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 num_pages = 0;
2746         int retries = 0;
2747         int ret = 0;
2748
2749         /*
2750          * If this block group is smaller than 100 megs don't bother caching the
2751          * block group.
2752          */
2753         if (block_group->key.offset < (100 * 1024 * 1024)) {
2754                 spin_lock(&block_group->lock);
2755                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2756                 spin_unlock(&block_group->lock);
2757                 return 0;
2758         }
2759
2760 again:
2761         inode = lookup_free_space_inode(root, block_group, path);
2762         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2763                 ret = PTR_ERR(inode);
2764                 btrfs_release_path(root, path);
2765                 goto out;
2766         }
2767
2768         if (IS_ERR(inode)) {
2769                 BUG_ON(retries);
2770                 retries++;
2771
2772                 if (block_group->ro)
2773                         goto out_free;
2774
2775                 ret = create_free_space_inode(root, trans, block_group, path);
2776                 if (ret)
2777                         goto out_free;
2778                 goto again;
2779         }
2780
2781         /*
2782          * We want to set the generation to 0, that way if anything goes wrong
2783          * from here on out we know not to trust this cache when we load up next
2784          * time.
2785          */
2786         BTRFS_I(inode)->generation = 0;
2787         ret = btrfs_update_inode(trans, root, inode);
2788         WARN_ON(ret);
2789
2790         if (i_size_read(inode) > 0) {
2791                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2792                                                       inode);
2793                 if (ret)
2794                         goto out_put;
2795         }
2796
2797         spin_lock(&block_group->lock);
2798         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2799                 spin_unlock(&block_group->lock);
2800                 goto out_put;
2801         }
2802         spin_unlock(&block_group->lock);
2803
2804         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2805         if (!num_pages)
2806                 num_pages = 1;
2807
2808         /*
2809          * Just to make absolutely sure we have enough space, we're going to
2810          * preallocate 12 pages worth of space for each block group.  In
2811          * practice we ought to use at most 8, but we need extra space so we can
2812          * add our header and have a terminator between the extents and the
2813          * bitmaps.
2814          */
2815         num_pages *= 16;
2816         num_pages *= PAGE_CACHE_SIZE;
2817
2818         ret = btrfs_check_data_free_space(inode, num_pages);
2819         if (ret)
2820                 goto out_put;
2821
2822         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2823                                               num_pages, num_pages,
2824                                               &alloc_hint);
2825         btrfs_free_reserved_data_space(inode, num_pages);
2826 out_put:
2827         iput(inode);
2828 out_free:
2829         btrfs_release_path(root, path);
2830 out:
2831         spin_lock(&block_group->lock);
2832         if (ret)
2833                 block_group->disk_cache_state = BTRFS_DC_ERROR;
2834         else
2835                 block_group->disk_cache_state = BTRFS_DC_SETUP;
2836         spin_unlock(&block_group->lock);
2837
2838         return ret;
2839 }
2840
2841 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2842                                    struct btrfs_root *root)
2843 {
2844         struct btrfs_block_group_cache *cache;
2845         int err = 0;
2846         struct btrfs_path *path;
2847         u64 last = 0;
2848
2849         path = btrfs_alloc_path();
2850         if (!path)
2851                 return -ENOMEM;
2852
2853 again:
2854         while (1) {
2855                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2856                 while (cache) {
2857                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2858                                 break;
2859                         cache = next_block_group(root, cache);
2860                 }
2861                 if (!cache) {
2862                         if (last == 0)
2863                                 break;
2864                         last = 0;
2865                         continue;
2866                 }
2867                 err = cache_save_setup(cache, trans, path);
2868                 last = cache->key.objectid + cache->key.offset;
2869                 btrfs_put_block_group(cache);
2870         }
2871
2872         while (1) {
2873                 if (last == 0) {
2874                         err = btrfs_run_delayed_refs(trans, root,
2875                                                      (unsigned long)-1);
2876                         BUG_ON(err);
2877                 }
2878
2879                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2880                 while (cache) {
2881                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2882                                 btrfs_put_block_group(cache);
2883                                 goto again;
2884                         }
2885
2886                         if (cache->dirty)
2887                                 break;
2888                         cache = next_block_group(root, cache);
2889                 }
2890                 if (!cache) {
2891                         if (last == 0)
2892                                 break;
2893                         last = 0;
2894                         continue;
2895                 }
2896
2897                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2898                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2899                 cache->dirty = 0;
2900                 last = cache->key.objectid + cache->key.offset;
2901
2902                 err = write_one_cache_group(trans, root, path, cache);
2903                 BUG_ON(err);
2904                 btrfs_put_block_group(cache);
2905         }
2906
2907         while (1) {
2908                 /*
2909                  * I don't think this is needed since we're just marking our
2910                  * preallocated extent as written, but just in case it can't
2911                  * hurt.
2912                  */
2913                 if (last == 0) {
2914                         err = btrfs_run_delayed_refs(trans, root,
2915                                                      (unsigned long)-1);
2916                         BUG_ON(err);
2917                 }
2918
2919                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2920                 while (cache) {
2921                         /*
2922                          * Really this shouldn't happen, but it could if we
2923                          * couldn't write the entire preallocated extent and
2924                          * splitting the extent resulted in a new block.
2925                          */
2926                         if (cache->dirty) {
2927                                 btrfs_put_block_group(cache);
2928                                 goto again;
2929                         }
2930                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2931                                 break;
2932                         cache = next_block_group(root, cache);
2933                 }
2934                 if (!cache) {
2935                         if (last == 0)
2936                                 break;
2937                         last = 0;
2938                         continue;
2939                 }
2940
2941                 btrfs_write_out_cache(root, trans, cache, path);
2942
2943                 /*
2944                  * If we didn't have an error then the cache state is still
2945                  * NEED_WRITE, so we can set it to WRITTEN.
2946                  */
2947                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2948                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2949                 last = cache->key.objectid + cache->key.offset;
2950                 btrfs_put_block_group(cache);
2951         }
2952
2953         btrfs_free_path(path);
2954         return 0;
2955 }
2956
2957 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2958 {
2959         struct btrfs_block_group_cache *block_group;
2960         int readonly = 0;
2961
2962         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2963         if (!block_group || block_group->ro)
2964                 readonly = 1;
2965         if (block_group)
2966                 btrfs_put_block_group(block_group);
2967         return readonly;
2968 }
2969
2970 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2971                              u64 total_bytes, u64 bytes_used,
2972                              struct btrfs_space_info **space_info)
2973 {
2974         struct btrfs_space_info *found;
2975         int i;
2976         int factor;
2977
2978         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2979                      BTRFS_BLOCK_GROUP_RAID10))
2980                 factor = 2;
2981         else
2982                 factor = 1;
2983
2984         found = __find_space_info(info, flags);
2985         if (found) {
2986                 spin_lock(&found->lock);
2987                 found->total_bytes += total_bytes;
2988                 found->disk_total += total_bytes * factor;
2989                 found->bytes_used += bytes_used;
2990                 found->disk_used += bytes_used * factor;
2991                 found->full = 0;
2992                 spin_unlock(&found->lock);
2993                 *space_info = found;
2994                 return 0;
2995         }
2996         found = kzalloc(sizeof(*found), GFP_NOFS);
2997         if (!found)
2998                 return -ENOMEM;
2999
3000         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3001                 INIT_LIST_HEAD(&found->block_groups[i]);
3002         init_rwsem(&found->groups_sem);
3003         spin_lock_init(&found->lock);
3004         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3005                                 BTRFS_BLOCK_GROUP_SYSTEM |
3006                                 BTRFS_BLOCK_GROUP_METADATA);
3007         found->total_bytes = total_bytes;
3008         found->disk_total = total_bytes * factor;
3009         found->bytes_used = bytes_used;
3010         found->disk_used = bytes_used * factor;
3011         found->bytes_pinned = 0;
3012         found->bytes_reserved = 0;
3013         found->bytes_readonly = 0;
3014         found->bytes_may_use = 0;
3015         found->full = 0;
3016         found->force_alloc = 0;
3017         *space_info = found;
3018         list_add_rcu(&found->list, &info->space_info);
3019         atomic_set(&found->caching_threads, 0);
3020         return 0;
3021 }
3022
3023 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3024 {
3025         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3026                                    BTRFS_BLOCK_GROUP_RAID1 |
3027                                    BTRFS_BLOCK_GROUP_RAID10 |
3028                                    BTRFS_BLOCK_GROUP_DUP);
3029         if (extra_flags) {
3030                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3031                         fs_info->avail_data_alloc_bits |= extra_flags;
3032                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3033                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3034                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3035                         fs_info->avail_system_alloc_bits |= extra_flags;
3036         }
3037 }
3038
3039 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3040 {
3041         u64 num_devices = root->fs_info->fs_devices->rw_devices;
3042
3043         if (num_devices == 1)
3044                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3045         if (num_devices < 4)
3046                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3047
3048         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3049             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3050                       BTRFS_BLOCK_GROUP_RAID10))) {
3051                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3052         }
3053
3054         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3055             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3056                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3057         }
3058
3059         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3060             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3061              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3062              (flags & BTRFS_BLOCK_GROUP_DUP)))
3063                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3064         return flags;
3065 }
3066
3067 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3068 {
3069         if (flags & BTRFS_BLOCK_GROUP_DATA)
3070                 flags |= root->fs_info->avail_data_alloc_bits &
3071                          root->fs_info->data_alloc_profile;
3072         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3073                 flags |= root->fs_info->avail_system_alloc_bits &
3074                          root->fs_info->system_alloc_profile;
3075         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3076                 flags |= root->fs_info->avail_metadata_alloc_bits &
3077                          root->fs_info->metadata_alloc_profile;
3078         return btrfs_reduce_alloc_profile(root, flags);
3079 }
3080
3081 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3082 {
3083         u64 flags;
3084
3085         if (data)
3086                 flags = BTRFS_BLOCK_GROUP_DATA;
3087         else if (root == root->fs_info->chunk_root)
3088                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3089         else
3090                 flags = BTRFS_BLOCK_GROUP_METADATA;
3091
3092         return get_alloc_profile(root, flags);
3093 }
3094
3095 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3096 {
3097         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3098                                                        BTRFS_BLOCK_GROUP_DATA);
3099 }
3100
3101 /*
3102  * This will check the space that the inode allocates from to make sure we have
3103  * enough space for bytes.
3104  */
3105 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3106 {
3107         struct btrfs_space_info *data_sinfo;
3108         struct btrfs_root *root = BTRFS_I(inode)->root;
3109         u64 used;
3110         int ret = 0, committed = 0, alloc_chunk = 1;
3111
3112         /* make sure bytes are sectorsize aligned */
3113         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3114
3115         if (root == root->fs_info->tree_root) {
3116                 alloc_chunk = 0;
3117                 committed = 1;
3118         }
3119
3120         data_sinfo = BTRFS_I(inode)->space_info;
3121         if (!data_sinfo)
3122                 goto alloc;
3123
3124 again:
3125         /* make sure we have enough space to handle the data first */
3126         spin_lock(&data_sinfo->lock);
3127         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3128                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3129                 data_sinfo->bytes_may_use;
3130
3131         if (used + bytes > data_sinfo->total_bytes) {
3132                 struct btrfs_trans_handle *trans;
3133
3134                 /*
3135                  * if we don't have enough free bytes in this space then we need
3136                  * to alloc a new chunk.
3137                  */
3138                 if (!data_sinfo->full && alloc_chunk) {
3139                         u64 alloc_target;
3140
3141                         data_sinfo->force_alloc = 1;
3142                         spin_unlock(&data_sinfo->lock);
3143 alloc:
3144                         alloc_target = btrfs_get_alloc_profile(root, 1);
3145                         trans = btrfs_join_transaction(root, 1);
3146                         if (IS_ERR(trans))
3147                                 return PTR_ERR(trans);
3148
3149                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3150                                              bytes + 2 * 1024 * 1024,
3151                                              alloc_target, 0);
3152                         btrfs_end_transaction(trans, root);
3153                         if (ret < 0)
3154                                 return ret;
3155
3156                         if (!data_sinfo) {
3157                                 btrfs_set_inode_space_info(root, inode);
3158                                 data_sinfo = BTRFS_I(inode)->space_info;
3159                         }
3160                         goto again;
3161                 }
3162                 spin_unlock(&data_sinfo->lock);
3163
3164                 /* commit the current transaction and try again */
3165                 if (!committed && !root->fs_info->open_ioctl_trans) {
3166                         committed = 1;
3167                         trans = btrfs_join_transaction(root, 1);
3168                         if (IS_ERR(trans))
3169                                 return PTR_ERR(trans);
3170                         ret = btrfs_commit_transaction(trans, root);
3171                         if (ret)
3172                                 return ret;
3173                         goto again;
3174                 }
3175
3176 #if 0 /* I hope we never need this code again, just in case */
3177                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3178                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3179                        "%llu bytes_readonly, %llu may use %llu total\n",
3180                        (unsigned long long)bytes,
3181                        (unsigned long long)data_sinfo->bytes_used,
3182                        (unsigned long long)data_sinfo->bytes_reserved,
3183                        (unsigned long long)data_sinfo->bytes_pinned,
3184                        (unsigned long long)data_sinfo->bytes_readonly,
3185                        (unsigned long long)data_sinfo->bytes_may_use,
3186                        (unsigned long long)data_sinfo->total_bytes);
3187 #endif
3188                 return -ENOSPC;
3189         }
3190         data_sinfo->bytes_may_use += bytes;
3191         BTRFS_I(inode)->reserved_bytes += bytes;
3192         spin_unlock(&data_sinfo->lock);
3193
3194         return 0;
3195 }
3196
3197 /*
3198  * called when we are clearing an delalloc extent from the
3199  * inode's io_tree or there was an error for whatever reason
3200  * after calling btrfs_check_data_free_space
3201  */
3202 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3203 {
3204         struct btrfs_root *root = BTRFS_I(inode)->root;
3205         struct btrfs_space_info *data_sinfo;
3206
3207         /* make sure bytes are sectorsize aligned */
3208         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3209
3210         data_sinfo = BTRFS_I(inode)->space_info;
3211         spin_lock(&data_sinfo->lock);
3212         data_sinfo->bytes_may_use -= bytes;
3213         BTRFS_I(inode)->reserved_bytes -= bytes;
3214         spin_unlock(&data_sinfo->lock);
3215 }
3216
3217 static void force_metadata_allocation(struct btrfs_fs_info *info)
3218 {
3219         struct list_head *head = &info->space_info;
3220         struct btrfs_space_info *found;
3221
3222         rcu_read_lock();
3223         list_for_each_entry_rcu(found, head, list) {
3224                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3225                         found->force_alloc = 1;
3226         }
3227         rcu_read_unlock();
3228 }
3229
3230 static int should_alloc_chunk(struct btrfs_root *root,
3231                               struct btrfs_space_info *sinfo, u64 alloc_bytes)
3232 {
3233         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3234         u64 thresh;
3235
3236         if (sinfo->bytes_used + sinfo->bytes_reserved +
3237             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3238                 return 0;
3239
3240         if (sinfo->bytes_used + sinfo->bytes_reserved +
3241             alloc_bytes < div_factor(num_bytes, 8))
3242                 return 0;
3243
3244         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3245         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3246
3247         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3248                 return 0;
3249
3250         return 1;
3251 }
3252
3253 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3254                           struct btrfs_root *extent_root, u64 alloc_bytes,
3255                           u64 flags, int force)
3256 {
3257         struct btrfs_space_info *space_info;
3258         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3259         int ret = 0;
3260
3261         mutex_lock(&fs_info->chunk_mutex);
3262
3263         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3264
3265         space_info = __find_space_info(extent_root->fs_info, flags);
3266         if (!space_info) {
3267                 ret = update_space_info(extent_root->fs_info, flags,
3268                                         0, 0, &space_info);
3269                 BUG_ON(ret);
3270         }
3271         BUG_ON(!space_info);
3272
3273         spin_lock(&space_info->lock);
3274         if (space_info->force_alloc)
3275                 force = 1;
3276         if (space_info->full) {
3277                 spin_unlock(&space_info->lock);
3278                 goto out;
3279         }
3280
3281         if (!force && !should_alloc_chunk(extent_root, space_info,
3282                                           alloc_bytes)) {
3283                 spin_unlock(&space_info->lock);
3284                 goto out;
3285         }
3286         spin_unlock(&space_info->lock);
3287
3288         /*
3289          * If we have mixed data/metadata chunks we want to make sure we keep
3290          * allocating mixed chunks instead of individual chunks.
3291          */
3292         if (btrfs_mixed_space_info(space_info))
3293                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3294
3295         /*
3296          * if we're doing a data chunk, go ahead and make sure that
3297          * we keep a reasonable number of metadata chunks allocated in the
3298          * FS as well.
