 if (ret < 0)
6661                         goto out;
6662
6663                 leaf = path->nodes[0];
6664                 nritems = btrfs_header_nritems(leaf);
6665                 if (path->slots[0] >= nritems) {
6666                         ret = btrfs_next_leaf(extent_root, path);
6667                         if (ret < 0)
6668                                 goto out;
6669                         if (ret > 0) {
6670                                 /* the extent was freed by someone */
6671                                 if (ref_path->lowest_level == level)
6672                                         goto out;
6673                                 btrfs_release_path(extent_root, path);
6674                                 goto walk_down;
6675                         }
6676                         leaf = path->nodes[0];
6677                 }
6678
6679                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6680                 if (found_key.objectid != bytenr ||
6681                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6682                         /* the extent was freed by someone */
6683                         if (ref_path->lowest_level == level) {
6684                                 ret = 1;
6685                                 goto out;
6686                         }
6687                         btrfs_release_path(extent_root, path);
6688                         goto walk_down;
6689                 }
6690 found:
6691                 ref = btrfs_item_ptr(leaf, path->slots[0],
6692                                 struct btrfs_extent_ref);
6693                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6694                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6695                         if (first_time) {
6696                                 level = (int)ref_objectid;
6697                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6698                                 ref_path->lowest_level = level;
6699                                 ref_path->current_level = level;
6700                                 ref_path->nodes[level] = bytenr;
6701                         } else {
6702                                 WARN_ON(ref_objectid != level);
6703                         }
6704                 } else {
6705                         WARN_ON(level != -1);
6706                 }
6707                 first_time = 0;
6708
6709                 if (ref_path->lowest_level == level) {
6710                         ref_path->owner_objectid = ref_objectid;
6711                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6712                 }
6713
6714                 /*
6715                  * the block is tree root or the block isn't in reference
6716                  * counted tree.
6717                  */
6718                 if (found_key.objectid == found_key.offset ||
6719                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6720                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6721                         ref_path->root_generation =
6722                                 btrfs_ref_generation(leaf, ref);
6723                         if (level < 0) {
6724                                 /* special reference from the tree log */
6725                                 ref_path->nodes[0] = found_key.offset;
6726                                 ref_path->current_level = 0;
6727                         }
6728                         ret = 0;
6729                         goto out;
6730                 }
6731
6732                 level++;
6733                 BUG_ON(ref_path->nodes[level] != 0);
6734                 ref_path->nodes[level] = found_key.offset;
6735                 ref_path->current_level = level;
6736
6737                 /*
6738                  * the reference was created in the running transaction,
6739                  * no need to continue walking up.
6740                  */
6741                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6742                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6743                         ref_path->root_generation =
6744                                 btrfs_ref_generation(leaf, ref);
6745                         ret = 0;
6746                         goto out;
6747                 }
6748
6749                 btrfs_release_path(extent_root, path);
6750                 cond_resched();
6751         }
6752         /* reached max tree level, but no tree root found. */
6753         BUG();
6754 out:
6755         btrfs_free_path(path);
6756         return ret;
6757 }
6758
6759 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6760                                 struct btrfs_root *extent_root,
6761                                 struct btrfs_ref_path *ref_path,
6762                                 u64 extent_start)
6763 {
6764         memset(ref_path, 0, sizeof(*ref_path));
6765         ref_path->extent_start = extent_start;
6766
6767         return __next_ref_path(trans, extent_root, ref_path, 1);
6768 }
6769
6770 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6771                                struct btrfs_root *extent_root,
6772                                struct btrfs_ref_path *ref_path)
6773 {
6774         return __next_ref_path(trans, extent_root, ref_path, 0);
6775 }
6776
6777 static noinline int get_new_locations(struct inode *reloc_inode,
6778                                       struct btrfs_key *extent_key,
6779                                       u64 offset, int no_fragment,
6780                                       struct disk_extent **extents,
6781                                       int *nr_extents)
6782 {
6783         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6784         struct btrfs_path *path;
6785         struct btrfs_file_extent_item *fi;
6786         struct extent_buffer *leaf;
6787         struct disk_extent *exts = *extents;
6788         struct btrfs_key found_key;
6789         u64 cur_pos;
6790         u64 last_byte;
6791         u32 nritems;
6792         int nr = 0;
6793         int max = *nr_extents;
6794         int ret;
6795
6796         WARN_ON(!no_fragment && *extents);
6797         if (!exts) {
6798                 max = 1;
6799                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6800                 if (!exts)
6801                         return -ENOMEM;
6802         }
6803
6804         path = btrfs_alloc_path();
6805         BUG_ON(!path);
6806
6807         cur_pos = extent_key->objectid - offset;
6808         last_byte = extent_key->objectid + extent_key->offset;
6809         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6810                                        cur_pos, 0);
6811         if (ret < 0)
6812                 goto out;
6813         if (ret > 0) {
6814                 ret = -ENOENT;
6815                 goto out;
6816         }
6817
6818         while (1) {
6819                 leaf = path->nodes[0];
6820                 nritems = btrfs_header_nritems(leaf);
6821                 if (path->slots[0] >= nritems) {
6822                         ret = btrfs_next_leaf(root, path);
6823                         if (ret < 0)
6824                                 goto out;
6825                         if (ret > 0)
6826                                 break;
6827                         leaf = path->nodes[0];
6828                 }
6829
6830                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6831                 if (found_key.offset != cur_pos ||
6832                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6833                     found_key.objectid != reloc_inode->i_ino)
6834                         break;
6835
6836                 fi = btrfs_item_ptr(leaf, path->slots[0],
6837                                     struct btrfs_file_extent_item);
6838                 if (btrfs_file_extent_type(leaf, fi) !=
6839                     BTRFS_FILE_EXTENT_REG ||
6840                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6841                         break;
6842
6843                 if (nr == max) {
6844                         struct disk_extent *old = exts;
6845                         max *= 2;
6846                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6847                         memcpy(exts, old, sizeof(*exts) * nr);
6848                         if (old != *extents)
6849                                 kfree(old);
6850                 }
6851
6852                 exts[nr].disk_bytenr =
6853                         btrfs_file_extent_disk_bytenr(leaf, fi);
6854                 exts[nr].disk_num_bytes =
6855                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6856                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6857                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6858                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6859                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6860                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6861                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6862                                                                            fi);
6863                 BUG_ON(exts[nr].offset > 0);
6864                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6865                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6866
6867                 cur_pos += exts[nr].num_bytes;
6868                 nr++;
6869
6870                 if (cur_pos + offset >= last_byte)
6871                         break;
6872
6873                 if (no_fragment) {
6874                         ret = 1;
6875                         goto out;
6876                 }
6877                 path->slots[0]++;
6878         }
6879
6880         BUG_ON(cur_pos + offset > last_byte);
6881         if (cur_pos + offset < last_byte) {
6882                 ret = -ENOENT;
6883                 goto out;
6884         }
6885         ret = 0;
6886 out:
6887         btrfs_free_path(path);
6888         if (ret) {
6889                 if (exts != *extents)
6890                         kfree(exts);
6891         } else {
6892                 *extents = exts;
6893                 *nr_extents = nr;
6894         }
6895         return ret;
6896 }
6897
6898 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6899                                         struct btrfs_root *root,
6900                                         struct btrfs_path *path,
6901                                         struct btrfs_key *extent_key,
6902                                         struct btrfs_key *leaf_key,
6903                                         struct btrfs_ref_path *ref_path,
6904                                         struct disk_extent *new_extents,
6905                                         int nr_extents)
6906 {
6907         struct extent_buffer *leaf;
6908         struct btrfs_file_extent_item *fi;
6909         struct inode *inode = NULL;
6910         struct btrfs_key key;
6911         u64 lock_start = 0;
6912         u64 lock_end = 0;
6913         u64 num_bytes;
6914         u64 ext_offset;
6915         u64 search_end = (u64)-1;
6916         u32 nritems;
6917         int nr_scaned = 0;
6918         int extent_locked = 0;
6919         int extent_type;
6920         int ret;
6921
6922         memcpy(&key, leaf_key, sizeof(key));
6923         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6924                 if (key.objectid < ref_path->owner_objectid ||
6925                     (key.objectid == ref_path->owner_objectid &&
6926                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6927                         key.objectid = ref_path->owner_objectid;
6928                         key.type = BTRFS_EXTENT_DATA_KEY;
6929                         key.offset = 0;
6930                 }
6931         }
6932
6933         while (1) {
6934                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6935                 if (ret < 0)
6936                         goto out;
6937
6938                 leaf = path->nodes[0];
6939                 nritems = btrfs_header_nritems(leaf);
6940 next:
6941                 if (extent_locked && ret > 0) {
6942                         /*
6943                          * the file extent item was modified by someone
6944                          * before the extent got locked.