3299          */
3300         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3301                 fs_info->data_chunk_allocations++;
3302                 if (!(fs_info->data_chunk_allocations %
3303                       fs_info->metadata_ratio))
3304                         force_metadata_allocation(fs_info);
3305         }
3306
3307         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3308         spin_lock(&space_info->lock);
3309         if (ret)
3310                 space_info->full = 1;
3311         else
3312                 ret = 1;
3313         space_info->force_alloc = 0;
3314         spin_unlock(&space_info->lock);
3315 out:
3316         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3317         return ret;
3318 }
3319
3320 /*
3321  * shrink metadata reservation for delalloc
3322  */
3323 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3324                            struct btrfs_root *root, u64 to_reclaim, int sync)
3325 {
3326         struct btrfs_block_rsv *block_rsv;
3327         struct btrfs_space_info *space_info;
3328         u64 reserved;
3329         u64 max_reclaim;
3330         u64 reclaimed = 0;
3331         int pause = 1;
3332         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3333
3334         block_rsv = &root->fs_info->delalloc_block_rsv;
3335         space_info = block_rsv->space_info;
3336
3337         smp_mb();
3338         reserved = space_info->bytes_reserved;
3339
3340         if (reserved == 0)
3341                 return 0;
3342
3343         max_reclaim = min(reserved, to_reclaim);
3344
3345         while (1) {
3346                 /* have the flusher threads jump in and do some IO */
3347                 smp_mb();
3348                 nr_pages = min_t(unsigned long, nr_pages,
3349                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3350                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3351
3352                 spin_lock(&space_info->lock);
3353                 if (reserved > space_info->bytes_reserved)
3354                         reclaimed += reserved - space_info->bytes_reserved;
3355                 reserved = space_info->bytes_reserved;
3356                 spin_unlock(&space_info->lock);
3357
3358                 if (reserved == 0 || reclaimed >= max_reclaim)
3359                         break;
3360
3361                 if (trans && trans->transaction->blocked)
3362                         return -EAGAIN;
3363
3364                 __set_current_state(TASK_INTERRUPTIBLE);
3365                 schedule_timeout(pause);
3366                 pause <<= 1;
3367                 if (pause > HZ / 10)
3368                         pause = HZ / 10;
3369
3370         }
3371         return reclaimed >= to_reclaim;
3372 }
3373
3374 /*
3375  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3376  * idea is if this is the first call (retries == 0) then we will add to our
3377  * reserved count if we can't make the allocation in order to hold our place
3378  * while we go and try and free up space.  That way for retries > 1 we don't try
3379  * and add space, we just check to see if the amount of unused space is >= the
3380  * total space, meaning that our reservation is valid.
3381  *
3382  * However if we don't intend to retry this reservation, pass -1 as retries so
3383  * that it short circuits this logic.
3384  */
3385 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3386                                   struct btrfs_root *root,
3387                                   struct btrfs_block_rsv *block_rsv,
3388                                   u64 orig_bytes, int flush)
3389 {
3390         struct btrfs_space_info *space_info = block_rsv->space_info;
3391         u64 unused;
3392         u64 num_bytes = orig_bytes;
3393         int retries = 0;
3394         int ret = 0;
3395         bool reserved = false;
3396         bool committed = false;
3397
3398 again:
3399         ret = -ENOSPC;
3400         if (reserved)
3401                 num_bytes = 0;
3402
3403         spin_lock(&space_info->lock);
3404         unused = space_info->bytes_used + space_info->bytes_reserved +
3405                  space_info->bytes_pinned + space_info->bytes_readonly +
3406                  space_info->bytes_may_use;
3407
3408         /*
3409          * The idea here is that we've not already over-reserved the block group
3410          * then we can go ahead and save our reservation first and then start
3411          * flushing if we need to.  Otherwise if we've already overcommitted
3412          * lets start flushing stuff first and then come back and try to make
3413          * our reservation.
3414          */
3415         if (unused <= space_info->total_bytes) {
3416                 unused -= space_info->total_bytes;
3417                 if (unused >= num_bytes) {
3418                         if (!reserved)
3419                                 space_info->bytes_reserved += orig_bytes;
3420                         ret = 0;
3421                 } else {
3422                         /*
3423                          * Ok set num_bytes to orig_bytes since we aren't
3424                          * overocmmitted, this way we only try and reclaim what
3425                          * we need.
3426                          */
3427                         num_bytes = orig_bytes;
3428                 }
3429         } else {
3430                 /*
3431                  * Ok we're over committed, set num_bytes to the overcommitted
3432                  * amount plus the amount of bytes that we need for this
3433                  * reservation.
3434                  */
3435                 num_bytes = unused - space_info->total_bytes +
3436                         (orig_bytes * (retries + 1));
3437         }
3438
3439         /*
3440          * Couldn't make our reservation, save our place so while we're trying
3441          * to reclaim space we can actually use it instead of somebody else
3442          * stealing it from us.
3443          */
3444         if (ret && !reserved) {
3445                 space_info->bytes_reserved += orig_bytes;
3446                 reserved = true;
3447         }
3448
3449         spin_unlock(&space_info->lock);
3450
3451         if (!ret)
3452                 return 0;
3453
3454         if (!flush)
3455                 goto out;
3456
3457         /*
3458          * We do synchronous shrinking since we don't actually unreserve
3459          * metadata until after the IO is completed.
3460          */
3461         ret = shrink_delalloc(trans, root, num_bytes, 1);
3462         if (ret > 0)
3463                 return 0;
3464         else if (ret < 0)
3465                 goto out;
3466
3467         /*
3468          * So if we were overcommitted it's possible that somebody else flushed
3469          * out enough space and we simply didn't have enough space to reclaim,
3470          * so go back around and try again.
3471          */
3472         if (retries < 2) {
3473                 retries++;
3474                 goto again;
3475         }
3476
3477         spin_lock(&space_info->lock);
3478         /*
3479          * Not enough space to be reclaimed, don't bother committing the
3480          * transaction.
3481          */
3482         if (space_info->bytes_pinned < orig_bytes)
3483                 ret = -ENOSPC;
3484         spin_unlock(&space_info->lock);
3485         if (ret)
3486                 goto out;
3487
3488         ret = -EAGAIN;
3489         if (trans || committed)
3490                 goto out;
3491
3492         ret = -ENOSPC;
3493         trans = btrfs_join_transaction(root, 1);
3494         if (IS_ERR(trans))
3495                 goto out;
3496         ret = btrfs_commit_transaction(trans, root);
3497         if (!ret) {
3498                 trans = NULL;
3499                 committed = true;
3500                 goto again;
3501         }
3502
3503 out:
3504         if (reserved) {
3505                 spin_lock(&space_info->lock);
3506                 space_info->bytes_reserved -= orig_bytes;
3507                 spin_unlock(&space_info->lock);
3508         }
3509
3510         return ret;
3511 }
3512
3513 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3514                                              struct btrfs_root *root)
3515 {
3516         struct btrfs_block_rsv *block_rsv;
3517         if (root->ref_cows)
3518                 block_rsv = trans->block_rsv;
3519         else
3520                 block_rsv = root->block_rsv;
3521
3522         if (!block_rsv)
3523                 block_rsv = &root->fs_info->empty_block_rsv;
3524
3525         return block_rsv;
3526 }
3527
3528 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3529                                u64 num_bytes)
3530 {
3531         int ret = -ENOSPC;
3532         spin_lock(&block_rsv->lock);
3533         if (block_rsv->reserved >= num_bytes) {
3534                 block_rsv->reserved -= num_bytes;
3535                 if (block_rsv->reserved < block_rsv->size)
3536                         block_rsv->full = 0;
3537                 ret = 0;
3538         }
3539         spin_unlock(&block_rsv->lock);
3540         return ret;
3541 }
3542
3543 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3544                                 u64 num_bytes, int update_size)
3545 {
3546         spin_lock(&block_rsv->lock);
3547         block_rsv->reserved += num_bytes;
3548         if (update_size)
3549                 block_rsv->size += num_bytes;
3550         else if (block_rsv->reserved >= block_rsv->size)
3551                 block_rsv->full = 1;
3552         spin_unlock(&block_rsv->lock);
3553 }
3554
3555 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3556                              struct btrfs_block_rsv *dest, u64 num_bytes)
3557 {
3558         struct btrfs_space_info *space_info = block_rsv->space_info;
3559
3560         spin_lock(&block_rsv->lock);
3561         if (num_bytes == (u64)-1)
3562                 num_bytes = block_rsv->size;
3563         block_rsv->size -= num_bytes;
3564         if (block_rsv->reserved >= block_rsv->size) {
3565                 num_bytes = block_rsv->reserved - block_rsv->size;
3566                 block_rsv->reserved = block_rsv->size;
3567                 block_rsv->full = 1;
3568         } else {
3569                 num_bytes = 0;
3570         }
3571         spin_unlock(&block_rsv->lock);
3572
3573         if (num_bytes > 0) {
3574                 if (dest) {
3575                         block_rsv_add_bytes(dest, num_bytes, 0);
3576                 } else {
3577                         spin_lock(&space_info->lock);
3578                         space_info->bytes_reserved -= num_bytes;
3579                         spin_unlock(&space_info->lock);
3580                 }
3581         }
3582 }
3583
3584 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3585                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3586 {
3587         int ret;
3588
3589         ret = block_rsv_use_bytes(src, num_bytes);
3590         if (ret)
3591                 return ret;
3592
3593         block_rsv_add_bytes(dst, num_bytes, 1);
3594         return 0;
3595 }
3596
3597 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3598 {
3599         memset(rsv, 0, sizeof(*rsv));
3600         spin_lock_init(&rsv->lock);
3601         atomic_set(&rsv->usage, 1);
3602         rsv->priority = 6;
3603         INIT_LIST_HEAD(&rsv->list);
3604 }
3605
3606 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3607 {
3608         struct btrfs_block_rsv *block_rsv;
3609         struct btrfs_fs_info *fs_info = root->fs_info;
3610         u64 alloc_target;
3611
3612         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3613         if (!block_rsv)
3614                 return NULL;
3615
3616         btrfs_init_block_rsv(block_rsv);
3617
3618         alloc_target = btrfs_get_alloc_profile(root, 0);
3619         block_rsv->space_info = __find_space_info(fs_info,
3620                                                   BTRFS_BLOCK_GROUP_METADATA);
3621
3622         return block_rsv;
3623 }
3624
3625 void btrfs_free_block_rsv(struct btrfs_root *root,
3626                           struct btrfs_block_rsv *rsv)
3627 {
3628         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3629                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3630                 if (!rsv->durable)
3631                         kfree(rsv);
3632         }
3633 }
3634
3635 /*
3636  * make the block_rsv struct be able to capture freed space.
3637  * the captured space will re-add to the the block_rsv struct
3638  * after transaction commit
3639  */
3640 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3641                                  struct btrfs_block_rsv *block_rsv)
3642 {
3643         block_rsv->durable = 1;
3644         mutex_lock(&fs_info->durable_block_rsv_mutex);
3645         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3646         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3647 }
3648
3649 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3650                         struct btrfs_root *root,
3651                         struct btrfs_block_rsv *block_rsv,
3652                         u64 num_bytes)
3653 {
3654         int ret;
3655
3656         if (num_bytes == 0)
3657                 return 0;
3658
3659         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3660         if (!ret) {
3661                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3662                 return 0;
3663         }
3664
3665         return ret;
3666 }
3667
3668 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3669                           struct btrfs_root *root,
3670                           struct btrfs_block_rsv *block_rsv,
3671                           u64 min_reserved, int min_factor)
3672 {
3673         u64 num_bytes = 0;
3674         int commit_trans = 0;
3675         int ret = -ENOSPC;
3676
3677         if (!block_rsv)
3678                 return 0;
3679
3680         spin_lock(&block_rsv->lock);
3681         if (min_factor > 0)
3682                 num_bytes = div_factor(block_rsv->size, min_factor);
3683         if (min_reserved > num_bytes)
3684                 num_bytes = min_reserved;
3685
3686         if (block_rsv->reserved >= num_bytes) {
3687                 ret = 0;
3688         } else {
3689                 num_bytes -= block_rsv->reserved;
3690                 if (block_rsv->durable &&
3691                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3692                         commit_trans = 1;
3693         }
3694         spin_unlock(&block_rsv->lock);
3695         if (!ret)
3696                 return 0;
3697
3698         if (block_rsv->refill_used) {
3699                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3700                                              num_bytes, 0);
3701                 if (!ret) {
3702                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3703                         return 0;
3704                 }
3705         }
3706
3707         if (commit_trans) {
3708                 if (trans)
3709                         return -EAGAIN;
3710
3711                 trans = btrfs_join_transaction(root, 1);
3712                 BUG_ON(IS_ERR(trans));
3713                 ret = btrfs_commit_transaction(trans, root);
3714                 return 0;
3715         }
3716
3717         WARN_ON(1);
3718         printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
3719                 block_rsv->size, block_rsv->reserved,
3720                 block_rsv->freed[0], block_rsv->freed[1]);
3721
3722         return -ENOSPC;
3723 }
3724
3725 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3726                             struct btrfs_block_rsv *dst_rsv,
3727                             u64 num_bytes)
3728 {
3729         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3730 }
3731
3732 void btrfs_block_rsv_release(struct btrfs_root *root,
3733                              struct btrfs_block_rsv *block_rsv,
3734                              u64 num_bytes)
3735 {
3736         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3737         if (global_rsv->full || global_rsv == block_rsv ||
3738             block_rsv->space_info != global_rsv->space_info)
3739                 global_rsv = NULL;
3740         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3741 }
3742
3743 /*
3744  * helper to calculate size of global block reservation.
3745  * the desired value is sum of space used by extent tree,
3746  * checksum tree and root tree
3747  */
3748 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3749 {
3750         struct btrfs_space_info *sinfo;
3751         u64 num_bytes;
3752         u64 meta_used;
3753         u64 data_used;
3754         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3755 #if 0
3756         /*
3757          * per tree used space accounting can be inaccuracy, so we
3758          * can't rely on it.
3759          */
3760         spin_lock(&fs_info->extent_root->accounting_lock);
3761         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3762         spin_unlock(&fs_info->extent_root->accounting_lock);
3763
3764         spin_lock(&fs_info->csum_root->accounting_lock);
3765         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3766         spin_unlock(&fs_info->csum_root->accounting_lock);
3767
3768         spin_lock(&fs_info->tree_root->accounting_lock);
3769         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3770         spin_unlock(&fs_info->tree_root->accounting_lock);
3771 #endif
3772         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3773         spin_lock(&sinfo->lock);
3774         data_used = sinfo->bytes_used;
3775         spin_unlock(&sinfo->lock);
3776
3777         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3778         spin_lock(&sinfo->lock);
3779         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3780                 data_used = 0;
3781         meta_used = sinfo->bytes_used;
3782         spin_unlock(&sinfo->lock);
3783
3784         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3785                     csum_size * 2;
3786         num_bytes += div64_u64(data_used + meta_used, 50);
3787
3788         if (num_bytes * 3 > meta_used)
3789                 num_bytes = div64_u64(meta_used, 3);
3790
3791         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3792 }
3793
3794 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3795 {
3796         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3797         struct btrfs_space_info *sinfo = block_rsv->space_info;
3798         u64 num_bytes;
3799
3800         num_bytes = calc_global_metadata_size(fs_info);
3801
3802         spin_lock(&block_rsv->lock);
3803         spin_lock(&sinfo->lock);
3804
3805         block_rsv->size = num_bytes;
3806
3807         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3808                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3809                     sinfo->bytes_may_use;
3810
3811         if (sinfo->total_bytes > num_bytes) {
3812                 num_bytes = sinfo->total_bytes - num_bytes;
3813                 block_rsv->reserved += num_bytes;
3814                 sinfo->bytes_reserved += num_bytes;
3815         }
3816
3817         if (block_rsv->reserved >= block_rsv->size) {
3818                 num_bytes = block_rsv->reserved - block_rsv->size;
3819                 sinfo->bytes_reserved -= num_bytes;
3820                 block_rsv->reserved = block_rsv->size;
3821                 block_rsv->full = 1;
3822         }
3823 #if 0
3824         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3825                 block_rsv->size, block_rsv->reserved);
3826 #endif
3827         spin_unlock(&sinfo->lock);
3828         spin_unlock(&block_rsv->lock);
3829 }
3830
3831 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3832 {
3833         struct btrfs_space_info *space_info;
3834
3835         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3836         fs_info->chunk_block_rsv.space_info = space_info;
3837         fs_info->chunk_block_rsv.priority = 10;
3838
3839         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3840         fs_info->global_block_rsv.space_info = space_info;
3841         fs_info->global_block_rsv.priority = 10;
3842         fs_info->global_block_rsv.refill_used = 1;
3843         fs_info->delalloc_block_rsv.space_info = space_info;
3844         fs_info->trans_block_rsv.space_info = space_info;
3845         fs_info->empty_block_rsv.space_info = space_info;
3846         fs_info->empty_block_rsv.priority = 10;
3847
3848         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3849         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3850         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3851         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3852         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3853
3854         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3855
3856         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3857
3858         update_global_block_rsv(fs_info);
3859 }
3860
3861 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3862 {
3863         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3864         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3865         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3866         WARN_ON(fs_info->trans_block_rsv.size > 0);
3867         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3868         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3869         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3870 }
3871
3872 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3873 {
3874         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3875                 3 * num_items;
3876 }
3877
3878 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3879                                  struct btrfs_root *root,
3880                                  int num_items)
3881 {
3882         u64 num_bytes;
3883         int ret;
3884
3885         if (num_items == 0 || root->fs_info->chunk_root == root)
3886                 return 0;
3887
3888         num_bytes = calc_trans_metadata_size(root, num_items);
3889         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3890                                   num_bytes);
3891         if (!ret) {
3892                 trans->bytes_reserved += num_bytes;
3893                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3894         }
3895         return ret;
3896 }
3897
3898 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3899                                   struct btrfs_root *root)
3900 {
3901         if (!trans->bytes_reserved)
3902                 return;
3903
3904         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3905         btrfs_block_rsv_release(root, trans->block_rsv,
3906                                 trans->bytes_reserved);
3907         trans->bytes_reserved = 0;
3908 }
3909
3910 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3911                                   struct inode *inode)
3912 {
3913         struct btrfs_root *root = BTRFS_I(inode)->root;
3914         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3915         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3916
3917         /*
3918          * one for deleting orphan item, one for updating inode and
3919          * two for calling btrfs_truncate_inode_items.
3920          *
3921          * btrfs_truncate_inode_items is a delete operation, it frees
3922          * more space than it uses in most cases. So two units of
3923          * metadata space should be enough for calling it many times.
3924          * If all of the metadata space is used, we can commit
3925          * transaction and use space it freed.
3926          */
3927         u64 num_bytes = calc_trans_metadata_size(root, 4);
3928         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3929 }
3930
3931 void btrfs_orphan_release_metadata(struct inode *inode)
3932 {
3933         struct btrfs_root *root = BTRFS_I(inode)->root;
3934         u64 num_bytes = calc_trans_metadata_size(root, 4);
3935         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3936 }
3937
3938 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3939                                 struct btrfs_pending_snapshot *pending)
3940 {
3941         struct btrfs_root *root = pending->root;
3942         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3943         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3944         /*
3945          * two for root back/forward refs, two for directory entries
3946          * and one for root of the snapshot.
3947          */
3948         u64 num_bytes = calc_trans_metadata_size(root, 5);
3949         dst_rsv->space_info = src_rsv->space_info;
3950         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3951 }
3952
3953 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3954 {
3955         return num_bytes >>= 3;
3956 }
3957
3958 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3959 {
3960         struct btrfs_root *root = BTRFS_I(inode)->root;
3961         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3962         u64 to_reserve;
3963         int nr_extents;
3964         int ret;
3965
3966         if (btrfs_transaction_in_commit(root->fs_info))
3967                 schedule_timeout(1);
3968
3969         num_bytes = ALIGN(num_bytes, root->sectorsize);
3970
3971         spin_lock(&BTRFS_I(inode)->accounting_lock);
3972         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
3973         if (nr_extents > BTRFS_I(inode)->reserved_extents) {
3974                 nr_extents -= BTRFS_I(inode)->reserved_extents;
3975                 to_reserve = calc_trans_metadata_size(root, nr_extents);
3976         } else {
3977                 nr_extents = 0;
3978                 to_reserve = 0;
3979         }
3980         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3981
3982         to_reserve += calc_csum_metadata_size(inode, num_bytes);
3983         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
3984         if (ret)
3985                 return ret;
3986
3987         spin_lock(&BTRFS_I(inode)->accounting_lock);
3988         BTRFS_I(inode)->reserved_extents += nr_extents;
3989         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
3990         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3991
3992         block_rsv_add_bytes(block_rsv, to_reserve, 1);
3993
3994         if (block_rsv->size > 512 * 1024 * 1024)
3995                 shrink_delalloc(NULL, root, to_reserve, 0);
3996
3997         return 0;
3998 }
3999
4000 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4001 {
4002         struct btrfs_root *root = BTRFS_I(inode)->root;
4003         u64 to_free;
4004         int nr_extents;
4005
4006         num_bytes = ALIGN(num_bytes, root->sectorsize);
4007         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4008
4009         spin_lock(&BTRFS_I(inode)->accounting_lock);
4010         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4011         if (nr_extents < BTRFS_I(inode)->reserved_extents) {
4012                 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
4013                 BTRFS_I(inode)->reserved_extents -= nr_extents;
4014         } else {
4015                 nr_extents = 0;
4016         }
4017         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4018
4019         to_free = calc_csum_metadata_size(inode, num_bytes);
4020         if (nr_extents > 0)
4021                 to_free += calc_trans_metadata_size(root, nr_extents);
4022
4023         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4024                                 to_free);
4025 }
4026
4027 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4028 {
4029         int ret;
4030
4031         ret = btrfs_check_data_free_space(inode, num_bytes);
4032         if (ret)
4033                 return ret;
4034
4035         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4036         if (ret) {
4037                 btrfs_free_reserved_data_space(inode, num_bytes);
4038                 return ret;
4039         }
4040
4041         return 0;
4042 }
4043
4044 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4045 {
4046         btrfs_delalloc_release_metadata(inode, num_bytes);
4047         btrfs_free_reserved_data_space(inode, num_bytes);
4048 }
4049
4050 static int update_block_group(struct btrfs_trans_handle *trans,
4051                               struct btrfs_root *root,
4052                               u64 bytenr, u64 num_bytes, int alloc)
4053 {
4054         struct btrfs_block_group_cache *cache = NULL;
4055         struct btrfs_fs_info *info = root->fs_info;
4056         u64 total = num_bytes;
4057         u64 old_val;
4058         u64 byte_in_group;
4059         int factor;
4060
4061         /* block accounting for super block */
4062         spin_lock(&info->delalloc_lock);
4063         old_val = btrfs_super_bytes_used(&info->super_copy);
4064         if (alloc)
4065                 old_val += num_bytes;
4066         else
4067                 old_val -= num_bytes;
4068         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4069         spin_unlock(&info->delalloc_lock);
4070
4071         while (total) {
4072                 cache = btrfs_lookup_block_group(info, bytenr);
4073                 if (!cache)
4074                         return -1;
4075                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4076                                     BTRFS_BLOCK_GROUP_RAID1 |
4077                                     BTRFS_BLOCK_GROUP_RAID10))
4078                         factor = 2;
4079                 else
4080                         factor = 1;
4081                 /*
4082                  * If this block group has free space cache written out, we
4083                  * need to make sure to load it if we are removing space.  This
4084                  * is because we need the unpinning stage to actually add the
4085                  * space back to the block group, otherwise we will leak space.
4086                  */
4087                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4088                         cache_block_group(cache, trans, 1);
4089
4090                 byte_in_group = bytenr - cache->key.objectid;
4091                 WARN_ON(byte_in_group > cache->key.offset);
4092
4093                 spin_lock(&cache->space_info->lock);
4094                 spin_lock(&cache->lock);
4095
4096                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4097                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4098                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4099
4100                 cache->dirty = 1;
4101                 old_val = btrfs_block_group_used(&cache->item);
4102                 num_bytes = min(total, cache->key.offset - byte_in_group);
4103                 if (alloc) {
4104                         old_val += num_bytes;
4105                         btrfs_set_block_group_used(&cache->item, old_val);
4106                         cache->reserved -= num_bytes;
4107                         cache->space_info->bytes_reserved -= num_bytes;
4108                         cache->space_info->bytes_used += num_bytes;
4109                         cache->space_info->disk_used += num_bytes * factor;
4110                         spin_unlock(&cache->lock);
4111                         spin_unlock(&cache->space_info->lock);
4112                 } else {
4113                         old_val -= num_bytes;
4114                         btrfs_set_block_group_used(&cache->item, old_val);
4115                         cache->pinned += num_bytes;
4116                         cache->space_info->bytes_pinned += num_bytes;
4117                         cache->space_info->bytes_used -= num_bytes;
4118                         cache->space_info->disk_used -= num_bytes * factor;
4119                         spin_unlock(&cache->lock);
4120                         spin_unlock(&cache->space_info->lock);
4121
4122                         set_extent_dirty(info->pinned_extents,
4123                                          bytenr, bytenr + num_bytes - 1,
4124                                          GFP_NOFS | __GFP_NOFAIL);
4125                 }
4126                 btrfs_put_block_group(cache);
4127                 total -= num_bytes;
4128                 bytenr += num_bytes;
4129         }
4130         return 0;
4131 }
4132
4133 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4134 {
4135         struct btrfs_block_group_cache *cache;
4136         u64 bytenr;
4137
4138         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4139         if (!cache)
4140                 return 0;
4141
4142         bytenr = cache->key.objectid;
4143         btrfs_put_block_group(cache);
4144
4145         return bytenr;
4146 }
4147
4148 static int pin_down_extent(struct btrfs_root *root,
4149                            struct btrfs_block_group_cache *cache,
4150                            u64 bytenr, u64 num_bytes, int reserved)
4151 {
4152         spin_lock(&cache->space_info->lock);
4153         spin_lock(&cache->lock);
4154         cache->pinned += num_bytes;
4155         cache->space_info->bytes_pinned += num_bytes;
4156         if (reserved) {
4157                 cache->reserved -= num_bytes;
4158                 cache->space_info->bytes_reserved -= num_bytes;
4159         }
4160         spin_unlock(&cache->lock);
4161         spin_unlock(&cache->space_info->lock);
4162
4163         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4164                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4165         return 0;
4166 }
4167
4168 /*
4169  * this function must be called within transaction
4170  */
4171 int btrfs_pin_extent(struct btrfs_root *root,
4172                      u64 bytenr, u64 num_bytes, int reserved)
4173 {
4174         struct btrfs_block_group_cache *cache;
4175
4176         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4177         BUG_ON(!cache);
4178
4179         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4180
4181         btrfs_put_block_group(cache);
4182         return 0;
4183 }
4184
4185 /*
4186  * update size of reserved extents. this function may return -EAGAIN
4187  * if 'reserve' is true or 'sinfo' is false.
4188  */
4189 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4190                                  u64 num_bytes, int reserve, int sinfo)
4191 {
4192         int ret = 0;
4193         if (sinfo) {
4194                 struct btrfs_space_info *space_info = cache->space_info;
4195                 spin_lock(&space_info->lock);
4196                 spin_lock(&cache->lock);
4197                 if (reserve) {
4198                         if (cache->ro) {
4199                                 ret = -EAGAIN;
4200                         } else {
4201                                 cache->reserved += num_bytes;
4202                                 space_info->bytes_reserved += num_bytes;
4203                         }
4204                 } else {
4205                         if (cache->ro)
4206                                 space_info->bytes_readonly += num_bytes;
4207                         cache->reserved -= num_bytes;
4208                         space_info->bytes_reserved -= num_bytes;
4209                 }
4210                 spin_unlock(&cache->lock);
4211                 spin_unlock(&space_info->lock);
4212         } else {
4213                 spin_lock(&cache->lock);
4214                 if (cache->ro) {
4215                         ret = -EAGAIN;
4216                 } else {
4217                         if (reserve)
4218                                 cache->reserved += num_bytes;
4219                         else
4220                                 cache->reserved -= num_bytes;
4221                 }
4222                 spin_unlock(&cache->lock);
4223         }
4224         return ret;
4225 }
4226
4227 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4228                                 struct btrfs_root *root)
4229 {
4230         struct btrfs_fs_info *fs_info = root->fs_info;
4231         struct btrfs_caching_control *next;
4232         struct btrfs_caching_control *caching_ctl;
4233         struct btrfs_block_group_cache *cache;
4234
4235         down_write(&fs_info->extent_commit_sem);
4236
4237         list_for_each_entry_safe(caching_ctl, next,
4238                                  &fs_info->caching_block_groups, list) {
4239                 cache = caching_ctl->block_group;
4240                 if (block_group_cache_done(cache)) {
4241                         cache->last_byte_to_unpin = (u64)-1;
4242                         list_del_init(&caching_ctl->list);
4243                         put_caching_control(caching_ctl);
4244                 } else {
4245                         cache->last_byte_to_unpin = caching_ctl->progress;
4246                 }
4247         }
4248
4249         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4250                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4251         else
4252                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4253
4254         up_write(&fs_info->extent_commit_sem);
4255
4256         update_global_block_rsv(fs_info);
4257         return 0;
4258 }
4259
4260 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4261 {
4262         struct btrfs_fs_info *fs_info = root->fs_info;
4263         struct btrfs_block_group_cache *cache = NULL;
4264         u64 len;
4265
4266         while (start <= end) {
4267                 if (!cache ||
4268                     start >= cache->key.objectid + cache->key.offset) {
4269                         if (cache)
4270                                 btrfs_put_block_group(cache);
4271                         cache = btrfs_lookup_block_group(fs_info, start);
4272                         BUG_ON(!cache);
4273                 }
4274
4275                 len = cache->key.objectid + cache->key.offset - start;
4276                 len = min(len, end + 1 - start);
4277
4278                 if (start < cache->last_byte_to_unpin) {
4279                         len = min(len, cache->last_byte_to_unpin - start);
4280                         btrfs_add_free_space(cache, start, len);
4281                 }
4282
4283                 start += len;
4284
4285                 spin_lock(&cache->space_info->lock);
4286                 spin_lock(&cache->lock);
4287                 cache->pinned -= len;
4288                 cache->space_info->bytes_pinned -= len;
4289                 if (cache->ro) {
4290                         cache->space_info->bytes_readonly += len;
4291                 } else if (cache->reserved_pinned > 0) {
4292                         len = min(len, cache->reserved_pinned);
4293                         cache->reserved_pinned -= len;
4294                         cache->space_info->bytes_reserved += len;
4295                 }
4296                 spin_unlock(&cache->lock);
4297                 spin_unlock(&cache->space_info->lock);
4298         }
4299
4300         if (cache)
4301                 btrfs_put_block_group(cache);
4302         return 0;
4303 }
4304
4305 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4306                                struct btrfs_root *root)
4307 {
4308         struct btrfs_fs_info *fs_info = root->fs_info;
4309         struct extent_io_tree *unpin;
4310         struct btrfs_block_rsv *block_rsv;
4311         struct btrfs_block_rsv *next_rsv;
4312         u64 start;
4313         u64 end;
4314         int idx;
4315         int ret;
4316
4317         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4318                 unpin = &fs_info->freed_extents[1];
4319         else
4320                 unpin = &fs_info->freed_extents[0];
4321
4322         while (1) {
4323                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4324                                             EXTENT_DIRTY);
4325                 if (ret)
4326                         break;
4327
4328                 ret = btrfs_discard_extent(root, start, end + 1 - start);
4329
4330                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4331                 unpin_extent_range(root, start, end);
4332                 cond_resched();
4333         }
4334
4335         mutex_lock(&fs_info->durable_block_rsv_mutex);
4336         list_for_each_entry_safe(block_rsv, next_rsv,
4337                                  &fs_info->durable_block_rsv_list, list) {
4338
4339                 idx = trans->transid & 0x1;
4340                 if (block_rsv->freed[idx] > 0) {
4341                         block_rsv_add_bytes(block_rsv,
4342                                             block_rsv->freed[idx], 0);
4343                         block_rsv->freed[idx] = 0;
4344                 }
4345                 if (atomic_read(&block_rsv->usage) == 0) {
4346                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4347
4348                         if (block_rsv->freed[0] == 0 &&
4349                             block_rsv->freed[1] == 0) {
4350                                 list_del_init(&block_rsv->list);
4351                                 kfree(block_rsv);
4352                         }
4353                 } else {
4354                         btrfs_block_rsv_release(root, block_rsv, 0);
4355                 }
4356         }
4357         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4358
4359         return 0;
4360 }
4361
4362 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4363                                 struct btrfs_root *root,
4364                                 u64 bytenr, u64 num_bytes, u64 parent,
4365                                 u64 root_objectid, u64 owner_objectid,
4366                                 u64 owner_offset, int refs_to_drop,
4367                                 struct btrfs_delayed_extent_op *extent_op)
4368 {
4369         struct btrfs_key key;
4370         struct btrfs_path *path;
4371         struct btrfs_fs_info *info = root->fs_info;
4372         struct btrfs_root *extent_root = info->extent_root;
4373         struct extent_buffer *leaf;
4374         struct btrfs_extent_item *ei;
4375         struct btrfs_extent_inline_ref *iref;
4376         int ret;
4377         int is_data;
4378         int extent_slot = 0;
4379         int found_extent = 0;
4380         int num_to_del = 1;
4381         u32 item_size;
4382         u64 refs;
4383
4384         path = btrfs_alloc_path();
4385         if (!path)
4386                 return -ENOMEM;
4387
4388         path->reada = 1;
4389         path->leave_spinning = 1;
4390
4391         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4392         BUG_ON(!is_data && refs_to_drop != 1);
4393
4394         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4395                                     bytenr, num_bytes, parent,
4396                                     root_objectid, owner_objectid,
4397                                     owner_offset);
4398         if (ret == 0) {
4399                 extent_slot = path->slots[0];
4400                 while (extent_slot >= 0) {
4401                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4402                                               extent_slot);
4403                         if (key.objectid != bytenr)
4404                                 break;
4405                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4406                             key.offset == num_bytes) {
4407                                 found_extent = 1;
4408                                 break;
4409                         }
4410                         if (path->slots[0] - extent_slot > 5)
4411                                 break;
4412                         extent_slot--;
4413                 }