6945                          */
6946                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6947                                       lock_end, GFP_NOFS);
6948                         extent_locked = 0;
6949                 }
6950
6951                 if (path->slots[0] >= nritems) {
6952                         if (++nr_scaned > 2)
6953                                 break;
6954
6955                         BUG_ON(extent_locked);
6956                         ret = btrfs_next_leaf(root, path);
6957                         if (ret < 0)
6958                                 goto out;
6959                         if (ret > 0)
6960                                 break;
6961                         leaf = path->nodes[0];
6962                         nritems = btrfs_header_nritems(leaf);
6963                 }
6964
6965                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6966
6967                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6968                         if ((key.objectid > ref_path->owner_objectid) ||
6969                             (key.objectid == ref_path->owner_objectid &&
6970                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6971                             key.offset >= search_end)
6972                                 break;
6973                 }
6974
6975                 if (inode && key.objectid != inode->i_ino) {
6976                         BUG_ON(extent_locked);
6977                         btrfs_release_path(root, path);
6978                         mutex_unlock(&inode->i_mutex);
6979                         iput(inode);
6980                         inode = NULL;
6981                         continue;
6982                 }
6983
6984                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6985                         path->slots[0]++;
6986                         ret = 1;
6987                         goto next;
6988                 }
6989                 fi = btrfs_item_ptr(leaf, path->slots[0],
6990                                     struct btrfs_file_extent_item);
6991                 extent_type = btrfs_file_extent_type(leaf, fi);
6992                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6993                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6994                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6995                      extent_key->objectid)) {
6996                         path->slots[0]++;
6997                         ret = 1;
6998                         goto next;
6999                 }
7000
7001                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7002                 ext_offset = btrfs_file_extent_offset(leaf, fi);
7003
7004                 if (search_end == (u64)-1) {
7005                         search_end = key.offset - ext_offset +
7006                                 btrfs_file_extent_ram_bytes(leaf, fi);
7007                 }
7008
7009                 if (!extent_locked) {
7010                         lock_start = key.offset;
7011                         lock_end = lock_start + num_bytes - 1;
7012                 } else {
7013                         if (lock_start > key.offset ||
7014                             lock_end + 1 < key.offset + num_bytes) {
7015                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7016                                               lock_start, lock_end, GFP_NOFS);
7017                                 extent_locked = 0;
7018                         }
7019                 }
7020
7021                 if (!inode) {
7022                         btrfs_release_path(root, path);
7023
7024                         inode = btrfs_iget_locked(root->fs_info->sb,
7025                                                   key.objectid, root);
7026                         if (inode->i_state & I_NEW) {
7027                                 BTRFS_I(inode)->root = root;
7028                                 BTRFS_I(inode)->location.objectid =
7029                                         key.objectid;
7030                                 BTRFS_I(inode)->location.type =
7031                                         BTRFS_INODE_ITEM_KEY;
7032                                 BTRFS_I(inode)->location.offset = 0;
7033                                 btrfs_read_locked_inode(inode);
7034                                 unlock_new_inode(inode);
7035                         }
7036                         /*
7037                          * some code call btrfs_commit_transaction while
7038                          * holding the i_mutex, so we can't use mutex_lock
7039                          * here.
7040                          */
7041                         if (is_bad_inode(inode) ||
7042                             !mutex_trylock(&inode->i_mutex)) {
7043                                 iput(inode);
7044                                 inode = NULL;
7045                                 key.offset = (u64)-1;
7046                                 goto skip;
7047                         }
7048                 }
7049
7050                 if (!extent_locked) {
7051                         struct btrfs_ordered_extent *ordered;
7052
7053                         btrfs_release_path(root, path);
7054
7055                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7056                                     lock_end, GFP_NOFS);
7057                         ordered = btrfs_lookup_first_ordered_extent(inode,
7058                                                                     lock_end);
7059                         if (ordered &&
7060                             ordered->file_offset <= lock_end &&
7061                             ordered->file_offset + ordered->len > lock_start) {
7062                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7063                                               lock_start, lock_end, GFP_NOFS);
7064                                 btrfs_start_ordered_extent(inode, ordered, 1);
7065                                 btrfs_put_ordered_extent(ordered);
7066                                 key.offset += num_bytes;
7067                                 goto skip;
7068                         }
7069                         if (ordered)
7070                                 btrfs_put_ordered_extent(ordered);
7071
7072                         extent_locked = 1;
7073                         continue;
7074                 }
7075
7076                 if (nr_extents == 1) {
7077                         /* update extent pointer in place */
7078                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7079                                                 new_extents[0].disk_bytenr);
7080                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7081                                                 new_extents[0].disk_num_bytes);
7082                         btrfs_mark_buffer_dirty(leaf);
7083
7084                         btrfs_drop_extent_cache(inode, key.offset,
7085                                                 key.offset + num_bytes - 1, 0);
7086
7087                         ret = btrfs_inc_extent_ref(trans, root,
7088                                                 new_extents[0].disk_bytenr,
7089                                                 new_extents[0].disk_num_bytes,
7090                                                 leaf->start,
7091                                                 root->root_key.objectid,
7092                                                 trans->transid,
7093                                                 key.objectid);
7094                         BUG_ON(ret);
7095
7096                         ret = btrfs_free_extent(trans, root,
7097                                                 extent_key->objectid,
7098                                                 extent_key->offset,
7099                                                 leaf->start,
7100                                                 btrfs_header_owner(leaf),
7101                                                 btrfs_header_generation(leaf),
7102                                                 key.objectid, 0);
7103                         BUG_ON(ret);
7104
7105                         btrfs_release_path(root, path);
7106                         key.offset += num_bytes;
7107                 } else {
7108                         BUG_ON(1);
7109 #if 0
7110                         u64 alloc_hint;
7111                         u64 extent_len;
7112                         int i;
7113                         /*
7114                          * drop old extent pointer at first, then insert the
7115                          * new pointers one bye one
7116                          */
7117                         btrfs_release_path(root, path);
7118                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7119                                                  key.offset + num_bytes,
7120                                                  key.offset, &alloc_hint);
7121                         BUG_ON(ret);
7122
7123                         for (i = 0; i < nr_extents; i++) {
7124                                 if (ext_offset >= new_extents[i].num_bytes) {
7125                                         ext_offset -= new_extents[i].num_bytes;
7126                                         continue;
7127                                 }
7128                                 extent_len = min(new_extents[i].num_bytes -
7129                                                  ext_offset, num_bytes);
7130
7131                                 ret = btrfs_insert_empty_item(trans, root,
7132                                                               path, &key,
7133                                                               sizeof(*fi));
7134                                 BUG_ON(ret);
7135
7136                                 leaf = path->nodes[0];