4414 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4415                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4416                 if (found_extent && item_size < sizeof(*ei))
4417                         found_extent = 0;
4418 #endif
4419                 if (!found_extent) {
4420                         BUG_ON(iref);
4421                         ret = remove_extent_backref(trans, extent_root, path,
4422                                                     NULL, refs_to_drop,
4423                                                     is_data);
4424                         BUG_ON(ret);
4425                         btrfs_release_path(extent_root, path);
4426                         path->leave_spinning = 1;
4427
4428                         key.objectid = bytenr;
4429                         key.type = BTRFS_EXTENT_ITEM_KEY;
4430                         key.offset = num_bytes;
4431
4432                         ret = btrfs_search_slot(trans, extent_root,
4433                                                 &key, path, -1, 1);
4434                         if (ret) {
4435                                 printk(KERN_ERR "umm, got %d back from search"
4436                                        ", was looking for %llu\n", ret,
4437                                        (unsigned long long)bytenr);
4438                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4439                         }
4440                         BUG_ON(ret);
4441                         extent_slot = path->slots[0];
4442                 }
4443         } else {
4444                 btrfs_print_leaf(extent_root, path->nodes[0]);
4445                 WARN_ON(1);
4446                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4447                        "parent %llu root %llu  owner %llu offset %llu\n",
4448                        (unsigned long long)bytenr,
4449                        (unsigned long long)parent,
4450                        (unsigned long long)root_objectid,
4451                        (unsigned long long)owner_objectid,
4452                        (unsigned long long)owner_offset);
4453         }
4454
4455         leaf = path->nodes[0];
4456         item_size = btrfs_item_size_nr(leaf, extent_slot);
4457 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4458         if (item_size < sizeof(*ei)) {
4459                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4460                 ret = convert_extent_item_v0(trans, extent_root, path,
4461                                              owner_objectid, 0);
4462                 BUG_ON(ret < 0);
4463
4464                 btrfs_release_path(extent_root, path);
4465                 path->leave_spinning = 1;
4466
4467                 key.objectid = bytenr;
4468                 key.type = BTRFS_EXTENT_ITEM_KEY;
4469                 key.offset = num_bytes;
4470
4471                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4472                                         -1, 1);
4473                 if (ret) {
4474                         printk(KERN_ERR "umm, got %d back from search"
4475                                ", was looking for %llu\n", ret,
4476                                (unsigned long long)bytenr);
4477                         btrfs_print_leaf(extent_root, path->nodes[0]);
4478                 }
4479                 BUG_ON(ret);
4480                 extent_slot = path->slots[0];
4481                 leaf = path->nodes[0];
4482                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4483         }
4484 #endif
4485         BUG_ON(item_size < sizeof(*ei));
4486         ei = btrfs_item_ptr(leaf, extent_slot,
4487                             struct btrfs_extent_item);
4488         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4489                 struct btrfs_tree_block_info *bi;
4490                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4491                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4492                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4493         }
4494
4495         refs = btrfs_extent_refs(leaf, ei);
4496         BUG_ON(refs < refs_to_drop);
4497         refs -= refs_to_drop;
4498
4499         if (refs > 0) {
4500                 if (extent_op)
4501                         __run_delayed_extent_op(extent_op, leaf, ei);
4502                 /*
4503                  * In the case of inline back ref, reference count will
4504                  * be updated by remove_extent_backref
4505                  */
4506                 if (iref) {
4507                         BUG_ON(!found_extent);
4508                 } else {
4509                         btrfs_set_extent_refs(leaf, ei, refs);
4510                         btrfs_mark_buffer_dirty(leaf);
4511                 }
4512                 if (found_extent) {
4513                         ret = remove_extent_backref(trans, extent_root, path,
4514                                                     iref, refs_to_drop,
4515                                                     is_data);
4516                         BUG_ON(ret);
4517                 }
4518         } else {
4519                 if (found_extent) {
4520                         BUG_ON(is_data && refs_to_drop !=
4521                                extent_data_ref_count(root, path, iref));
4522                         if (iref) {
4523                                 BUG_ON(path->slots[0] != extent_slot);
4524                         } else {
4525                                 BUG_ON(path->slots[0] != extent_slot + 1);
4526                                 path->slots[0] = extent_slot;
4527                                 num_to_del = 2;
4528                         }
4529                 }
4530
4531                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4532                                       num_to_del);
4533                 BUG_ON(ret);
4534                 btrfs_release_path(extent_root, path);
4535
4536                 if (is_data) {
4537                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4538                         BUG_ON(ret);
4539                 } else {
4540                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4541                              bytenr >> PAGE_CACHE_SHIFT,
4542                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4543                 }
4544
4545                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4546                 BUG_ON(ret);
4547         }
4548         btrfs_free_path(path);
4549         return ret;
4550 }
4551
4552 /*
4553  * when we free an block, it is possible (and likely) that we free the last
4554  * delayed ref for that extent as well.  This searches the delayed ref tree for
4555  * a given extent, and if there are no other delayed refs to be processed, it
4556  * removes it from the tree.
4557  */
4558 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4559                                       struct btrfs_root *root, u64 bytenr)
4560 {
4561         struct btrfs_delayed_ref_head *head;
4562         struct btrfs_delayed_ref_root *delayed_refs;
4563         struct btrfs_delayed_ref_node *ref;
4564         struct rb_node *node;
4565         int ret = 0;
4566
4567         delayed_refs = &trans->transaction->delayed_refs;
4568         spin_lock(&delayed_refs->lock);
4569         head = btrfs_find_delayed_ref_head(trans, bytenr);
4570         if (!head)
4571                 goto out;
4572
4573         node = rb_prev(&head->node.rb_node);
4574         if (!node)
4575                 goto out;
4576
4577         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4578
4579         /* there are still entries for this ref, we can't drop it */
4580         if (ref->bytenr == bytenr)
4581                 goto out;
4582
4583         if (head->extent_op) {
4584                 if (!head->must_insert_reserved)
4585                         goto out;
4586                 kfree(head->extent_op);
4587                 head->extent_op = NULL;
4588         }
4589
4590         /*
4591          * waiting for the lock here would deadlock.  If someone else has it
4592          * locked they are already in the process of dropping it anyway
4593          */
4594         if (!mutex_trylock(&head->mutex))
4595                 goto out;
4596
4597         /*
4598          * at this point we have a head with no other entries.  Go
4599          * ahead and process it.
4600          */
4601         head->node.in_tree = 0;
4602         rb_erase(&head->node.rb_node, &delayed_refs->root);
4603
4604         delayed_refs->num_entries--;
4605
4606         /*
4607          * we don't take a ref on the node because we're removing it from the
4608          * tree, so we just steal the ref the tree was holding.
4609          */
4610         delayed_refs->num_heads--;
4611         if (list_empty(&head->cluster))
4612                 delayed_refs->num_heads_ready--;
4613
4614         list_del_init(&head->cluster);
4615         spin_unlock(&delayed_refs->lock);
4616
4617         BUG_ON(head->extent_op);
4618         if (head->must_insert_reserved)
4619                 ret = 1;
4620
4621         mutex_unlock(&head->mutex);
4622         btrfs_put_delayed_ref(&head->node);
4623         return ret;
4624 out:
4625         spin_unlock(&delayed_refs->lock);
4626         return 0;
4627 }
4628
4629 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4630                            struct btrfs_root *root,
4631                            struct extent_buffer *buf,
4632                            u64 parent, int last_ref)
4633 {
4634         struct btrfs_block_rsv *block_rsv;
4635         struct btrfs_block_group_cache *cache = NULL;
4636         int ret;
4637
4638         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4639                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4640                                                 parent, root->root_key.objectid,
4641                                                 btrfs_header_level(buf),
4642                                                 BTRFS_DROP_DELAYED_REF, NULL);
4643                 BUG_ON(ret);
4644         }
4645
4646         if (!last_ref)
4647                 return;
4648
4649         block_rsv = get_block_rsv(trans, root);
4650         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4651         if (block_rsv->space_info != cache->space_info)
4652                 goto out;
4653
4654         if (btrfs_header_generation(buf) == trans->transid) {
4655                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4656                         ret = check_ref_cleanup(trans, root, buf->start);
4657                         if (!ret)
4658                                 goto pin;
4659                 }
4660
4661                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4662                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4663                         goto pin;
4664                 }
4665
4666                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4667
4668                 btrfs_add_free_space(cache, buf->start, buf->len);
4669                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4670                 if (ret == -EAGAIN) {
4671                         /* block group became read-only */
4672                         update_reserved_bytes(cache, buf->len, 0, 1);
4673                         goto out;
4674                 }
4675
4676                 ret = 1;
4677                 spin_lock(&block_rsv->lock);
4678                 if (block_rsv->reserved < block_rsv->size) {
4679                         block_rsv->reserved += buf->len;
4680                         ret = 0;
4681                 }
4682                 spin_unlock(&block_rsv->lock);
4683
4684                 if (ret) {
4685                         spin_lock(&cache->space_info->lock);
4686                         cache->space_info->bytes_reserved -= buf->len;
4687                         spin_unlock(&cache->space_info->lock);
4688                 }
4689                 goto out;
4690         }
4691 pin:
4692         if (block_rsv->durable && !cache->ro) {
4693                 ret = 0;
4694                 spin_lock(&cache->lock);
4695                 if (!cache->ro) {
4696                         cache->reserved_pinned += buf->len;
4697                         ret = 1;
4698                 }
4699                 spin_unlock(&cache->lock);
4700
4701                 if (ret) {
4702                         spin_lock(&block_rsv->lock);
4703                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4704                         spin_unlock(&block_rsv->lock);
4705                 }
4706         }
4707 out:
4708         btrfs_put_block_group(cache);
4709 }
4710
4711 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4712                       struct btrfs_root *root,
4713                       u64 bytenr, u64 num_bytes, u64 parent,
4714                       u64 root_objectid, u64 owner, u64 offset)
4715 {
4716         int ret;
4717
4718         /*
4719          * tree log blocks never actually go into the extent allocation
4720          * tree, just update pinning info and exit early.
4721          */
4722         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4723                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4724                 /* unlocks the pinned mutex */
4725                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4726                 ret = 0;
4727         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4728                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4729                                         parent, root_objectid, (int)owner,
4730                                         BTRFS_DROP_DELAYED_REF, NULL);
4731                 BUG_ON(ret);
4732         } else {
4733                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4734                                         parent, root_objectid, owner,
4735                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4736                 BUG_ON(ret);
4737         }
4738         return ret;
4739 }
4740
4741 static u64 stripe_align(struct btrfs_root *root, u64 val)
4742 {
4743         u64 mask = ((u64)root->stripesize - 1);
4744         u64 ret = (val + mask) & ~mask;
4745         return ret;
4746 }
4747
4748 /*
4749  * when we wait for progress in the block group caching, its because
4750  * our allocation attempt failed at least once.  So, we must sleep
4751  * and let some progress happen before we try again.
4752  *
4753  * This function will sleep at least once waiting for new free space to
4754  * show up, and then it will check the block group free space numbers
4755  * for our min num_bytes.  Another option is to have it go ahead
4756  * and look in the rbtree for a free extent of a given size, but this
4757  * is a good start.
4758  */
4759 static noinline int
4760 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4761                                 u64 num_bytes)
4762 {
4763         struct btrfs_caching_control *caching_ctl;
4764         DEFINE_WAIT(wait);
4765
4766         caching_ctl = get_caching_control(cache);
4767         if (!caching_ctl)
4768                 return 0;
4769
4770         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4771                    (cache->free_space >= num_bytes));
4772
4773         put_caching_control(caching_ctl);
4774         return 0;
4775 }
4776
4777 static noinline int
4778 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4779 {
4780         struct btrfs_caching_control *caching_ctl;
4781         DEFINE_WAIT(wait);
4782
4783         caching_ctl = get_caching_control(cache);
4784         if (!caching_ctl)
4785                 return 0;
4786
4787         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4788
4789         put_caching_control(caching_ctl);
4790         return 0;
4791 }
4792
4793 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4794 {
4795         int index;
4796         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4797                 index = 0;
4798         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4799                 index = 1;
4800         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4801                 index = 2;
4802         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4803                 index = 3;
4804         else
4805                 index = 4;
4806         return index;
4807 }
4808
4809 enum btrfs_loop_type {
4810         LOOP_FIND_IDEAL = 0,
4811         LOOP_CACHING_NOWAIT = 1,
4812         LOOP_CACHING_WAIT = 2,
4813         LOOP_ALLOC_CHUNK = 3,
4814         LOOP_NO_EMPTY_SIZE = 4,
4815 };
4816
4817 /*
4818  * walks the btree of allocated extents and find a hole of a given size.
4819  * The key ins is changed to record the hole:
4820  * ins->objectid == block start
4821  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4822  * ins->offset == number of blocks
4823  * Any available blocks before search_start are skipped.
4824  */
4825 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4826                                      struct btrfs_root *orig_root,
4827                                      u64 num_bytes, u64 empty_size,
4828                                      u64 search_start, u64 search_end,
4829                                      u64 hint_byte, struct btrfs_key *ins,
4830                                      int data)
4831 {
4832         int ret = 0;
4833         struct btrfs_root *root = orig_root->fs_info->extent_root;
4834         struct btrfs_free_cluster *last_ptr = NULL;
4835         struct btrfs_block_group_cache *block_group = NULL;
4836         int empty_cluster = 2 * 1024 * 1024;
4837         int allowed_chunk_alloc = 0;
4838         int done_chunk_alloc = 0;
4839         struct btrfs_space_info *space_info;
4840         int last_ptr_loop = 0;
4841         int loop = 0;
4842         int index = 0;
4843         bool found_uncached_bg = false;
4844         bool failed_cluster_refill = false;
4845         bool failed_alloc = false;
4846         bool use_cluster = true;
4847         u64 ideal_cache_percent = 0;
4848         u64 ideal_cache_offset = 0;
4849
4850         WARN_ON(num_bytes < root->sectorsize);
4851         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4852         ins->objectid = 0;
4853         ins->offset = 0;
4854
4855         space_info = __find_space_info(root->fs_info, data);
4856         if (!space_info) {
4857                 printk(KERN_ERR "No space info for %d\n", data);
4858                 return -ENOSPC;
4859         }
4860
4861         /*
4862          * If the space info is for both data and metadata it means we have a
4863          * small filesystem and we can't use the clustering stuff.
4864          */
4865         if (btrfs_mixed_space_info(space_info))
4866                 use_cluster = false;
4867
4868         if (orig_root->ref_cows || empty_size)
4869                 allowed_chunk_alloc = 1;
4870
4871         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4872                 last_ptr = &root->fs_info->meta_alloc_cluster;
4873                 if (!btrfs_test_opt(root, SSD))
4874                         empty_cluster = 64 * 1024;
4875         }
4876
4877         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4878             btrfs_test_opt(root, SSD)) {
4879                 last_ptr = &root->fs_info->data_alloc_cluster;
4880         }
4881
4882         if (last_ptr) {
4883                 spin_lock(&last_ptr->lock);
4884                 if (last_ptr->block_group)
4885                         hint_byte = last_ptr->window_start;
4886                 spin_unlock(&last_ptr->lock);
4887         }
4888
4889         search_start = max(search_start, first_logical_byte(root, 0));
4890         search_start = max(search_start, hint_byte);
4891
4892         if (!last_ptr)
4893                 empty_cluster = 0;
4894
4895         if (search_start == hint_byte) {
4896 ideal_cache:
4897                 block_group = btrfs_lookup_block_group(root->fs_info,
4898                                                        search_start);
4899                 /*
4900                  * we don't want to use the block group if it doesn't match our
4901                  * allocation bits, or if its not cached.
4902                  *
4903                  * However if we are re-searching with an ideal block group
4904                  * picked out then we don't care that the block group is cached.
4905                  */
4906                 if (block_group && block_group_bits(block_group, data) &&
4907                     (block_group->cached != BTRFS_CACHE_NO ||
4908                      search_start == ideal_cache_offset)) {
4909                         down_read(&space_info->groups_sem);
4910                         if (list_empty(&block_group->list) ||
4911                             block_group->ro) {
4912                                 /*
4913                                  * someone is removing this block group,
4914                                  * we can't jump into the have_block_group
4915                                  * target because our list pointers are not
4916                                  * valid
4917                                  */
4918                                 btrfs_put_block_group(block_group);
4919                                 up_read(&space_info->groups_sem);
4920                         } else {
4921                                 index = get_block_group_index(block_group);
4922                                 goto have_block_group;
4923                         }
4924                 } else if (block_group) {
4925                         btrfs_put_block_group(block_group);
4926                 }
4927         }
4928 search:
4929         down_read(&space_info->groups_sem);
4930         list_for_each_entry(block_group, &space_info->block_groups[index],
4931                             list) {
4932                 u64 offset;
4933                 int cached;
4934
4935                 btrfs_get_block_group(block_group);
4936                 search_start = block_group->key.objectid;
4937
4938 have_block_group:
4939                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4940                         u64 free_percent;
4941
4942                         ret = cache_block_group(block_group, trans, 1);
4943                         if (block_group->cached == BTRFS_CACHE_FINISHED)
4944                                 goto have_block_group;
4945
4946                         free_percent = btrfs_block_group_used(&block_group->item);
4947                         free_percent *= 100;
4948                         free_percent = div64_u64(free_percent,
4949                                                  block_group->key.offset);
4950                         free_percent = 100 - free_percent;
4951                         if (free_percent > ideal_cache_percent &&
4952                             likely(!block_group->ro)) {
4953                                 ideal_cache_offset = block_group->key.objectid;
4954                                 ideal_cache_percent = free_percent;
4955                         }
4956
4957                         /*
4958                          * We only want to start kthread caching if we are at
4959                          * the point where we will wait for caching to make
4960                          * progress, or if our ideal search is over and we've
4961                          * found somebody to start caching.
4962                          */
4963                         if (loop > LOOP_CACHING_NOWAIT ||
4964                             (loop > LOOP_FIND_IDEAL &&
4965                              atomic_read(&space_info->caching_threads) < 2)) {
4966                                 ret = cache_block_group(block_group, trans, 0);
4967                                 BUG_ON(ret);
4968                         }
4969                         found_uncached_bg = true;
4970
4971                         /*
4972                          * If loop is set for cached only, try the next block
4973                          * group.
4974                          */
4975                         if (loop == LOOP_FIND_IDEAL)
4976                                 goto loop;
4977                 }
4978
4979                 cached = block_group_cache_done(block_group);
4980                 if (unlikely(!cached))
4981                         found_uncached_bg = true;
4982
4983                 if (unlikely(block_group->ro))
4984                         goto loop;
4985
4986                 /*
4987                  * Ok we want to try and use the cluster allocator, so lets look
4988                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4989                  * have tried the cluster allocator plenty of times at this
4990                  * point and not have found anything, so we are likely way too
4991                  * fragmented for the clustering stuff to find anything, so lets
4992                  * just skip it and let the allocator find whatever block it can
4993                  * find
4994                  */
4995                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4996                         /*
4997                          * the refill lock keeps out other
4998                          * people trying to start a new cluster
4999                          */
5000                         spin_lock(&last_ptr->refill_lock);
5001                         if (last_ptr->block_group &&
5002                             (last_ptr->block_group->ro ||
5003                             !block_group_bits(last_ptr->block_group, data))) {
5004                                 offset = 0;
5005                                 goto refill_cluster;
5006                         }
5007
5008                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5009                                                  num_bytes, search_start);
5010                         if (offset) {
5011                                 /* we have a block, we're done */
5012                                 spin_unlock(&last_ptr->refill_lock);
5013                                 goto checks;
5014                         }
5015
5016                         spin_lock(&last_ptr->lock);
5017                         /*
5018                          * whoops, this cluster doesn't actually point to
5019                          * this block group.  Get a ref on the block
5020                          * group is does point to and try again
5021                          */
5022                         if (!last_ptr_loop && last_ptr->block_group &&
5023                             last_ptr->block_group != block_group) {
5024
5025                                 btrfs_put_block_group(block_group);
5026                                 block_group = last_ptr->block_group;
5027                                 btrfs_get_block_group(block_group);
5028                                 spin_unlock(&last_ptr->lock);
5029                                 spin_unlock(&last_ptr->refill_lock);
5030
5031                                 last_ptr_loop = 1;
5032                                 search_start = block_group->key.objectid;
5033                                 /*
5034                                  * we know this block group is properly
5035                                  * in the list because
5036                                  * btrfs_remove_block_group, drops the
5037                                  * cluster before it removes the block
5038                                  * group from the list
5039                                  */
5040                                 goto have_block_group;
5041                         }
5042                         spin_unlock(&last_ptr->lock);
5043 refill_cluster:
5044                         /*
5045                          * this cluster didn't work out, free it and
5046                          * start over
5047                          */
5048                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5049
5050                         last_ptr_loop = 0;
5051
5052                         /* allocate a cluster in this block group */
5053                         ret = btrfs_find_space_cluster(trans, root,
5054                                                block_group, last_ptr,
5055                                                offset, num_bytes,
5056                                                empty_cluster + empty_size);
5057                         if (ret == 0) {
5058                                 /*
5059                                  * now pull our allocation out of this
5060                                  * cluster
5061                                  */
5062                                 offset = btrfs_alloc_from_cluster(block_group,
5063                                                   last_ptr, num_bytes,
5064                                                   search_start);
5065                                 if (offset) {
5066                                         /* we found one, proceed */
5067                                         spin_unlock(&last_ptr->refill_lock);
5068                                         goto checks;
5069                                 }
5070                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5071                                    && !failed_cluster_refill) {
5072                                 spin_unlock(&last_ptr->refill_lock);
5073
5074                                 failed_cluster_refill = true;
5075                                 wait_block_group_cache_progress(block_group,
5076                                        num_bytes + empty_cluster + empty_size);
5077                                 goto have_block_group;
5078                         }
5079
5080                         /*
5081                          * at this point we either didn't find a cluster
5082                          * or we weren't able to allocate a block from our
5083                          * cluster.  Free the cluster we've been trying
5084                          * to use, and go to the next block group
5085                          */
5086                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5087                         spin_unlock(&last_ptr->refill_lock);
5088                         goto loop;
5089                 }
5090
5091                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5092                                                     num_bytes, empty_size);
5093                 /*
5094                  * If we didn't find a chunk, and we haven't failed on this
5095                  * block group before, and this block group is in the middle of
5096                  * caching and we are ok with waiting, then go ahead and wait
5097                  * for progress to be made, and set failed_alloc to true.
5098                  *
5099                  * If failed_alloc is true then we've already waited on this
5100                  * block group once and should move on to the next block group.
5101                  */
5102                 if (!offset && !failed_alloc && !cached &&
5103                     loop > LOOP_CACHING_NOWAIT) {
5104                         wait_block_group_cache_progress(block_group,
5105                                                 num_bytes + empty_size);
5106                         failed_alloc = true;
5107                         goto have_block_group;
5108                 } else if (!offset) {
5109                         goto loop;
5110                 }
5111 checks:
5112                 search_start = stripe_align(root, offset);
5113                 /* move on to the next group */
5114                 if (search_start + num_bytes >= search_end) {
5115                         btrfs_add_free_space(block_group, offset, num_bytes);
5116                         goto loop;
5117                 }
5118
5119                 /* move on to the next group */
5120                 if (search_start + num_bytes >
5121                     block_group->key.objectid + block_group->key.offset) {
5122                         btrfs_add_free_space(block_group, offset, num_bytes);
5123                         goto loop;
5124                 }
5125
5126                 ins->objectid = search_start;
5127                 ins->offset = num_bytes;
5128
5129                 if (offset < search_start)
5130                         btrfs_add_free_space(block_group, offset,
5131                                              search_start - offset);
5132                 BUG_ON(offset > search_start);
5133
5134                 ret = update_reserved_bytes(block_group, num_bytes, 1,
5135                                             (data & BTRFS_BLOCK_GROUP_DATA));
5136                 if (ret == -EAGAIN) {
5137                         btrfs_add_free_space(block_group, offset, num_bytes);
5138                         goto loop;
5139                 }
5140
5141                 /* we are all good, lets return */
5142                 ins->objectid = search_start;
5143                 ins->offset = num_bytes;
5144
5145                 if (offset < search_start)
5146                         btrfs_add_free_space(block_group, offset,
5147                                              search_start - offset);
5148                 BUG_ON(offset > search_start);
5149                 break;
5150 loop:
5151                 failed_cluster_refill = false;
5152                 failed_alloc = false;
5153                 BUG_ON(index != get_block_group_index(block_group));
5154                 btrfs_put_block_group(block_group);
5155         }
5156         up_read(&space_info->groups_sem);
5157
5158         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5159                 goto search;
5160
5161         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5162          *                      for them to make caching progress.  Also
5163          *                      determine the best possible bg to cache
5164          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5165          *                      caching kthreads as we move along
5166          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5167          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5168          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5169          *                      again
5170          */
5171         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5172             (found_uncached_bg || empty_size || empty_cluster ||
5173              allowed_chunk_alloc)) {
5174                 index = 0;
5175                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5176                         found_uncached_bg = false;
5177                         loop++;
5178                         if (!ideal_cache_percent &&
5179                             atomic_read(&space_info->caching_threads))
5180                                 goto search;
5181
5182                         /*
5183                          * 1 of the following 2 things have happened so far
5184                          *
5185                          * 1) We found an ideal block group for caching that
5186                          * is mostly full and will cache quickly, so we might
5187                          * as well wait for it.
5188                          *
5189                          * 2) We searched for cached only and we didn't find
5190                          * anything, and we didn't start any caching kthreads
5191                          * either, so chances are we will loop through and
5192                          * start a couple caching kthreads, and then come back
5193                          * around and just wait for them.  This will be slower
5194                          * because we will have 2 caching kthreads reading at
5195                          * the same time when we could have just started one
5196                          * and waited for it to get far enough to give us an
5197                          * allocation, so go ahead and go to the wait caching
5198                          * loop.
5199                          */
5200                         loop = LOOP_CACHING_WAIT;
5201                         search_start = ideal_cache_offset;
5202                         ideal_cache_percent = 0;
5203                         goto ideal_cache;
5204                 } else if (loop == LOOP_FIND_IDEAL) {
5205                         /*
5206                          * Didn't find a uncached bg, wait on anything we find
5207                          * next.
5208                          */
5209                         loop = LOOP_CACHING_WAIT;
5210                         goto search;
5211                 }
5212
5213                 if (loop < LOOP_CACHING_WAIT) {
5214                         loop++;
5215                         goto search;
5216                 }
5217
5218                 if (loop == LOOP_ALLOC_CHUNK) {
5219                         empty_size = 0;
5220                         empty_cluster = 0;
5221                 }
5222
5223                 if (allowed_chunk_alloc) {
5224                         ret = do_chunk_alloc(trans, root, num_bytes +
5225                                              2 * 1024 * 1024, data, 1);
5226                         allowed_chunk_alloc = 0;
5227                         done_chunk_alloc = 1;
5228                 } else if (!done_chunk_alloc) {
5229                         space_info->force_alloc = 1;
5230                 }
5231
5232                 if (loop < LOOP_NO_EMPTY_SIZE) {
5233                         loop++;
5234                         goto search;
5235                 }
5236                 ret = -ENOSPC;
5237         } else if (!ins->objectid) {
5238                 ret = -ENOSPC;
5239         }
5240
5241         /* we found what we needed */
5242         if (ins->objectid) {
5243                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5244                         trans->block_group = block_group->key.objectid;
5245
5246                 btrfs_put_block_group(block_group);
5247                 ret = 0;
5248         }
5249
5250         return ret;
5251 }
5252
5253 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5254                             int dump_block_groups)
5255 {
5256         struct btrfs_block_group_cache *cache;
5257         int index = 0;
5258
5259         spin_lock(&info->lock);
5260         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5261                (unsigned long long)(info->total_bytes - info->bytes_used -
5262                                     info->bytes_pinned - info->bytes_reserved -
5263                                     info->bytes_readonly),
5264                (info->full) ? "" : "not ");
5265         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5266                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5267                (unsigned long long)info->total_bytes,
5268                (unsigned long long)info->bytes_used,
5269                (unsigned long long)info->bytes_pinned,
5270                (unsigned long long)info->bytes_reserved,
5271                (unsigned long long)info->bytes_may_use,
5272                (unsigned long long)info->bytes_readonly);
5273         spin_unlock(&info->lock);
5274
5275         if (!dump_block_groups)
5276                 return;
5277
5278         down_read(&info->groups_sem);
5279 again:
5280         list_for_each_entry(cache, &info->block_groups[index], list) {
5281                 spin_lock(&cache->lock);
5282                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5283                        "%llu pinned %llu reserved\n",
5284                        (unsigned long long)cache->key.objectid,
5285                        (unsigned long long)cache->key.offset,
5286                        (unsigned long long)btrfs_block_group_used(&cache->item),
5287                        (unsigned long long)cache->pinned,
5288                        (unsigned long long)cache->reserved);
5289                 btrfs_dump_free_space(cache, bytes);
5290                 spin_unlock(&cache->lock);
5291         }
5292         if (++index < BTRFS_NR_RAID_TYPES)
5293                 goto again;
5294         up_read(&info->groups_sem);
5295 }
5296
5297 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5298                          struct btrfs_root *root,
5299                          u64 num_bytes, u64 min_alloc_size,
5300                          u64 empty_size, u64 hint_byte,
5301                          u64 search_end, struct btrfs_key *ins,
5302                          u64 data)
5303 {
5304         int ret;
5305         u64 search_start = 0;
5306
5307         data = btrfs_get_alloc_profile(root, data);
5308 again:
5309         /*
5310          * the only place that sets empty_size is btrfs_realloc_node, which
5311          * is not called recursively on allocations
5312          */
5313         if (empty_size || root->ref_cows)
5314                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5315                                      num_bytes + 2 * 1024 * 1024, data, 0);
5316
5317         WARN_ON(num_bytes < root->sectorsize);
5318         ret = find_free_extent(trans, root, num_bytes, empty_size,
5319                                search_start, search_end, hint_byte,
5320                                ins, data);
5321
5322         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5323                 num_bytes = num_bytes >> 1;
5324                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5325                 num_bytes = max(num_bytes, min_alloc_size);
5326                 do_chunk_alloc(trans, root->fs_info->extent_root,
5327                                num_bytes, data, 1);
5328                 goto again;
5329         }
5330         if (ret == -ENOSPC) {
5331                 struct btrfs_space_info *sinfo;
5332
5333                 sinfo = __find_space_info(root->fs_info, data);
5334                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5335                        "wanted %llu\n", (unsigned long long)data,
5336                        (unsigned long long)num_bytes);
5337                 dump_space_info(sinfo, num_bytes, 1);
5338         }
5339
5340         return ret;
5341 }
5342
5343 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5344 {
5345         struct btrfs_block_group_cache *cache;
5346         int ret = 0;
5347
5348         cache = btrfs_lookup_block_group(root->fs_info, start);
5349         if (!cache) {
5350                 printk(KERN_ERR "Unable to find block group for %llu\n",
5351                        (unsigned long long)start);
5352                 return -ENOSPC;
5353         }
5354
5355         ret = btrfs_discard_extent(root, start, len);
5356
5357         btrfs_add_free_space(cache, start, len);
5358         update_reserved_bytes(cache, len, 0, 1);
5359         btrfs_put_block_group(cache);
5360
5361         return ret;
5362 }
5363
5364 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5365                                       struct btrfs_root *root,
5366                                       u64 parent, u64 root_objectid,
5367                                       u64 flags, u64 owner, u64 offset,
5368                                       struct btrfs_key *ins, int ref_mod)
5369 {
5370         int ret;
5371         struct btrfs_fs_info *fs_info = root->fs_info;
5372         struct btrfs_extent_item *extent_item;
5373         struct btrfs_extent_inline_ref *iref;
5374         struct btrfs_path *path;
5375         struct extent_buffer *leaf;
5376         int type;
5377         u32 size;
5378
5379         if (parent > 0)
5380                 type = BTRFS_SHARED_DATA_REF_KEY;
5381         else
5382                 type = BTRFS_EXTENT_DATA_REF_KEY;
5383
5384         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5385
5386         path = btrfs_alloc_path();
5387         BUG_ON(!path);
5388
5389         path->leave_spinning = 1;
5390         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5391                                       ins, size);
5392         BUG_ON(ret);
5393
5394         leaf = path->nodes[0];
5395         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5396                                      struct btrfs_extent_item);
5397         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5398         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5399         btrfs_set_extent_flags(leaf, extent_item,
5400                                flags | BTRFS_EXTENT_FLAG_DATA);
5401
5402         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5403         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5404         if (parent > 0) {
5405                 struct btrfs_shared_data_ref *ref;
5406                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5407                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5408                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5409         } else {