7137                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7138                                                 struct btrfs_file_extent_item);
7139                                 btrfs_set_file_extent_generation(leaf, fi,
7140                                                         trans->transid);
7141                                 btrfs_set_file_extent_type(leaf, fi,
7142                                                         BTRFS_FILE_EXTENT_REG);
7143                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7144                                                 new_extents[i].disk_bytenr);
7145                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7146                                                 new_extents[i].disk_num_bytes);
7147                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7148                                                 new_extents[i].ram_bytes);
7149
7150                                 btrfs_set_file_extent_compression(leaf, fi,
7151                                                 new_extents[i].compression);
7152                                 btrfs_set_file_extent_encryption(leaf, fi,
7153                                                 new_extents[i].encryption);
7154                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7155                                                 new_extents[i].other_encoding);
7156
7157                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7158                                                         extent_len);
7159                                 ext_offset += new_extents[i].offset;
7160                                 btrfs_set_file_extent_offset(leaf, fi,
7161                                                         ext_offset);
7162                                 btrfs_mark_buffer_dirty(leaf);
7163
7164                                 btrfs_drop_extent_cache(inode, key.offset,
7165                                                 key.offset + extent_len - 1, 0);
7166
7167                                 ret = btrfs_inc_extent_ref(trans, root,
7168                                                 new_extents[i].disk_bytenr,
7169                                                 new_extents[i].disk_num_bytes,
7170                                                 leaf->start,
7171                                                 root->root_key.objectid,
7172                                                 trans->transid, key.objectid);
7173                                 BUG_ON(ret);
7174                                 btrfs_release_path(root, path);
7175
7176                                 inode_add_bytes(inode, extent_len);
7177
7178                                 ext_offset = 0;
7179                                 num_bytes -= extent_len;
7180                                 key.offset += extent_len;
7181
7182                                 if (num_bytes == 0)
7183                                         break;
7184                         }
7185                         BUG_ON(i >= nr_extents);
7186 #endif
7187                 }
7188
7189                 if (extent_locked) {
7190                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7191                                       lock_end, GFP_NOFS);
7192                         extent_locked = 0;
7193                 }
7194 skip:
7195                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7196                     key.offset >= search_end)
7197                         break;
7198
7199                 cond_resched();
7200         }
7201         ret = 0;
7202 out:
7203         btrfs_release_path(root, path);
7204         if (inode) {
7205                 mutex_unlock(&inode->i_mutex);
7206                 if (extent_locked) {
7207                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7208                                       lock_end, GFP_NOFS);
7209                 }
7210                 iput(inode);
7211         }
7212         return ret;
7213 }
7214
7215 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7216                                struct btrfs_root *root,
7217                                struct extent_buffer *buf, u64 orig_start)
7218 {
7219         int level;
7220         int ret;
7221
7222         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7223         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7224
7225         level = btrfs_header_level(buf);
7226         if (level == 0) {
7227                 struct btrfs_leaf_ref *ref;
7228                 struct btrfs_leaf_ref *orig_ref;
7229
7230                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7231                 if (!orig_ref)
7232                         return -ENOENT;
7233
7234                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7235                 if (!ref) {
7236                         btrfs_free_leaf_ref(root, orig_ref);
7237                         return -ENOMEM;
7238                 }
7239
7240                 ref->nritems = orig_ref->nritems;
7241                 memcpy(ref->extents, orig_ref->extents,
7242                         sizeof(ref->extents[0]) * ref->nritems);
7243
7244                 btrfs_free_leaf_ref(root, orig_ref);
7245
7246                 ref->root_gen = trans->transid;
7247                 ref->bytenr = buf->start;
7248                 ref->owner = btrfs_header_owner(buf);
7249                 ref->generation = btrfs_header_generation(buf);
7250
7251                 ret = btrfs_add_leaf_ref(root, ref, 0);
7252                 WARN_ON(ret);
7253                 btrfs_free_leaf_ref(root, ref);
7254         }
7255         return 0;
7256 }
7257
7258 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7259                                         struct extent_buffer *leaf,
7260                                         struct btrfs_block_group_cache *group,
7261                                         struct btrfs_root *target_root)
7262 {
7263         struct btrfs_key key;
7264         struct inode *inode = NULL;
7265         struct btrfs_file_extent_item *fi;
7266         struct extent_state *cached_state = NULL;
7267         u64 num_bytes;
7268         u64 skip_objectid = 0;
7269         u32 nritems;
7270         u32 i;
7271
7272         nritems = btrfs_header_nritems(leaf);
7273         for (i = 0; i < nritems; i++) {
7274                 btrfs_item_key_to_cpu(leaf, &key, i);
7275                 if (key.objectid == skip_objectid ||
7276                     key.type != BTRFS_EXTENT_DATA_KEY)
7277                         continue;
7278                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7279                 if (btrfs_file_extent_type(leaf, fi) ==
7280                     BTRFS_FILE_EXTENT_INLINE)
7281                         continue;
7282                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7283                         continue;
7284                 if (!inode || inode->i_ino != key.objectid) {
7285                         iput(inode);
7286                         inode = btrfs_ilookup(target_root->fs_info->sb,
7287                                               key.objectid, target_root, 1);
7288                 }
7289                 if (!inode) {
7290                         skip_objectid = key.objectid;
7291                         continue;
7292                 }
7293                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7294
7295                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7296                                  key.offset + num_bytes - 1, 0, &cached_state,
7297                                  GFP_NOFS);
7298                 btrfs_drop_extent_cache(inode, key.offset,
7299                                         key.offset + num_bytes - 1, 1);
7300                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7301                                      key.offset + num_bytes - 1, &cached_state,
7302                                      GFP_NOFS);
7303                 cond_resched();
7304         }
7305         iput(inode);
7306         return 0;
7307 }
7308
7309 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7310                                         struct btrfs_root *root,
7311                                         struct extent_buffer *leaf,
7312                                         struct btrfs_block_group_cache *group,
7313                                         struct inode *reloc_inode)
7314 {
7315         struct btrfs_key key;
7316         struct btrfs_key extent_key;
7317         struct btrfs_file_extent_item *fi;
7318         struct btrfs_leaf_ref *ref;
7319         struct disk_extent *new_extent;
7320         u64 bytenr;
7321         u64 num_bytes;
7322         u32 nritems;
7323         u32 i;
7324         int ext_index;
7325         int nr_extent;
7326         int ret;
7327
7328         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7329         BUG_ON(!new_extent);
7330
7331         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7332         BUG_ON(!ref);
7333
7334         ext_index = -1;
7335         nritems = btrfs_header_nritems(leaf);
7336         for (i = 0; i < nritems; i++) {
7337                 btrfs_item_key_to_cpu(leaf, &key, i);
7338                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7339                         continue;
7340                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7341                 if (btrfs_file_extent_type(leaf, fi) ==
7342                     BTRFS_FILE_EXTENT_INLINE)
7343                         continue;
7344                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7345                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7346                 if (bytenr == 0)
7347                         continue;
7348
7349                 ext_index++;
7350                 if (bytenr >= group->key.objectid + group->key.offset ||