5410                 struct btrfs_extent_data_ref *ref;
5411                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5412                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5413                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5414                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5415                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5416         }
5417
5418         btrfs_mark_buffer_dirty(path->nodes[0]);
5419         btrfs_free_path(path);
5420
5421         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5422         if (ret) {
5423                 printk(KERN_ERR "btrfs update block group failed for %llu "
5424                        "%llu\n", (unsigned long long)ins->objectid,
5425                        (unsigned long long)ins->offset);
5426                 BUG();
5427         }
5428         return ret;
5429 }
5430
5431 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5432                                      struct btrfs_root *root,
5433                                      u64 parent, u64 root_objectid,
5434                                      u64 flags, struct btrfs_disk_key *key,
5435                                      int level, struct btrfs_key *ins)
5436 {
5437         int ret;
5438         struct btrfs_fs_info *fs_info = root->fs_info;
5439         struct btrfs_extent_item *extent_item;
5440         struct btrfs_tree_block_info *block_info;
5441         struct btrfs_extent_inline_ref *iref;
5442         struct btrfs_path *path;
5443         struct extent_buffer *leaf;
5444         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5445
5446         path = btrfs_alloc_path();
5447         BUG_ON(!path);
5448
5449         path->leave_spinning = 1;
5450         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5451                                       ins, size);
5452         BUG_ON(ret);
5453
5454         leaf = path->nodes[0];
5455         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5456                                      struct btrfs_extent_item);
5457         btrfs_set_extent_refs(leaf, extent_item, 1);
5458         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5459         btrfs_set_extent_flags(leaf, extent_item,
5460                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5461         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5462
5463         btrfs_set_tree_block_key(leaf, block_info, key);
5464         btrfs_set_tree_block_level(leaf, block_info, level);
5465
5466         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5467         if (parent > 0) {
5468                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5469                 btrfs_set_extent_inline_ref_type(leaf, iref,
5470                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5471                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5472         } else {
5473                 btrfs_set_extent_inline_ref_type(leaf, iref,
5474                                                  BTRFS_TREE_BLOCK_REF_KEY);
5475                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5476         }
5477
5478         btrfs_mark_buffer_dirty(leaf);
5479         btrfs_free_path(path);
5480
5481         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5482         if (ret) {
5483                 printk(KERN_ERR "btrfs update block group failed for %llu "
5484                        "%llu\n", (unsigned long long)ins->objectid,
5485                        (unsigned long long)ins->offset);
5486                 BUG();
5487         }
5488         return ret;
5489 }
5490
5491 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5492                                      struct btrfs_root *root,
5493                                      u64 root_objectid, u64 owner,
5494                                      u64 offset, struct btrfs_key *ins)
5495 {
5496         int ret;
5497
5498         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5499
5500         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5501                                          0, root_objectid, owner, offset,
5502                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5503         return ret;
5504 }
5505
5506 /*
5507  * this is used by the tree logging recovery code.  It records that
5508  * an extent has been allocated and makes sure to clear the free
5509  * space cache bits as well
5510  */
5511 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5512                                    struct btrfs_root *root,
5513                                    u64 root_objectid, u64 owner, u64 offset,
5514                                    struct btrfs_key *ins)
5515 {
5516         int ret;
5517         struct btrfs_block_group_cache *block_group;
5518         struct btrfs_caching_control *caching_ctl;
5519         u64 start = ins->objectid;
5520         u64 num_bytes = ins->offset;
5521
5522         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5523         cache_block_group(block_group, trans, 0);
5524         caching_ctl = get_caching_control(block_group);
5525
5526         if (!caching_ctl) {
5527                 BUG_ON(!block_group_cache_done(block_group));
5528                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5529                 BUG_ON(ret);
5530         } else {
5531                 mutex_lock(&caching_ctl->mutex);
5532
5533                 if (start >= caching_ctl->progress) {
5534                         ret = add_excluded_extent(root, start, num_bytes);
5535                         BUG_ON(ret);
5536                 } else if (start + num_bytes <= caching_ctl->progress) {
5537                         ret = btrfs_remove_free_space(block_group,
5538                                                       start, num_bytes);
5539                         BUG_ON(ret);
5540                 } else {
5541                         num_bytes = caching_ctl->progress - start;
5542                         ret = btrfs_remove_free_space(block_group,
5543                                                       start, num_bytes);
5544                         BUG_ON(ret);
5545
5546                         start = caching_ctl->progress;
5547                         num_bytes = ins->objectid + ins->offset -
5548                                     caching_ctl->progress;
5549                         ret = add_excluded_extent(root, start, num_bytes);
5550                         BUG_ON(ret);
5551                 }
5552
5553                 mutex_unlock(&caching_ctl->mutex);
5554                 put_caching_control(caching_ctl);
5555         }
5556
5557         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5558         BUG_ON(ret);
5559         btrfs_put_block_group(block_group);
5560         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5561                                          0, owner, offset, ins, 1);
5562         return ret;
5563 }
5564
5565 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5566                                             struct btrfs_root *root,
5567                                             u64 bytenr, u32 blocksize,
5568                                             int level)
5569 {
5570         struct extent_buffer *buf;
5571
5572         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5573         if (!buf)
5574                 return ERR_PTR(-ENOMEM);
5575         btrfs_set_header_generation(buf, trans->transid);
5576         btrfs_set_buffer_lockdep_class(buf, level);
5577         btrfs_tree_lock(buf);
5578         clean_tree_block(trans, root, buf);
5579
5580         btrfs_set_lock_blocking(buf);
5581         btrfs_set_buffer_uptodate(buf);
5582
5583         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5584                 /*
5585                  * we allow two log transactions at a time, use different
5586                  * EXENT bit to differentiate dirty pages.
5587                  */
5588                 if (root->log_transid % 2 == 0)
5589                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5590                                         buf->start + buf->len - 1, GFP_NOFS);
5591                 else
5592                         set_extent_new(&root->dirty_log_pages, buf->start,
5593                                         buf->start + buf->len - 1, GFP_NOFS);
5594         } else {
5595                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5596                          buf->start + buf->len - 1, GFP_NOFS);
5597         }
5598         trans->blocks_used++;
5599         /* this returns a buffer locked for blocking */
5600         return buf;
5601 }
5602
5603 static struct btrfs_block_rsv *
5604 use_block_rsv(struct btrfs_trans_handle *trans,
5605               struct btrfs_root *root, u32 blocksize)
5606 {
5607         struct btrfs_block_rsv *block_rsv;
5608         int ret;
5609
5610         block_rsv = get_block_rsv(trans, root);
5611
5612         if (block_rsv->size == 0) {
5613                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5614                                              blocksize, 0);
5615                 if (ret)
5616                         return ERR_PTR(ret);
5617                 return block_rsv;
5618         }
5619
5620         ret = block_rsv_use_bytes(block_rsv, blocksize);
5621         if (!ret)
5622                 return block_rsv;
5623
5624         return ERR_PTR(-ENOSPC);
5625 }
5626
5627 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5628 {
5629         block_rsv_add_bytes(block_rsv, blocksize, 0);
5630         block_rsv_release_bytes(block_rsv, NULL, 0);
5631 }
5632
5633 /*
5634  * finds a free extent and does all the dirty work required for allocation
5635  * returns the key for the extent through ins, and a tree buffer for
5636  * the first block of the extent through buf.
5637  *
5638  * returns the tree buffer or NULL.
5639  */
5640 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5641                                         struct btrfs_root *root, u32 blocksize,
5642                                         u64 parent, u64 root_objectid,
5643                                         struct btrfs_disk_key *key, int level,
5644                                         u64 hint, u64 empty_size)
5645 {
5646         struct btrfs_key ins;
5647         struct btrfs_block_rsv *block_rsv;
5648         struct extent_buffer *buf;
5649         u64 flags = 0;
5650         int ret;
5651
5652
5653         block_rsv = use_block_rsv(trans, root, blocksize);
5654         if (IS_ERR(block_rsv))
5655                 return ERR_CAST(block_rsv);
5656
5657         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5658                                    empty_size, hint, (u64)-1, &ins, 0);
5659         if (ret) {
5660                 unuse_block_rsv(block_rsv, blocksize);
5661                 return ERR_PTR(ret);
5662         }
5663
5664         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5665                                     blocksize, level);
5666         BUG_ON(IS_ERR(buf));
5667
5668         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5669                 if (parent == 0)
5670                         parent = ins.objectid;
5671                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5672         } else
5673                 BUG_ON(parent > 0);
5674
5675         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5676                 struct btrfs_delayed_extent_op *extent_op;
5677                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5678                 BUG_ON(!extent_op);
5679                 if (key)
5680                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5681                 else
5682                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5683                 extent_op->flags_to_set = flags;
5684                 extent_op->update_key = 1;
5685                 extent_op->update_flags = 1;
5686                 extent_op->is_data = 0;
5687
5688                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5689                                         ins.offset, parent, root_objectid,
5690                                         level, BTRFS_ADD_DELAYED_EXTENT,
5691                                         extent_op);
5692                 BUG_ON(ret);
5693         }
5694         return buf;
5695 }
5696
5697 struct walk_control {
5698         u64 refs[BTRFS_MAX_LEVEL];
5699         u64 flags[BTRFS_MAX_LEVEL];
5700         struct btrfs_key update_progress;
5701         int stage;
5702         int level;
5703         int shared_level;
5704         int update_ref;
5705         int keep_locks;
5706         int reada_slot;
5707         int reada_count;
5708 };
5709
5710 #define DROP_REFERENCE  1
5711 #define UPDATE_BACKREF  2
5712
5713 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5714                                      struct btrfs_root *root,
5715                                      struct walk_control *wc,
5716                                      struct btrfs_path *path)
5717 {
5718         u64 bytenr;
5719         u64 generation;
5720         u64 refs;
5721         u64 flags;
5722         u32 nritems;
5723         u32 blocksize;
5724         struct btrfs_key key;
5725         struct extent_buffer *eb;
5726         int ret;
5727         int slot;
5728         int nread = 0;
5729
5730         if (path->slots[wc->level] < wc->reada_slot) {
5731                 wc->reada_count = wc->reada_count * 2 / 3;
5732                 wc->reada_count = max(wc->reada_count, 2);
5733         } else {
5734                 wc->reada_count = wc->reada_count * 3 / 2;
5735                 wc->reada_count = min_t(int, wc->reada_count,
5736                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5737         }
5738
5739         eb = path->nodes[wc->level];
5740         nritems = btrfs_header_nritems(eb);
5741         blocksize = btrfs_level_size(root, wc->level - 1);
5742
5743         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5744                 if (nread >= wc->reada_count)
5745                         break;
5746
5747                 cond_resched();
5748                 bytenr = btrfs_node_blockptr(eb, slot);
5749                 generation = btrfs_node_ptr_generation(eb, slot);
5750
5751                 if (slot == path->slots[wc->level])
5752                         goto reada;
5753
5754                 if (wc->stage == UPDATE_BACKREF &&
5755                     generation <= root->root_key.offset)
5756                         continue;
5757
5758                 /* We don't lock the tree block, it's OK to be racy here */
5759                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5760                                                &refs, &flags);
5761                 BUG_ON(ret);
5762                 BUG_ON(refs == 0);
5763
5764                 if (wc->stage == DROP_REFERENCE) {
5765                         if (refs == 1)
5766                                 goto reada;
5767
5768                         if (wc->level == 1 &&
5769                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5770                                 continue;
5771                         if (!wc->update_ref ||
5772                             generation <= root->root_key.offset)
5773                                 continue;
5774                         btrfs_node_key_to_cpu(eb, &key, slot);
5775                         ret = btrfs_comp_cpu_keys(&key,
5776                                                   &wc->update_progress);
5777                         if (ret < 0)
5778                                 continue;
5779                 } else {
5780                         if (wc->level == 1 &&
5781                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5782                                 continue;
5783                 }
5784 reada:
5785                 ret = readahead_tree_block(root, bytenr, blocksize,
5786                                            generation);
5787                 if (ret)
5788                         break;
5789                 nread++;
5790         }
5791         wc->reada_slot = slot;
5792 }
5793
5794 /*
5795  * hepler to process tree block while walking down the tree.
5796  *
5797  * when wc->stage == UPDATE_BACKREF, this function updates
5798  * back refs for pointers in the block.
5799  *
5800  * NOTE: return value 1 means we should stop walking down.
5801  */
5802 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5803                                    struct btrfs_root *root,
5804                                    struct btrfs_path *path,
5805                                    struct walk_control *wc, int lookup_info)
5806 {
5807         int level = wc->level;
5808         struct extent_buffer *eb = path->nodes[level];
5809         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5810         int ret;
5811
5812         if (wc->stage == UPDATE_BACKREF &&
5813             btrfs_header_owner(eb) != root->root_key.objectid)
5814                 return 1;
5815
5816         /*
5817          * when reference count of tree block is 1, it won't increase
5818          * again. once full backref flag is set, we never clear it.
5819          */
5820         if (lookup_info &&
5821             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5822              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5823                 BUG_ON(!path->locks[level]);
5824                 ret = btrfs_lookup_extent_info(trans, root,
5825                                                eb->start, eb->len,
5826                                                &wc->refs[level],
5827                                                &wc->flags[level]);
5828                 BUG_ON(ret);
5829                 BUG_ON(wc->refs[level] == 0);
5830         }
5831
5832         if (wc->stage == DROP_REFERENCE) {
5833                 if (wc->refs[level] > 1)
5834                         return 1;
5835
5836                 if (path->locks[level] && !wc->keep_locks) {
5837                         btrfs_tree_unlock(eb);
5838                         path->locks[level] = 0;
5839                 }
5840                 return 0;
5841         }
5842
5843         /* wc->stage == UPDATE_BACKREF */
5844         if (!(wc->flags[level] & flag)) {
5845                 BUG_ON(!path->locks[level]);
5846                 ret = btrfs_inc_ref(trans, root, eb, 1);
5847                 BUG_ON(ret);
5848                 ret = btrfs_dec_ref(trans, root, eb, 0);
5849                 BUG_ON(ret);
5850                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5851                                                   eb->len, flag, 0);
5852                 BUG_ON(ret);
5853                 wc->flags[level] |= flag;
5854         }
5855
5856         /*
5857          * the block is shared by multiple trees, so it's not good to
5858          * keep the tree lock
5859          */
5860         if (path->locks[level] && level > 0) {
5861                 btrfs_tree_unlock(eb);
5862                 path->locks[level] = 0;
5863         }
5864         return 0;
5865 }
5866
5867 /*
5868  * hepler to process tree block pointer.
5869  *
5870  * when wc->stage == DROP_REFERENCE, this function checks
5871  * reference count of the block pointed to. if the block
5872  * is shared and we need update back refs for the subtree
5873  * rooted at the block, this function changes wc->stage to
5874  * UPDATE_BACKREF. if the block is shared and there is no
5875  * need to update back, this function drops the reference
5876  * to the block.
5877  *
5878  * NOTE: return value 1 means we should stop walking down.
5879  */
5880 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5881                                  struct btrfs_root *root,
5882                                  struct btrfs_path *path,
5883                                  struct walk_control *wc, int *lookup_info)
5884 {
5885         u64 bytenr;
5886         u64 generation;
5887         u64 parent;
5888         u32 blocksize;
5889         struct btrfs_key key;
5890         struct extent_buffer *next;
5891         int level = wc->level;
5892         int reada = 0;
5893         int ret = 0;
5894
5895         generation = btrfs_node_ptr_generation(path->nodes[level],
5896                                                path->slots[level]);
5897         /*
5898          * if the lower level block was created before the snapshot
5899          * was created, we know there is no need to update back refs
5900          * for the subtree
5901          */
5902         if (wc->stage == UPDATE_BACKREF &&
5903             generation <= root->root_key.offset) {
5904                 *lookup_info = 1;
5905                 return 1;
5906         }
5907
5908         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5909         blocksize = btrfs_level_size(root, level - 1);
5910
5911         next = btrfs_find_tree_block(root, bytenr, blocksize);
5912         if (!next) {
5913                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5914                 if (!next)
5915                         return -ENOMEM;
5916                 reada = 1;
5917         }
5918         btrfs_tree_lock(next);
5919         btrfs_set_lock_blocking(next);
5920
5921         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5922                                        &wc->refs[level - 1],
5923                                        &wc->flags[level - 1]);
5924         BUG_ON(ret);
5925         BUG_ON(wc->refs[level - 1] == 0);
5926         *lookup_info = 0;
5927
5928         if (wc->stage == DROP_REFERENCE) {
5929                 if (wc->refs[level - 1] > 1) {
5930                         if (level == 1 &&
5931                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5932                                 goto skip;
5933
5934                         if (!wc->update_ref ||
5935                             generation <= root->root_key.offset)
5936                                 goto skip;
5937
5938                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5939                                               path->slots[level]);
5940                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5941                         if (ret < 0)
5942                                 goto skip;
5943
5944                         wc->stage = UPDATE_BACKREF;
5945                         wc->shared_level = level - 1;
5946                 }
5947         } else {
5948                 if (level == 1 &&
5949                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5950                         goto skip;
5951         }
5952
5953         if (!btrfs_buffer_uptodate(next, generation)) {
5954                 btrfs_tree_unlock(next);
5955                 free_extent_buffer(next);
5956                 next = NULL;
5957                 *lookup_info = 1;
5958         }
5959
5960         if (!next) {
5961                 if (reada && level == 1)
5962                         reada_walk_down(trans, root, wc, path);
5963                 next = read_tree_block(root, bytenr, blocksize, generation);
5964                 btrfs_tree_lock(next);
5965                 btrfs_set_lock_blocking(next);
5966         }
5967
5968         level--;
5969         BUG_ON(level != btrfs_header_level(next));
5970         path->nodes[level] = next;
5971         path->slots[level] = 0;
5972         path->locks[level] = 1;
5973         wc->level = level;
5974         if (wc->level == 1)
5975                 wc->reada_slot = 0;
5976         return 0;
5977 skip:
5978         wc->refs[level - 1] = 0;
5979         wc->flags[level - 1] = 0;
5980         if (wc->stage == DROP_REFERENCE) {
5981                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5982                         parent = path->nodes[level]->start;
5983                 } else {
5984                         BUG_ON(root->root_key.objectid !=
5985                                btrfs_header_owner(path->nodes[level]));
5986                         parent = 0;
5987                 }
5988
5989                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5990                                         root->root_key.objectid, level - 1, 0);
5991                 BUG_ON(ret);
5992         }
5993         btrfs_tree_unlock(next);
5994         free_extent_buffer(next);
5995         *lookup_info = 1;
5996         return 1;
5997 }
5998
5999 /*
6000  * hepler to process tree block while walking up the tree.
6001  *
6002  * when wc->stage == DROP_REFERENCE, this function drops
6003  * reference count on the block.
6004  *
6005  * when wc->stage == UPDATE_BACKREF, this function changes
6006  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6007  * to UPDATE_BACKREF previously while processing the block.
6008  *
6009  * NOTE: return value 1 means we should stop walking up.
6010  */
6011 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6012                                  struct btrfs_root *root,
6013                                  struct btrfs_path *path,
6014                                  struct walk_control *wc)
6015 {
6016         int ret;
6017         int level = wc->level;
6018         struct extent_buffer *eb = path->nodes[level];
6019         u64 parent = 0;
6020
6021         if (wc->stage == UPDATE_BACKREF) {
6022                 BUG_ON(wc->shared_level < level);
6023                 if (level < wc->shared_level)
6024                         goto out;
6025
6026                 ret = find_next_key(path, level + 1, &wc->update_progress);
6027                 if (ret > 0)
6028                         wc->update_ref = 0;
6029
6030                 wc->stage = DROP_REFERENCE;
6031                 wc->shared_level = -1;
6032                 path->slots[level] = 0;
6033
6034                 /*
6035                  * check reference count again if the block isn't locked.
6036                  * we should start walking down the tree again if reference
6037                  * count is one.
6038                  */
6039                 if (!path->locks[level]) {
6040                         BUG_ON(level == 0);
6041                         btrfs_tree_lock(eb);
6042                         btrfs_set_lock_blocking(eb);
6043                         path->locks[level] = 1;
6044
6045                         ret = btrfs_lookup_extent_info(trans, root,
6046                                                        eb->start, eb->len,
6047                                                        &wc->refs[level],
6048                                                        &wc->flags[level]);
6049                         BUG_ON(ret);
6050                         BUG_ON(wc->refs[level] == 0);
6051                         if (wc->refs[level] == 1) {
6052                                 btrfs_tree_unlock(eb);
6053                                 path->locks[level] = 0;
6054                                 return 1;
6055                         }
6056                 }
6057         }
6058
6059         /* wc->stage == DROP_REFERENCE */
6060         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6061
6062         if (wc->refs[level] == 1) {
6063                 if (level == 0) {
6064                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6065                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6066                         else
6067                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6068                         BUG_ON(ret);
6069                 }
6070                 /* make block locked assertion in clean_tree_block happy */
6071                 if (!path->locks[level] &&
6072                     btrfs_header_generation(eb) == trans->transid) {
6073                         btrfs_tree_lock(eb);
6074                         btrfs_set_lock_blocking(eb);
6075                         path->locks[level] = 1;
6076                 }
6077                 clean_tree_block(trans, root, eb);
6078         }
6079
6080         if (eb == root->node) {
6081                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6082                         parent = eb->start;
6083                 else
6084                         BUG_ON(root->root_key.objectid !=
6085                                btrfs_header_owner(eb));
6086         } else {
6087                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6088                         parent = path->nodes[level + 1]->start;
6089                 else
6090                         BUG_ON(root->root_key.objectid !=
6091                                btrfs_header_owner(path->nodes[level + 1]));
6092         }
6093
6094         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6095 out:
6096         wc->refs[level] = 0;
6097         wc->flags[level] = 0;
6098         return 0;
6099 }
6100
6101 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6102                                    struct btrfs_root *root,
6103                                    struct btrfs_path *path,
6104                                    struct walk_control *wc)
6105 {
6106         int level = wc->level;
6107         int lookup_info = 1;
6108         int ret;
6109
6110         while (level >= 0) {
6111                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6112                 if (ret > 0)
6113                         break;
6114
6115                 if (level == 0)
6116                         break;
6117
6118                 if (path->slots[level] >=
6119                     btrfs_header_nritems(path->nodes[level]))
6120                         break;
6121
6122                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6123                 if (ret > 0) {
6124                         path->slots[level]++;
6125                         continue;
6126                 } else if (ret < 0)
6127                         return ret;
6128                 level = wc->level;
6129         }
6130         return 0;
6131 }
6132
6133 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6134                                  struct btrfs_root *root,
6135                                  struct btrfs_path *path,
6136                                  struct walk_control *wc, int max_level)
6137 {
6138         int level = wc->level;
6139         int ret;
6140
6141         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6142         while (level < max_level && path->nodes[level]) {
6143                 wc->level = level;
6144                 if (path->slots[level] + 1 <
6145                     btrfs_header_nritems(path->nodes[level])) {
6146                         path->slots[level]++;
6147                         return 0;
6148                 } else {
6149                         ret = walk_up_proc(trans, root, path, wc);
6150                         if (ret > 0)
6151                                 return 0;
6152
6153                         if (path->locks[level]) {
6154                                 btrfs_tree_unlock(path->nodes[level]);
6155                                 path->locks[level] = 0;
6156                         }
6157                         free_extent_buffer(path->nodes[level]);
6158                         path->nodes[level] = NULL;
6159                         level++;
6160                 }
6161         }
6162         return 1;
6163 }
6164
6165 /*
6166  * drop a subvolume tree.
6167  *
6168  * this function traverses the tree freeing any blocks that only
6169  * referenced by the tree.
6170  *
6171  * when a shared tree block is found. this function decreases its
6172  * reference count by one. if update_ref is true, this function
6173  * also make sure backrefs for the shared block and all lower level
6174  * blocks are properly updated.
6175  */
6176 int btrfs_drop_snapshot(struct btrfs_root *root,
6177                         struct btrfs_block_rsv *block_rsv, int update_ref)
6178 {
6179         struct btrfs_path *path;
6180         struct btrfs_trans_handle *trans;
6181         struct btrfs_root *tree_root = root->fs_info->tree_root;
6182         struct btrfs_root_item *root_item = &root->root_item;
6183         struct walk_control *wc;
6184         struct btrfs_key key;
6185         int err = 0;
6186         int ret;
6187         int level;
6188
6189         path = btrfs_alloc_path();
6190         BUG_ON(!path);
6191
6192         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6193         BUG_ON(!wc);
6194
6195         trans = btrfs_start_transaction(tree_root, 0);
6196         if (block_rsv)
6197                 trans->block_rsv = block_rsv;
6198
6199         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6200                 level = btrfs_header_level(root->node);
6201                 path->nodes[level] = btrfs_lock_root_node(root);
6202                 btrfs_set_lock_blocking(path->nodes[level]);
6203                 path->slots[level] = 0;
6204                 path->locks[level] = 1;
6205                 memset(&wc->update_progress, 0,
6206                        sizeof(wc->update_progress));
6207         } else {
6208                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6209                 memcpy(&wc->update_progress, &key,
6210                        sizeof(wc->update_progress));
6211
6212                 level = root_item->drop_level;
6213                 BUG_ON(level == 0);
6214                 path->lowest_level = level;
6215                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6216                 path->lowest_level = 0;
6217                 if (ret < 0) {
6218                         err = ret;
6219                         goto out;
6220                 }
6221                 WARN_ON(ret > 0);
6222
6223                 /*
6224                  * unlock our path, this is safe because only this
6225                  * function is allowed to delete this snapshot
6226                  */
6227                 btrfs_unlock_up_safe(path, 0);
6228
6229                 level = btrfs_header_level(root->node);
6230                 while (1) {
6231                         btrfs_tree_lock(path->nodes[level]);
6232                         btrfs_set_lock_blocking(path->nodes[level]);
6233
6234                         ret = btrfs_lookup_extent_info(trans, root,
6235                                                 path->nodes[level]->start,
6236                                                 path->nodes[level]->len,
6237                                                 &wc->refs[level],
6238                                                 &wc->flags[level]);
6239                         BUG_ON(ret);
6240                         BUG_ON(wc->refs[level] == 0);
6241
6242                         if (level == root_item->drop_level)
6243                                 break;
6244
6245                         btrfs_tree_unlock(path->nodes[level]);
6246                         WARN_ON(wc->refs[level] != 1);
6247                         level--;
6248                 }
6249         }
6250
6251         wc->level = level;
6252         wc->shared_level = -1;
6253         wc->stage = DROP_REFERENCE;
6254         wc->update_ref = update_ref;
6255         wc->keep_locks = 0;
6256         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6257
6258         while (1) {
6259                 ret = walk_down_tree(trans, root, path, wc);
6260                 if (ret < 0) {
6261                         err = ret;
6262                         break;
6263                 }
6264
6265                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6266                 if (ret < 0) {
6267                         err = ret;
6268                         break;
6269                 }
6270
6271                 if (ret > 0) {
6272                         BUG_ON(wc->stage != DROP_REFERENCE);
6273                         break;
6274                 }
6275
6276                 if (wc->stage == DROP_REFERENCE) {
6277                         level = wc->level;
6278                         btrfs_node_key(path->nodes[level],
6279                                        &root_item->drop_progress,
6280                                        path->slots[level]);
6281                         root_item->drop_level = level;
6282                 }
6283
6284                 BUG_ON(wc->level == 0);
6285                 if (btrfs_should_end_transaction(trans, tree_root)) {
6286                         ret = btrfs_update_root(trans, tree_root,
6287                                                 &root->root_key,
6288                                                 root_item);
6289                         BUG_ON(ret);
6290
6291                         btrfs_end_transaction_throttle(trans, tree_root);
6292                         trans = btrfs_start_transaction(tree_root, 0);
6293                         if (block_rsv)
6294                                 trans->block_rsv = block_rsv;
6295                 }
6296         }
6297         btrfs_release_path(root, path);
6298         BUG_ON(err);
6299
6300         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6301         BUG_ON(ret);
6302
6303         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6304                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6305                                            NULL, NULL);
6306                 BUG_ON(ret < 0);
6307                 if (ret > 0) {
6308                         ret = btrfs_del_orphan_item(trans, tree_root,
6309                                                     root->root_key.objectid);
6310                         BUG_ON(ret);
6311                 }
6312         }
6313
6314         if (root->in_radix) {
6315                 btrfs_free_fs_root(tree_root->fs_info, root);
6316         } else {
6317                 free_extent_buffer(root->node);
6318                 free_extent_buffer(root->commit_root);
6319                 kfree(root);
6320         }
6321 out:
6322         btrfs_end_transaction_throttle(trans, tree_root);
6323         kfree(wc);
6324         btrfs_free_path(path);
6325         return err;
6326 }
6327
6328 /*
6329  * drop subtree rooted at tree block 'node'.
6330  *
6331  * NOTE: this function will unlock and release tree block 'node'
6332  */
6333 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6334                         struct btrfs_root *root,
6335                         struct extent_buffer *node,
6336                         struct extent_buffer *parent)
6337 {
6338         struct btrfs_path *path;
6339         struct walk_control *wc;
6340         int level;
6341         int parent_level;
6342         int ret = 0;
6343         int wret;
6344
6345         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6346
6347         path = btrfs_alloc_path();
6348         BUG_ON(!path);
6349
6350         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6351         BUG_ON(!wc);
6352
6353         btrfs_assert_tree_locked(parent);
6354         parent_level = btrfs_header_level(parent);
6355         extent_buffer_get(parent);
6356         path->nodes[parent_level] = parent;
6357         path->slots[parent_level] = btrfs_header_nritems(parent);
6358
6359         btrfs_assert_tree_locked(node);
6360         level = btrfs_header_level(node);
6361         path->nodes[level] = node;
6362         path->slots[level] = 0;
6363         path->locks[level] = 1;
6364
6365         wc->refs[parent_level] = 1;
6366         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6367         wc->level = level;
6368         wc->shared_level = -1;
6369         wc->stage = DROP_REFERENCE;
6370         wc->update_ref = 0;
6371         wc->keep_locks = 1;
6372         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6373
6374         while (1) {
6375                 wret = walk_down_tree(trans, root, path, wc);
6376                 if (wret < 0) {
6377                         ret = wret;
6378                         break;
6379                 }
6380
6381                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6382                 if (wret < 0)
6383                         ret = wret;
6384                 if (wret != 0)
6385                         break;
6386         }
6387
6388         kfree(wc);
6389         btrfs_free_path(path);
6390         return ret;
6391 }
6392
6393 #if 0
6394 static unsigned long calc_ra(unsigned long start, unsigned long last,
6395                              unsigned long nr)
6396 {
6397         return min(last, start + nr - 1);
6398 }
6399
6400 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6401                                          u64 len)
6402 {
6403         u64 page_start;
6404         u64 page_end;
6405         unsigned long first_index;
6406         unsigned long last_index;
6407         unsigned long i;
6408         struct page *page;
6409         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6410         struct file_ra_state *ra;
6411         struct btrfs_ordered_extent *ordered;
6412         unsigned int total_read = 0;
6413         unsigned int total_dirty = 0;
6414         int ret = 0;
6415
6416         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6417
6418         mutex_lock(&inode->i_mutex);
6419         first_index = start >> PAGE_CACHE_SHIFT;
6420         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6421
6422         /* make sure the dirty trick played by the caller work */
6423         ret = invalidate_inode_pages2_range(inode->i_mapping,
6424                                             first_index, last_index);
6425         if (ret)
6426                 goto out_unlock;
6427
6428         file_ra_state_init(ra, inode->i_mapping);
6429
6430         for (i = first_index ; i <= last_index; i++) {
6431                 if (total_read % ra->ra_pages == 0) {
6432                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6433                                        calc_ra(i, last_index, ra->ra_pages));
6434                 }
6435                 total_read++;
6436 again:
6437                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6438                         BUG_ON(1);
6439                 page = grab_cache_page(inode->i_mapping, i);
6440                 if (!page) {
6441                         ret = -ENOMEM;
6442                         goto out_unlock;
6443                 }
6444                 if (!PageUptodate(page)) {
6445                         btrfs_readpage(NULL, page);
6446                         lock_page(page);
6447                         if (!PageUptodate(page)) {
6448                                 unlock_page(page);
6449                                 page_cache_release(page);
6450                                 ret = -EIO;
6451                                 goto out_unlock;
6452                         }
6453                 }
6454                 wait_on_page_writeback(page);
6455
6456                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6457                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6458                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6459
6460                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6461                 if (ordered) {
6462                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6463                         unlock_page(page);
6464                         page_cache_release(page);
6465                         btrfs_start_ordered_extent(inode, ordered, 1);
6466                         btrfs_put_ordered_extent(ordered);
6467                         goto again;
6468                 }
6469                 set_page_extent_mapped(page);
6470
6471                 if (i == first_index)
6472                         set_extent_bits(io_tree, page_start, page_end,
6473                                         EXTENT_BOUNDARY, GFP_NOFS);
6474                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6475
6476                 set_page_dirty(page);
6477                 total_dirty++;
6478
6479                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6480                 unlock_page(page);
6481                 page_cache_release(page);
6482         }
6483
6484 out_unlock:
6485         kfree(ra);
6486         mutex_unlock(&inode->i_mutex);
6487         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6488         return ret;
6489 }
6490
6491 static noinline int relocate_data_extent(struct inode *reloc_inode,
6492                                          struct btrfs_key *extent_key,
6493                                          u64 offset)
6494 {
6495         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6496         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6497         struct extent_map *em;
6498         u64 start = extent_key->objectid - offset;
6499         u64 end = start + extent_key->offset - 1;
6500
6501         em = alloc_extent_map(GFP_NOFS);
6502         BUG_ON(!em || IS_ERR(em));
6503
6504         em->start = start;
6505         em->len = extent_key->offset;
6506         em->block_len = extent_key->offset;
6507         em->block_start = extent_key->objectid;
6508         em->bdev = root->fs_info->fs_devices->latest_bdev;
6509         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6510
6511         /* setup extent map to cheat btrfs_readpage */
6512         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6513         while (1) {
6514                 int ret;
6515                 write_lock(&em_tree->lock);
6516                 ret = add_extent_mapping(em_tree, em);
6517                 write_unlock(&em_tree->lock);
6518                 if (ret != -EEXIST) {
6519                         free_extent_map(em);
6520                         break;
6521                 }
6522                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6523         }
6524         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6525
6526         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6527 }
6528
6529 struct btrfs_ref_path {
6530         u64 extent_start;
6531         u64 nodes[BTRFS_MAX_LEVEL];
6532         u64 root_objectid;
6533         u64 root_generation;
6534         u64 owner_objectid;
6535         u32 num_refs;
6536         int lowest_level;
6537         int current_level;
6538         int shared_level;
6539
6540         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6541         u64 new_nodes[BTRFS_MAX_LEVEL];
6542 };
6543
6544 struct disk_extent {
6545         u64 ram_bytes;
6546         u64 disk_bytenr;
6547         u64 disk_num_bytes;
6548         u64 offset;
6549         u64 num_bytes;
6550         u8 compression;
6551         u8 encryption;
6552         u16 other_encoding;
6553 };
6554
6555 static int is_cowonly_root(u64 root_objectid)
6556 {