7351                     bytenr + num_bytes <= group->key.objectid)
7352                         continue;
7353
7354                 extent_key.objectid = bytenr;
7355                 extent_key.offset = num_bytes;
7356                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7357                 nr_extent = 1;
7358                 ret = get_new_locations(reloc_inode, &extent_key,
7359                                         group->key.objectid, 1,
7360                                         &new_extent, &nr_extent);
7361                 if (ret > 0)
7362                         continue;
7363                 BUG_ON(ret < 0);
7364
7365                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7366                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7367                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7368                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7369
7370                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7371                                                 new_extent->disk_bytenr);
7372                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7373                                                 new_extent->disk_num_bytes);
7374                 btrfs_mark_buffer_dirty(leaf);
7375
7376                 ret = btrfs_inc_extent_ref(trans, root,
7377                                         new_extent->disk_bytenr,
7378                                         new_extent->disk_num_bytes,
7379                                         leaf->start,
7380                                         root->root_key.objectid,
7381                                         trans->transid, key.objectid);
7382                 BUG_ON(ret);
7383
7384                 ret = btrfs_free_extent(trans, root,
7385                                         bytenr, num_bytes, leaf->start,
7386                                         btrfs_header_owner(leaf),
7387                                         btrfs_header_generation(leaf),
7388                                         key.objectid, 0);
7389                 BUG_ON(ret);
7390                 cond_resched();
7391         }
7392         kfree(new_extent);
7393         BUG_ON(ext_index + 1 != ref->nritems);
7394         btrfs_free_leaf_ref(root, ref);
7395         return 0;
7396 }
7397
7398 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7399                           struct btrfs_root *root)
7400 {
7401         struct btrfs_root *reloc_root;
7402         int ret;
7403
7404         if (root->reloc_root) {
7405                 reloc_root = root->reloc_root;
7406                 root->reloc_root = NULL;
7407                 list_add(&reloc_root->dead_list,
7408                          &root->fs_info->dead_reloc_roots);
7409
7410                 btrfs_set_root_bytenr(&reloc_root->root_item,
7411                                       reloc_root->node->start);
7412                 btrfs_set_root_level(&root->root_item,
7413                                      btrfs_header_level(reloc_root->node));
7414                 memset(&reloc_root->root_item.drop_progress, 0,
7415                         sizeof(struct btrfs_disk_key));
7416                 reloc_root->root_item.drop_level = 0;
7417
7418                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7419                                         &reloc_root->root_key,
7420                                         &reloc_root->root_item);
7421                 BUG_ON(ret);
7422         }
7423         return 0;
7424 }
7425
7426 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7427 {
7428         struct btrfs_trans_handle *trans;
7429         struct btrfs_root *reloc_root;
7430         struct btrfs_root *prev_root = NULL;
7431         struct list_head dead_roots;
7432         int ret;
7433         unsigned long nr;
7434
7435         INIT_LIST_HEAD(&dead_roots);
7436         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7437
7438         while (!list_empty(&dead_roots)) {
7439                 reloc_root = list_entry(dead_roots.prev,
7440                                         struct btrfs_root, dead_list);
7441                 list_del_init(&reloc_root->dead_list);
7442
7443                 BUG_ON(reloc_root->commit_root != NULL);
7444                 while (1) {
7445                         trans = btrfs_join_transaction(root, 1);
7446                         BUG_ON(!trans);
7447
7448                         mutex_lock(&root->fs_info->drop_mutex);
7449                         ret = btrfs_drop_snapshot(trans, reloc_root);
7450                         if (ret != -EAGAIN)
7451                                 break;
7452                         mutex_unlock(&root->fs_info->drop_mutex);
7453
7454                         nr = trans->blocks_used;
7455                         ret = btrfs_end_transaction(trans, root);
7456                         BUG_ON(ret);
7457                         btrfs_btree_balance_dirty(root, nr);
7458                 }
7459
7460                 free_extent_buffer(reloc_root->node);
7461
7462                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7463                                      &reloc_root->root_key);
7464                 BUG_ON(ret);
7465                 mutex_unlock(&root->fs_info->drop_mutex);
7466
7467                 nr = trans->blocks_used;
7468                 ret = btrfs_end_transaction(trans, root);
7469                 BUG_ON(ret);
7470                 btrfs_btree_balance_dirty(root, nr);
7471
7472                 kfree(prev_root);
7473                 prev_root = reloc_root;
7474         }
7475         if (prev_root) {
7476                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7477                 kfree(prev_root);
7478         }
7479         return 0;
7480 }
7481
7482 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7483 {
7484         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7485         return 0;
7486 }
7487
7488 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7489 {
7490         struct btrfs_root *reloc_root;
7491         struct btrfs_trans_handle *trans;
7492         struct btrfs_key location;
7493         int found;
7494         int ret;
7495
7496         mutex_lock(&root->fs_info->tree_reloc_mutex);
7497         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7498         BUG_ON(ret);
7499         found = !list_empty(&root->fs_info->dead_reloc_roots);
7500         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7501
7502         if (found) {
7503                 trans = btrfs_start_transaction(root, 1);
7504                 BUG_ON(!trans);
7505                 ret = btrfs_commit_transaction(trans, root);
7506                 BUG_ON(ret);
7507         }
7508
7509         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7510         location.offset = (u64)-1;
7511         location.type = BTRFS_ROOT_ITEM_KEY;
7512
7513         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7514         BUG_ON(!reloc_root);
7515         btrfs_orphan_cleanup(reloc_root);
7516         return 0;
7517 }
7518
7519 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7520                                     struct btrfs_root *root)
7521 {
7522         struct btrfs_root *reloc_root;
7523         struct extent_buffer *eb;
7524         struct btrfs_root_item *root_item;
7525         struct btrfs_key root_key;
7526         int ret;
7527
7528         BUG_ON(!root->ref_cows);
7529         if (root->reloc_root)
7530                 return 0;
7531
7532         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7533         BUG_ON(!root_item);
7534
7535         ret = btrfs_copy_root(trans, root, root->commit_root,
7536                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7537         BUG_ON(ret);
7538
7539         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7540         root_key.offset = root->root_key.objectid;
7541         root_key.type = BTRFS_ROOT_ITEM_KEY;
7542
7543         memcpy(root_item, &root->root_item, sizeof(root_item));
7544         btrfs_set_root_refs(root_item, 0);
7545         btrfs_set_root_bytenr(root_item, eb->start);
7546         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7547         btrfs_set_root_generation(root_item, trans->transid);
7548
7549         btrfs_tree_unlock(eb);
7550         free_extent_buffer(eb);
7551
7552         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7553                                 &root_key, root_item);
7554         BUG_ON(ret);
7555         kfree(root_item);
7556
7557         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7558                                                  &root_key);
7559         BUG_ON(!reloc_root);
7560         reloc_root->last_trans = trans->transid;
7561         reloc_root->commit_root = NULL;
7562         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7563
7564         root->reloc_root = reloc_root;
7565         return 0;
7566 }
7567
7568 /*
7569  * Core function of space balance.
7570  *
7571  * The idea is using reloc trees to relocate tree blocks in reference
7572  * counted roots. There is one reloc tree for each subvol, and all
7573  * reloc trees share same root key objectid. Reloc trees are snapshots
7574  * of the latest committed roots of subvols (root->commit_root).
7575  *
7576  * To relocate a tree block referenced by a subvol, there are two steps.
7577  * COW the block through subvol's reloc tree, then update block pointer
7578  * in the subvol to point to the new block. Since all reloc trees share
7579  * same root key objectid, doing special handing for tree blocks owned
7580  * by them is easy. Once a tree block has been COWed in one reloc tree,
7581  * we can use the resulting new block directly when the same block is
7582  * required to COW again through other reloc trees. By this way, relocated
7583  * tree blocks are shared between reloc trees, so they are also shared
7584  * between subvols.
7585  */
7586 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7587                                       struct btrfs_root *root,
7588                                       struct btrfs_path *path,
7589                                       struct btrfs_key *first_key,
7590                                       struct btrfs_ref_path *ref_path,
7591                                       struct btrfs_block_group_cache *group,
7592                                       struct inode *reloc_inode)
7593 {
7594         struct btrfs_root *reloc_root;
7595         struct extent_buffer *eb = NULL;
7596         struct btrfs_key *keys;
7597         u64 *nodes;
7598         int level;
7599         int shared_level;
7600         int lowest_level = 0;
7601         int ret;
7602
7603         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7604                 lowest_level = ref_path->owner_objectid;
7605
7606         if (!root->ref_cows) {
7607                 path->lowest_level = lowest_level;
7608                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7609                 BUG_ON(ret < 0);
7610                 path->lowest_level = 0;
7611                 btrfs_release_path(root, path);
7612                 return 0;
7613         }
7614
7615         mutex_lock(&root->fs_info->tree_reloc_mutex);
7616         ret = init_reloc_tree(trans, root);
7617         BUG_ON(ret);
7618         reloc_root = root->reloc_root;
7619
7620         shared_level = ref_path->shared_level;
7621         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7622
7623         keys = ref_path->node_keys;
7624         nodes = ref_path->new_nodes;
7625         memset(&keys[shared_level + 1], 0,
7626                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7627         memset(&nodes[shared_level + 1], 0,
7628                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7629
7630         if (nodes[lowest_level] == 0) {
7631                 path->lowest_level = lowest_level;
7632                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7633                                         0, 1);
7634                 BUG_ON(ret);
7635                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7636                         eb = path->nodes[level];
7637                         if (!eb || eb == reloc_root->node)
7638                                 break;
7639                         nodes[level] = eb->start;
7640                         if (level == 0)
7641                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7642                         else
7643                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7644                 }
7645                 if (nodes[0] &&
7646                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7647                         eb = path->nodes[0];
7648                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7649                                                       group, reloc_inode);
7650                         BUG_ON(ret);
7651                 }
7652                 btrfs_release_path(reloc_root, path);
7653         } else {
7654                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7655                                        lowest_level);
7656                 BUG_ON(ret);
7657         }
7658
7659         /*
7660          * replace tree blocks in the fs tree with tree blocks in
7661          * the reloc tree.
7662          */
7663         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7664         BUG_ON(ret < 0);
7665
7666         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7667                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7668                                         0, 0);
7669                 BUG_ON(ret);
7670                 extent_buffer_get(path->nodes[0]);
7671                 eb = path->nodes[0];
7672                 btrfs_release_path(reloc_root, path);
7673                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7674                 BUG_ON(ret);
7675                 free_extent_buffer(eb);
7676         }
7677
7678         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7679         path->lowest_level = 0;
7680         return 0;
7681 }
7682
7683 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7684                                         struct btrfs_root *root,
7685                                         struct btrfs_path *path,
7686                                         struct btrfs_key *first_key,
7687                                         struct btrfs_ref_path *ref_path)
7688 {
7689         int ret;
7690
7691         ret = relocate_one_path(trans, root, path, first_key,
7692                                 ref_path, NULL, NULL);
7693         BUG_ON(ret);
7694
7695         return 0;
7696 }
7697
7698 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7699                                     struct btrfs_root *extent_root,
7700                                     struct btrfs_path *path,
7701                                     struct btrfs_key *extent_key)
7702 {
7703         int ret;
7704
7705         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7706         if (ret)
7707                 goto out;
7708         ret = btrfs_del_item(trans, extent_root, path);
7709 out:
7710         btrfs_release_path(extent_root, path);
7711         return ret;
7712 }
7713
7714 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7715                                                 struct btrfs_ref_path *ref_path)
7716 {
7717         struct btrfs_key root_key;
7718
7719         root_key.objectid = ref_path->root_objectid;
7720         root_key.type = BTRFS_ROOT_ITEM_KEY;
7721         if (is_cowonly_root(ref_path->root_objectid))
7722                 root_key.offset = 0;
7723         else
7724                 root_key.offset = (u64)-1;
7725
7726         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7727 }
7728
7729 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7730                                         struct btrfs_path *path,
7731                                         struct btrfs_key *extent_key,
7732                                         struct btrfs_block_group_cache *group,
7733                                         struct inode *reloc_inode, int pass)
7734 {
7735         struct btrfs_trans_handle *trans;
7736         struct btrfs_root *found_root;
7737         struct btrfs_ref_path *ref_path = NULL;
7738         struct disk_extent *new_extents = NULL;
7739         int nr_extents = 0;
7740         int loops;
7741         int ret;
7742         int level;
7743         struct btrfs_key first_key;
7744         u64 prev_block = 0;
7745
7746
7747         trans = btrfs_start_transaction(extent_root, 1);
7748         BUG_ON(!trans);
7749
7750         if (extent_key->objectid == 0) {
7751                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7752                 goto out;
7753         }
7754
7755         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7756         if (!ref_path) {
7757                 ret = -ENOMEM;
7758                 goto out;
7759         }
7760
7761         for (loops = 0; ; loops++) {
7762                 if (loops == 0) {
7763                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7764                                                    extent_key->objectid);
7765                 } else {
7766                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7767                 }
7768                 if (ret < 0)
7769                         goto out;
7770                 if (ret > 0)
7771                         break;
7772
7773                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7774                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7775                         continue;
7776
7777                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7778                 BUG_ON(!found_root);
7779                 /*
7780                  * for reference counted tree, only process reference paths
7781                  * rooted at the latest committed root.
7782                  */
7783                 if (found_root->ref_cows &&
7784                     ref_path->root_generation != found_root->root_key.offset)
7785                         continue;
7786
7787                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7788                         if (pass == 0) {
7789                                 /*
7790                                  * copy data extents to new locations
7791                                  */
7792                                 u64 group_start = group->key.objectid;
7793                                 ret = relocate_data_extent(reloc_inode,
7794                                                            extent_key,
7795                                                            group_start);
7796                                 if (ret < 0)
7797                                         goto out;
7798                                 break;
7799                         }
7800                         level = 0;
7801                 } else {
7802                         level = ref_path->owner_objectid;
7803                 }
7804
7805                 if (prev_block != ref_path->nodes[level]) {
7806                         struct extent_buffer *eb;
7807                         u64 block_start = ref_path->nodes[level];
7808                         u64 block_size = btrfs_level_size(found_root, level);
7809
7810                         eb = read_tree_block(found_root, block_start,
7811                                              block_size, 0);
7812                         btrfs_tree_lock(eb);
7813                         BUG_ON(level != btrfs_header_level(eb));
7814
7815                         if (level == 0)
7816                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7817                         else
7818                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7819
7820                         btrfs_tree_unlock(eb);
7821                         free_extent_buffer(eb);
7822                         prev_block = block_start;
7823                 }
7824
7825                 mutex_lock(&extent_root->fs_info->trans_mutex);
7826                 btrfs_record_root_in_trans(found_root);
7827                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7828                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7829                         /*
7830                          * try to update data extent references while
7831                          * keeping metadata shared between snapshots.
7832                          */
7833                         if (pass == 1) {
7834                                 ret = relocate_one_path(trans, found_root,
7835                                                 path, &first_key, ref_path,
7836                                                 group, reloc_inode);
7837                                 if (ret < 0)
7838                                         goto out;
7839                                 continue;
7840                         }
7841                         /*
7842                          * use fallback method to process the remaining
7843                          * references.
7844                          */
7845                         if (!new_extents) {
7846                                 u64 group_start = group->key.objectid;
7847                                 new_extents = kmalloc(sizeof(*new_extents),
7848                                                       GFP_NOFS);
7849                                 nr_extents = 1;
7850                                 ret = get_new_locations(reloc_inode,
7851                                                         extent_key,
7852                                                         group_start, 1,
7853                                                         &new_extents,
7854                                                         &nr_extents);
7855                                 if (ret)
7856                                         goto out;
7857                         }
7858                         ret = replace_one_extent(trans, found_root,
7859                                                 path, extent_key,
7860                                                 &first_key, ref_path,
7861                                                 new_extents, nr_extents);
7862                 } else {
7863                         ret = relocate_tree_block(trans, found_root, path,
7864                                                   &first_key, ref_path);
7865                 }
7866                 if (ret < 0)
7867                         goto out;
7868         }
7869         ret = 0;
7870 out:
7871         btrfs_end_transaction(trans, extent_root);
7872         kfree(new_extents);
7873         kfree(ref_path);
7874         return ret;
7875 }
7876 #endif
7877
7878 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7879 {
7880         u64 num_devices;
7881         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7882                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7883
7884         num_devices = root->fs_info->fs_devices->rw_devices;
7885         if (num_devices == 1) {
7886                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7887                 stripped = flags & ~stripped;
7888
7889                 /* turn raid0 into single device chunks */
7890                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7891                         return stripped;
7892
7893                 /* turn mirroring into duplication */
7894                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7895                              BTRFS_BLOCK_GROUP_RAID10))
7896                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7897                 return flags;
7898         } else {
7899                 /* they already had raid on here, just return */
7900                 if (flags & stripped)
7901                         return flags;
7902
7903                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7904                 stripped = flags & ~stripped;
7905
7906                 /* switch duplicated blocks with raid1 */
7907                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7908                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7909
7910                 /* turn single device chunks into raid0 */
7911                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7912         }
7913         return flags;
7914 }
7915
7916 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
7917 {
7918         struct btrfs_space_info *sinfo = cache->space_info;
7919         u64 num_bytes;
7920         int ret = -ENOSPC;
7921
7922         if (cache->ro)
7923                 return 0;
7924
7925         spin_lock(&sinfo->lock);
7926         spin_lock(&cache->lock);
7927         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7928                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7929
7930         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7931             sinfo->bytes_may_use + sinfo->bytes_readonly +
7932             cache->reserved_pinned + num_bytes < sinfo->total_bytes) {
7933                 sinfo->bytes_readonly += num_bytes;
7934                 sinfo->bytes_reserved += cache->reserved_pinned;
7935                 cache->reserved_pinned = 0;
7936                 cache->ro = 1;
7937                 ret = 0;
7938         }
7939         spin_unlock(&cache->lock);
7940         spin_unlock(&sinfo->lock);
7941         return ret;
7942 }
7943
7944 int btrfs_set_block_group_ro(struct btrfs_root *root,
7945                              struct btrfs_block_group_cache *cache)
7946
7947 {
7948         struct btrfs_trans_handle *trans;
7949         u64 alloc_flags;
7950         int ret;
7951
7952         BUG_ON(cache->ro);
7953
7954         trans = btrfs_join_transaction(root, 1);
7955         BUG_ON(IS_ERR(trans));
7956
7957         alloc_flags = update_block_group_flags(root, cache->flags);
7958         if (alloc_flags != cache->flags)
7959                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7960
7961         ret = set_block_group_ro(cache);
7962         if (!ret)
7963                 goto out;
7964         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7965         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7966         if (ret < 0)
7967                 goto out;
7968         ret = set_block_group_ro(cache);
7969 out:
7970         btrfs_end_transaction(trans, root);
7971         return ret;
7972 }
7973
7974 int btrfs_set_block_group_rw(struct btrfs_root *root,
7975                               struct btrfs_block_group_cache *cache)
7976 {
7977         struct btrfs_space_info *sinfo = cache->space_info;
7978         u64 num_bytes;
7979
7980         BUG_ON(!cache->ro);
7981
7982         spin_lock(&sinfo->lock);
7983         spin_lock(&cache->lock);
7984         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7985                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7986         sinfo->bytes_readonly -= num_bytes;
7987         cache->ro = 0;
7988         spin_unlock(&cache->lock);
7989         spin_unlock(&sinfo->lock);
7990         return 0;
7991 }
7992
7993 /*
7994  * checks to see if its even possible to relocate this block group.
7995  *
7996  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7997  * ok to go ahead and try.
7998  */
7999 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8000 {
8001         struct btrfs_block_group_cache *block_group;
8002         struct btrfs_space_info *space_info;
8003         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8004         struct btrfs_device *device;
8005         int full = 0;
8006         int ret = 0;
8007
8008         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8009
8010         /* odd, couldn't find the block group, leave it alone */
8011         if (!block_group)
8012                 return -1;
8013
8014         /* no bytes used, we're good */
8015         if (!btrfs_block_group_used(&block_group->item))
8016                 goto out;
8017
8018         space_info = block_group->space_info;
8019         spin_lock(&space_info->lock);
8020
8021         full = space_info->full;
8022
8023         /*
8024          * if this is the last block group we have in this space, we can't
8025          * relocate it unless we're able to allocate a new chunk below.
8026          *
8027          * Otherwise, we need to make sure we have room in the space to handle
8028          * all of the extents from this block group.  If we can, we're good
8029          */
8030         if ((space_info->total_bytes != block_group->key.offset) &&
8031            (space_info->bytes_used + space_info->bytes_reserved +
8032             space_info->bytes_pinned + space_info->bytes_readonly +
8033             btrfs_block_group_used(&block_group->item) <
8034             space_info->total_bytes)) {
8035                 spin_unlock(&space_info->lock);
8036                 goto out;
8037         }
8038         spin_unlock(&space_info->lock);
8039
8040         /*
8041          * ok we don't have enough space, but maybe we have free space on our
8042          * devices to allocate new chunks for relocation, so loop through our
8043          * alloc devices and guess if we have enough space.  However, if we
8044          * were marked as full, then we know there aren't enough chunks, and we
8045          * can just return.
8046          */
8047         ret = -1;
8048         if (full)
8049                 goto out;
8050
8051         mutex_lock(&root->fs_info->chunk_mutex);
8052         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8053                 u64 min_free = btrfs_block_group_used(&block_group->item);
8054                 u64 dev_offset, max_avail;
8055
8056                 /*
8057                  * check to make sure we can actually find a chunk with enough
8058                  * space to fit our block group in.
8059                  */
8060                 if (device->total_bytes > device->bytes_used + min_free) {
8061                         ret = find_free_dev_extent(NULL, device, min_free,
8062                                                    &dev_offset, &max_avail);
8063                         if (!ret)
8064                                 break;
8065                         ret = -1;
8066                 }
8067         }
8068         mutex_unlock(&root->fs_info->chunk_mutex);
8069 out:
8070         btrfs_put_block_group(block_group);
8071         return ret;
8072 }
8073
8074 static int find_first_block_group(struct btrfs_root *root,
8075                 struct btrfs_path *path, struct btrfs_key *key)
8076 {
8077         int ret = 0;
8078         struct btrfs_key found_key;
8079         struct extent_buffer *leaf;
8080         int slot;
8081
8082         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8083         if (ret < 0)
8084                 goto out;
8085
8086         while (1) {
8087                 slot = path->slots[0];
8088                 leaf = path->nodes[0];
8089                 if (slot >= btrfs_header_nritems(leaf)) {
8090                         ret = btrfs_next_leaf(root, path);
8091                         if (ret == 0)
8092                                 continue;
8093                         if (ret < 0)
8094                                 goto out;
8095                         break;
8096                 }
8097                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8098
8099                 if (found_key.objectid >= key->objectid &&
8100                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8101                         ret = 0;
8102                         goto out;
8103                 }
8104                 path->slots[0]++;
8105         }
8106 out:
8107         return ret;
8108 }
8109
8110 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8111 {
8112         struct btrfs_block_group_cache *block_group;
8113         u64 last = 0;
8114
8115         while (1) {
8116                 struct inode *inode;
8117
8118                 block_group = btrfs_lookup_first_block_group(info, last);
8119                 while (block_group) {
8120                         spin_lock(&block_group->lock);
8121                         if (block_group->iref)
8122                                 break;
8123                         spin_unlock(&block_group->lock);
8124                         block_group = next_block_group(info->tree_root,
8125                                                        block_group);
8126                 }
8127                 if (!block_group) {
8128                         if (last == 0)
8129                                 break;
8130                         last = 0;
8131                         continue;
8132                 }
8133
8134                 inode = block_group->inode;
8135                 block_group->iref = 0;
8136                 block_group->inode = NULL;
8137                 spin_unlock(&block_group->lock);
8138                 iput(inode);
8139                 last = block_group->key.objectid + block_group->key.offset;
8140                 btrfs_put_block_group(block_group);
8141         }
8142 }
8143
8144 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8145 {
8146         struct btrfs_block_group_cache *block_group;
8147         struct btrfs_space_info *space_info;
8148         struct btrfs_caching_control *caching_ctl;
8149         struct rb_node *n;
8150
8151         down_write(&info->extent_commit_sem);
8152         while (!list_empty(&info->caching_block_groups)) {
8153                 caching_ctl = list_entry(info->caching_block_groups.next,
8154                                          struct btrfs_caching_control, list);
8155                 list_del(&caching_ctl->list);
8156                 put_caching_control(caching_ctl);
8157         }
8158         up_write(&info->extent_commit_sem);
8159
8160         spin_lock(&info->block_group_cache_lock);
8161         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8162                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8163                                        cache_node);
8164                 rb_erase(&block_group->cache_node,
8165                          &info->block_group_cache_tree);
8166                 spin_unlock(&info->block_group_cache_lock);
8167
8168                 down_write(&block_group->space_info->groups_sem);
8169                 list_del(&block_group->list);
8170                 up_write(&block_group->space_info->groups_sem);
8171
8172                 if (block_group->cached == BTRFS_CACHE_STARTED)
8173                         wait_block_group_cache_done(block_group);
8174
8175                 btrfs_remove_free_space_cache(block_group);
8176                 btrfs_put_block_group(block_group);
8177
8178                 spin_lock(&info->block_group_cache_lock);
8179         }
8180         spin_unlock(&info->block_group_cache_lock);
8181
8182         /* now that all the block groups are freed, go through and
8183          * free all the space_info structs.  This is only called during
8184          * the final stages of unmount, and so we know nobody is
8185          * using them.  We call synchronize_rcu() once before we start,
8186          * just to be on the safe side.
8187          */
8188         synchronize_rcu();
8189
8190         release_global_block_rsv(info);
8191
8192         while(!list_empty(&info->space_info)) {
8193                 space_info = list_entry(info->space_info.next,
8194                                         struct btrfs_space_info,
8195                                         list);
8196                 if (space_info->bytes_pinned > 0 ||
8197                     space_info->bytes_reserved > 0) {
8198                         WARN_ON(1);
8199                         dump_space_info(space_info, 0, 0);
8200                 }
8201                 list_del(&space_info->list);
8202                 kfree(space_info);
8203         }
8204         return 0;
8205 }
8206
8207 static void __link_block_group(struct btrfs_space_info *space_info,
8208                                struct btrfs_block_group_cache *cache)
8209 {
8210         int index = get_block_group_index(cache);
8211
8212         down_write(&space_info->groups_sem);
8213         list_add_tail(&cache->list, &space_info->block_groups[index]);
8214         up_write(&space_info->groups_sem);
8215 }
8216
8217 int btrfs_read_block_groups(struct btrfs_root *root)
8218 {
8219         struct btrfs_path *path;
8220         int ret;
8221         struct btrfs_block_group_cache *cache;
8222         struct btrfs_fs_info *info = root->fs_info;
8223         struct btrfs_space_info *space_info;
8224         struct btrfs_key key;
8225         struct btrfs_key found_key;
8226         struct extent_buffer *leaf;
8227         int need_clear = 0;
8228         u64 cache_gen;
8229
8230         root = info->extent_root;
8231         key.objectid = 0;
8232         key.offset = 0;
8233         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8234         path = btrfs_alloc_path();
8235         if (!path)
8236                 return -ENOMEM;
8237
8238         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8239         if (cache_gen != 0 &&
8240             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8241                 need_clear = 1;
8242         if (btrfs_test_opt(root, CLEAR_CACHE))
8243                 need_clear = 1;
8244         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8245                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8246
8247         while (1) {
8248                 ret = find_first_block_group(root, path, &key);
8249                 if (ret > 0)
8250                         break;
8251                 if (ret != 0)
8252                         goto error;
8253
8254                 leaf = path->nodes[0];
8255                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8256                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8257                 if (!cache) {
8258                         ret = -ENOMEM;
8259                         goto error;
8260                 }
8261
8262                 atomic_set(&cache->count, 1);
8263                 spin_lock_init(&cache->lock);
8264                 spin_lock_init(&cache->tree_lock);
8265                 cache->fs_info = info;
8266                 INIT_LIST_HEAD(&cache->list);
8267                 INIT_LIST_HEAD(&cache->cluster_list);
8268
8269                 if (need_clear)
8270                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8271
8272                 /*
8273                  * we only want to have 32k of ram per block group for keeping
8274                  * track of free space, and if we pass 1/2 of that we want to
8275                  * start converting things over to using bitmaps
8276                  */
8277                 cache->extents_thresh = ((1024 * 32) / 2) /
8278                         sizeof(struct btrfs_free_space);
8279
8280                 read_extent_buffer(leaf, &cache->item,
8281                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8282                                    sizeof(cache->item));
8283                 memcpy(&cache->key, &found_key, sizeof(found_key));
8284
8285                 key.objectid = found_key.objectid + found_key.offset;
8286                 btrfs_release_path(root, path);
8287                 cache->flags = btrfs_block_group_flags(&cache->item);
8288                 cache->sectorsize = root->sectorsize;
8289
8290                 /*
8291                  * check for two cases, either we are full, and therefore
8292                  * don't need to bother with the caching work since we won't
8293                  * find any space, or we are empty, and we can just add all
8294                  * the space in and be done with it.  This saves us _alot_ of
8295                  * time, particularly in the full case.
8296                  */
8297                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8298                         exclude_super_stripes(root, cache);
8299                         cache->last_byte_to_unpin = (u64)-1;
8300                         cache->cached = BTRFS_CACHE_FINISHED;
8301                         free_excluded_extents(root, cache);
8302                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8303                         exclude_super_stripes(root, cache);
8304                         cache->last_byte_to_unpin = (u64)-1;
8305                         cache->cached = BTRFS_CACHE_FINISHED;
8306                         add_new_free_space(cache, root->fs_info,
8307                                            found_key.objectid,
8308                                            found_key.objectid +
8309                                            found_key.offset);
8310                         free_excluded_extents(root, cache);
8311                 }
8312
8313                 ret = update_space_info(info, cache->flags, found_key.offset,
8314                                         btrfs_block_group_used(&cache->item),
8315                                         &space_info);
8316                 BUG_ON(ret);
8317                 cache->space_info = space_info;
8318                 spin_lock(&cache->space_info->lock);
8319                 cache->space_info->bytes_readonly += cache->bytes_super;
8320                 spin_unlock(&cache->space_info->lock);
8321
8322                 __link_block_group(space_info, cache);
8323
8324                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8325                 BUG_ON(ret);
8326
8327                 set_avail_alloc_bits(root->fs_info, cache->flags);
8328                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8329                         set_block_group_ro(cache);
8330         }
8331
8332         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8333                 if (!(get_alloc_profile(root, space_info->flags) &
8334                       (BTRFS_BLOCK_GROUP_RAID10 |
8335                        BTRFS_BLOCK_GROUP_RAID1 |
8336                        BTRFS_BLOCK_GROUP_DUP)))
8337                         continue;
8338                 /*
8339                  * avoid allocating from un-mirrored block group if there are
8340                  * mirrored block groups.
8341                  */
8342                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8343                         set_block_group_ro(cache);
8344                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8345                         set_block_group_ro(cache);
8346         }
8347
8348         init_global_block_rsv(info);
8349         ret = 0;
8350 error:
8351         btrfs_free_path(path);
8352         return ret;
8353 }
8354
8355 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8356                            struct btrfs_root *root, u64 bytes_used,
8357                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8358                            u64 size)
8359 {
8360         int ret;
8361         struct btrfs_root *extent_root;
8362         struct btrfs_block_group_cache *cache;
8363
8364         extent_root = root->fs_info->extent_root;
8365
8366         root->fs_info->last_trans_log_full_commit = trans->transid;
8367
8368         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8369         if (!cache)
8370                 return -ENOMEM;
8371
8372         cache->key.objectid = chunk_offset;
8373         cache->key.offset = size;
8374         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8375         cache->sectorsize = root->sectorsize;
8376         cache->fs_info = root->fs_info;
8377
8378         /*
8379          * we only want to have 32k of ram per block group for keeping track
8380          * of free space, and if we pass 1/2 of that we want to start
8381          * converting things over to using bitmaps
8382          */
8383         cache->extents_thresh = ((1024 * 32) / 2) /
8384                 sizeof(struct btrfs_free_space);
8385         atomic_set(&cache->count, 1);
8386         spin_lock_init(&cache->lock);
8387         spin_lock_init(&cache->tree_lock);
8388         INIT_LIST_HEAD(&cache->list);
8389         INIT_LIST_HEAD(&cache->cluster_list);
8390
8391         btrfs_set_block_group_used(&cache->item, bytes_used);
8392         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8393         cache->flags = type;
8394         btrfs_set_block_group_flags(&cache->item, type);
8395
8396         cache->last_byte_to_unpin = (u64)-1;
8397         cache->cached = BTRFS_CACHE_FINISHED;
8398         exclude_super_stripes(root, cache);
8399
8400         add_new_free_space(cache, root->fs_info, chunk_offset,
8401                            chunk_offset + size);
8402
8403         free_excluded_extents(root, cache);
8404
8405         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8406                                 &cache->space_info);
8407         BUG_ON(ret);
8408
8409         spin_lock(&cache->space_info->lock);
8410         cache->space_info->bytes_readonly += cache->bytes_super;
8411         spin_unlock(&cache->space_info->lock);
8412
8413         __link_block_group(cache->space_info, cache);
8414
8415         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8416         BUG_ON(ret);
8417
8418         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8419                                 sizeof(cache->item));
8420         BUG_ON(ret);
8421
8422         set_avail_alloc_bits(extent_root->fs_info, type);
8423
8424         return 0;
8425 }
8426
8427 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8428                              struct btrfs_root *root, u64 group_start)
8429 {
8430         struct btrfs_path *path;
8431         struct btrfs_block_group_cache *block_group;
8432         struct btrfs_free_cluster *cluster;
8433         struct btrfs_root *tree_root = root->fs_info->tree_root;
8434         struct btrfs_key key;
8435         struct inode *inode;
8436         int ret;
8437         int factor;
8438
8439         root = root->fs_info->extent_root;
8440
8441         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8442         BUG_ON(!block_group);
8443         BUG_ON(!block_group->ro);
8444
8445         memcpy(&key, &block_group->key, sizeof(key));
8446         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8447                                   BTRFS_BLOCK_GROUP_RAID1 |
8448                                   BTRFS_BLOCK_GROUP_RAID10))
8449                 factor = 2;
8450         else
8451                 factor = 1;
8452
8453         /* make sure this block group isn't part of an allocation cluster */
8454         cluster = &root->fs_info->data_alloc_cluster;
8455         spin_lock(&cluster->refill_lock);
8456         btrfs_return_cluster_to_free_space(block_group, cluster);
8457         spin_unlock(&cluster->refill_lock);
8458
8459         /*
8460          * make sure this block group isn't part of a metadata
8461          * allocation cluster
8462          */
8463         cluster = &root->fs_info->meta_alloc_cluster;
8464         spin_lock(&cluster->refill_lock);
8465         btrfs_return_cluster_to_free_space(block_group, cluster);
8466         spin_unlock(&cluster->refill_lock);
8467
8468         path = btrfs_alloc_path();
8469         BUG_ON(!path);
8470
8471         inode = lookup_free_space_inode(root, block_group, path);
8472         if (!IS_ERR(inode)) {
8473                 btrfs_orphan_add(trans, inode);
8474                 clear_nlink(inode);
8475                 /* One for the block groups ref */
8476                 spin_lock(&block_group->lock);
8477                 if (block_group->iref) {
8478                         block_group->iref = 0;
8479                         block_group->inode = NULL;
8480                         spin_unlock(&block_group->lock);
8481                         iput(inode);
8482                 } else {
8483                         spin_unlock(&block_group->lock);
8484                 }
8485                 /* One for our lookup ref */
8486                 iput(inode);
8487         }
8488
8489         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8490         key.offset = block_group->key.objectid;
8491         key.type = 0;
8492
8493         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8494         if (ret < 0)
8495                 goto out;
8496         if (ret > 0)
8497                 btrfs_release_path(tree_root, path);
8498         if (ret == 0) {
8499                 ret = btrfs_del_item(trans, tree_root, path);
8500                 if (ret)
8501                         goto out;
8502                 btrfs_release_path(tree_root, path);
8503         }
8504
8505         spin_lock(&root->fs_info->block_group_cache_lock);
8506         rb_erase(&block_group->cache_node,
8507                  &root->fs_info->block_group_cache_tree);
8508         spin_unlock(&root->fs_info->block_group_cache_lock);
8509
8510         down_write(&block_group->space_info->groups_sem);
8511         /*
8512          * we must use list_del_init so people can check to see if they
8513          * are still on the list after taking the semaphore
8514          */
8515         list_del_init(&block_group->list);
8516         up_write(&block_group->space_info->groups_sem);
8517
8518         if (block_group->cached == BTRFS_CACHE_STARTED)
8519                 wait_block_group_cache_done(block_group);
8520
8521         btrfs_remove_free_space_cache(block_group);
8522
8523         spin_lock(&block_group->space_info->lock);
8524         block_group->space_info->total_bytes -= block_group->key.offset;
8525         block_group->space_info->bytes_readonly -= block_group->key.offset;
8526         block_group->space_info->disk_total -= block_group->key.offset * factor;
8527         spin_unlock(&block_group->space_info->lock);
8528
8529         memcpy(&key, &block_group->key, sizeof(key));
8530
8531         btrfs_clear_space_info_full(root->fs_info);
8532
8533         btrfs_put_block_group(block_group);
8534         btrfs_put_block_group(block_group);
8535
8536         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8537         if (ret > 0)
8538                 ret = -EIO;
8539         if (ret < 0)
8540                 goto out;
8541
8542         ret = btrfs_del_item(trans, root, path);
8543 out:
8544         btrfs_free_path(path);
8545         return ret;
8546 }