6557         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6558             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6559             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6560             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6561             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6562             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6563                 return 1;
6564         return 0;
6565 }
6566
6567 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6568                                     struct btrfs_root *extent_root,
6569                                     struct btrfs_ref_path *ref_path,
6570                                     int first_time)
6571 {
6572         struct extent_buffer *leaf;
6573         struct btrfs_path *path;
6574         struct btrfs_extent_ref *ref;
6575         struct btrfs_key key;
6576         struct btrfs_key found_key;
6577         u64 bytenr;
6578         u32 nritems;
6579         int level;
6580         int ret = 1;
6581
6582         path = btrfs_alloc_path();
6583         if (!path)
6584                 return -ENOMEM;
6585
6586         if (first_time) {
6587                 ref_path->lowest_level = -1;
6588                 ref_path->current_level = -1;
6589                 ref_path->shared_level = -1;
6590                 goto walk_up;
6591         }
6592 walk_down:
6593         level = ref_path->current_level - 1;
6594         while (level >= -1) {
6595                 u64 parent;
6596                 if (level < ref_path->lowest_level)
6597                         break;
6598
6599                 if (level >= 0)
6600                         bytenr = ref_path->nodes[level];
6601                 else
6602                         bytenr = ref_path->extent_start;
6603                 BUG_ON(bytenr == 0);
6604
6605                 parent = ref_path->nodes[level + 1];
6606                 ref_path->nodes[level + 1] = 0;
6607                 ref_path->current_level = level;
6608                 BUG_ON(parent == 0);
6609
6610                 key.objectid = bytenr;
6611                 key.offset = parent + 1;
6612                 key.type = BTRFS_EXTENT_REF_KEY;
6613
6614                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6615                 if (ret < 0)
6616                         goto out;
6617                 BUG_ON(ret == 0);
6618
6619                 leaf = path->nodes[0];
6620                 nritems = btrfs_header_nritems(leaf);
6621                 if (path->slots[0] >= nritems) {
6622                         ret = btrfs_next_leaf(extent_root, path);
6623                         if (ret < 0)
6624                                 goto out;
6625                         if (ret > 0)
6626                                 goto next;
6627                         leaf = path->nodes[0];
6628                 }
6629
6630                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6631                 if (found_key.objectid == bytenr &&
6632                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6633                         if (level < ref_path->shared_level)
6634                                 ref_path->shared_level = level;
6635                         goto found;
6636                 }
6637 next:
6638                 level--;
6639                 btrfs_release_path(extent_root, path);
6640                 cond_resched();
6641         }
6642         /* reached lowest level */
6643         ret = 1;
6644         goto out;
6645 walk_up:
6646         level = ref_path->current_level;
6647         while (level < BTRFS_MAX_LEVEL - 1) {
6648                 u64 ref_objectid;
6649
6650                 if (level >= 0)
6651                         bytenr = ref_path->nodes[level];
6652                 else
6653                         bytenr = ref_path->extent_start;
6654
6655                 BUG_ON(bytenr == 0);
6656
6657                 key.objectid = bytenr;
6658                 key.offset = 0;
6659                 key.type = BTRFS_EXTENT_REF_KEY;
6660
6661                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6662                 if (ret < 0)
6663                         goto out;
6664
6665                 leaf = path->nodes[0];
6666                 nritems = btrfs_header_nritems(leaf);
6667                 if (path->slots[0] >= nritems) {
6668                         ret = btrfs_next_leaf(extent_root, path);
6669                         if (ret < 0)
6670                                 goto out;
6671                         if (ret > 0) {
6672                                 /* the extent was freed by someone */
6673                                 if (ref_path->lowest_level == level)
6674                                         goto out;
6675                                 btrfs_release_path(extent_root, path);
6676                                 goto walk_down;
6677                         }
6678                         leaf = path->nodes[0];
6679                 }
6680
6681                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6682                 if (found_key.objectid != bytenr ||
6683                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6684                         /* the extent was freed by someone */
6685                         if (ref_path->lowest_level == level) {
6686                                 ret = 1;
6687                                 goto out;
6688                         }
6689                         btrfs_release_path(extent_root, path);
6690                         goto walk_down;
6691                 }
6692 found:
6693                 ref = btrfs_item_ptr(leaf, path->slots[0],
6694                                 struct btrfs_extent_ref);
6695                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6696                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6697                         if (first_time) {
6698                                 level = (int)ref_objectid;
6699                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6700                                 ref_path->lowest_level = level;
6701                                 ref_path->current_level = level;
6702                                 ref_path->nodes[level] = bytenr;
6703                         } else {
6704                                 WARN_ON(ref_objectid != level);
6705                         }
6706                 } else {
6707                         WARN_ON(level != -1);
6708                 }
6709                 first_time = 0;
6710
6711                 if (ref_path->lowest_level == level) {
6712                         ref_path->owner_objectid = ref_objectid;
6713                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6714                 }
6715
6716                 /*
6717                  * the block is tree root or the block isn't in reference
6718                  * counted tree.
6719                  */
6720                 if (found_key.objectid == found_key.offset ||
6721                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6722                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6723                         ref_path->root_generation =
6724                                 btrfs_ref_generation(leaf, ref);
6725                         if (level < 0) {
6726                                 /* special reference from the tree log */
6727                                 ref_path->nodes[0] = found_key.offset;
6728                                 ref_path->current_level = 0;
6729                         }
6730                         ret = 0;
6731                         goto out;
6732                 }
6733
6734                 level++;
6735                 BUG_ON(ref_path->nodes[level] != 0);
6736                 ref_path->nodes[level] = found_key.offset;
6737                 ref_path->current_level = level;
6738
6739                 /*
6740                  * the reference was created in the running transaction,
6741                  * no need to continue walking up.
6742                  */
6743                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6744                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6745                         ref_path->root_generation =
6746                                 btrfs_ref_generation(leaf, ref);
6747                         ret = 0;
6748                         goto out;
6749                 }
6750
6751                 btrfs_release_path(extent_root, path);
6752                 cond_resched();
6753         }
6754         /* reached max tree level, but no tree root found. */
6755         BUG();
6756 out:
6757         btrfs_free_path(path);
6758         return ret;
6759 }
6760
6761 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6762                                 struct btrfs_root *extent_root,
6763                                 struct btrfs_ref_path *ref_path,
6764                                 u64 extent_start)
6765 {
6766         memset(ref_path, 0, sizeof(*ref_path));
6767         ref_path->extent_start = extent_start;
6768
6769         return __next_ref_path(trans, extent_root, ref_path, 1);
6770 }
6771
6772 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6773                                struct btrfs_root *extent_root,
6774                                struct btrfs_ref_path *ref_path)
6775 {
6776         return __next_ref_path(trans, extent_root, ref_path, 0);
6777 }
6778
6779 static noinline int get_new_locations(struct inode *reloc_inode,
6780                                       struct btrfs_key *extent_key,
6781                                       u64 offset, int no_fragment,
6782                                       struct disk_extent **extents,
6783                                       int *nr_extents)
6784 {
6785         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6786         struct btrfs_path *path;
6787         struct btrfs_file_extent_item *fi;
6788         struct extent_buffer *leaf;
6789         struct disk_extent *exts = *extents;
6790         struct btrfs_key found_key;
6791         u64 cur_pos;
6792         u64 last_byte;
6793         u32 nritems;
6794         int nr = 0;
6795         int max = *nr_extents;
6796         int ret;
6797
6798         WARN_ON(!no_fragment && *extents);
6799         if (!exts) {
6800                 max = 1;
6801                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6802                 if (!exts)
6803                         return -ENOMEM;
6804         }
6805
6806         path = btrfs_alloc_path();
6807         BUG_ON(!path);
6808
6809         cur_pos = extent_key->objectid - offset;
6810         last_byte = extent_key->objectid + extent_key->offset;
6811         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6812                                        cur_pos, 0);
6813         if (ret < 0)
6814                 goto out;
6815         if (ret > 0) {
6816                 ret = -ENOENT;
6817                 goto out;
6818         }
6819
6820         while (1) {
6821                 leaf = path->nodes[0];
6822                 nritems = btrfs_header_nritems(leaf);
6823                 if (path->slots[0] >= nritems) {
6824                         ret = btrfs_next_leaf(root, path);
6825                         if (ret < 0)
6826                                 goto out;
6827                         if (ret > 0)
6828                                 break;
6829                         leaf = path->nodes[0];
6830                 }
6831
6832                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6833                 if (found_key.offset != cur_pos ||
6834                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6835                     found_key.objectid != reloc_inode->i_ino)
6836                         break;
6837
6838                 fi = btrfs_item_ptr(leaf, path->slots[0],
6839                                     struct btrfs_file_extent_item);
6840                 if (btrfs_file_extent_type(leaf, fi) !=
6841                     BTRFS_FILE_EXTENT_REG ||
6842                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6843                         break;
6844
6845                 if (nr == max) {
6846                         struct disk_extent *old = exts;
6847                         max *= 2;
6848                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6849                         memcpy(exts, old, sizeof(*exts) * nr);
6850                         if (old != *extents)
6851                                 kfree(old);
6852                 }
6853
6854                 exts[nr].disk_bytenr =
6855                         btrfs_file_extent_disk_bytenr(leaf, fi);
6856                 exts[nr].disk_num_bytes =
6857                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6858                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6859                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6860                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6861                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6862                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6863                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6864                                                                            fi);
6865                 BUG_ON(exts[nr].offset > 0);
6866                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6867                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6868
6869                 cur_pos += exts[nr].num_bytes;
6870                 nr++;
6871
6872                 if (cur_pos + offset >= last_byte)
6873                         break;
6874
6875                 if (no_fragment) {
6876                         ret = 1;
6877                         goto out;
6878                 }
6879                 path->slots[0]++;
6880         }
6881
6882         BUG_ON(cur_pos + offset > last_byte);
6883         if (cur_pos + offset < last_byte) {
6884                 ret = -ENOENT;
6885                 goto out;
6886         }
6887         ret = 0;
6888 out:
6889         btrfs_free_path(path);
6890         if (ret) {
6891                 if (exts != *extents)
6892                         kfree(exts);
6893         } else {
6894                 *extents = exts;
6895                 *nr_extents = nr;
6896         }
6897         return ret;
6898 }
6899
6900 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6901                                         struct btrfs_root *root,
6902                                         struct btrfs_path *path,
6903                                         struct btrfs_key *extent_key,
6904                                         struct btrfs_key *leaf_key,
6905                                         struct btrfs_ref_path *ref_path,
6906                                         struct disk_extent *new_extents,
6907                                         int nr_extents)
6908 {
6909         struct extent_buffer *leaf;
6910         struct btrfs_file_extent_item *fi;
6911         struct inode *inode = NULL;
6912         struct btrfs_key key;
6913         u64 lock_start = 0;
6914         u64 lock_end = 0;
6915         u64 num_bytes;
6916         u64 ext_offset;
6917         u64 search_end = (u64)-1;
6918         u32 nritems;
6919         int nr_scaned = 0;
6920         int extent_locked = 0;
6921         int extent_type;
6922         int ret;
6923
6924         memcpy(&key, leaf_key, sizeof(key));
6925         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6926                 if (key.objectid < ref_path->owner_objectid ||
6927                     (key.objectid == ref_path->owner_objectid &&
6928                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6929                         key.objectid = ref_path->owner_objectid;
6930                         key.type = BTRFS_EXTENT_DATA_KEY;
6931                         key.offset = 0;
6932                 }
6933         }
6934
6935         while (1) {
6936                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6937                 if (ret < 0)
6938                         goto out;
6939
6940                 leaf = path->nodes[0];
6941                 nritems = btrfs_header_nritems(leaf);
6942 next:
6943                 if (extent_locked && ret > 0) {
6944                         /*
6945                          * the file extent item was modified by someone
6946                          * before the extent got locked.
6947                          */
6948                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6949                                       lock_end, GFP_NOFS);
6950                         extent_locked = 0;
6951                 }
6952
6953                 if (path->slots[0] >= nritems) {
6954                         if (++nr_scaned > 2)
6955                                 break;
6956
6957                         BUG_ON(extent_locked);
6958                         ret = btrfs_next_leaf(root, path);
6959                         if (ret < 0)
6960                                 goto out;
6961                         if (ret > 0)
6962                                 break;
6963                         leaf = path->nodes[0];
6964                         nritems = btrfs_header_nritems(leaf);
6965                 }
6966
6967                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6968
6969                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6970                         if ((key.objectid > ref_path->owner_objectid) ||
6971                             (key.objectid == ref_path->owner_objectid &&
6972                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6973                             key.offset >= search_end)
6974                                 break;
6975                 }
6976
6977                 if (inode && key.objectid != inode->i_ino) {
6978                         BUG_ON(extent_locked);
6979                         btrfs_release_path(root, path);
6980                         mutex_unlock(&inode->i_mutex);
6981                         iput(inode);
6982                         inode = NULL;
6983                         continue;
6984                 }
6985
6986                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6987                         path->slots[0]++;
6988                         ret = 1;
6989                         goto next;
6990                 }
6991                 fi = btrfs_item_ptr(leaf, path->slots[0],
6992                                     struct btrfs_file_extent_item);
6993                 extent_type = btrfs_file_extent_type(leaf, fi);
6994                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6995                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6996                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6997                      extent_key->objectid)) {
6998                         path->slots[0]++;
6999                         ret = 1;
7000                         goto next;
7001                 }
7002
7003                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7004                 ext_offset = btrfs_file_extent_offset(leaf, fi);
7005
7006                 if (search_end == (u64)-1) {
7007                         search_end = key.offset - ext_offset +
7008                                 btrfs_file_extent_ram_bytes(leaf, fi);
7009                 }
7010
7011                 if (!extent_locked) {
7012                         lock_start = key.offset;
7013                         lock_end = lock_start + num_bytes - 1;
7014                 } else {
7015                         if (lock_start > key.offset ||
7016                             lock_end + 1 < key.offset + num_bytes) {
7017                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7018                                               lock_start, lock_end, GFP_NOFS);
7019                                 extent_locked = 0;
7020                         }
7021                 }
7022
7023                 if (!inode) {
7024                         btrfs_release_path(root, path);
7025
7026                         inode = btrfs_iget_locked(root->fs_info->sb,
7027                                                   key.objectid, root);
7028                         if (inode->i_state & I_NEW) {
7029                                 BTRFS_I(inode)->root = root;
7030                                 BTRFS_I(inode)->location.objectid =
7031                                         key.objectid;
7032                                 BTRFS_I(inode)->location.type =
7033                                         BTRFS_INODE_ITEM_KEY;
7034                                 BTRFS_I(inode)->location.offset = 0;
7035                                 btrfs_read_locked_inode(inode);
7036                                 unlock_new_inode(inode);
7037                         }
7038                         /*
7039                          * some code call btrfs_commit_transaction while
7040                          * holding the i_mutex, so we can't use mutex_lock
7041                          * here.
7042                          */
7043                         if (is_bad_inode(inode) ||
7044                             !mutex_trylock(&inode->i_mutex)) {
7045                                 iput(inode);
7046                                 inode = NULL;
7047                                 key.offset = (u64)-1;
7048                                 goto skip;
7049                         }
7050                 }
7051
7052                 if (!extent_locked) {
7053                         struct btrfs_ordered_extent *ordered;
7054
7055                         btrfs_release_path(root, path);
7056
7057                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7058                                     lock_end, GFP_NOFS);
7059                         ordered = btrfs_lookup_first_ordered_extent(inode,
7060                                                                     lock_end);
7061                         if (ordered &&
7062                             ordered->file_offset <= lock_end &&
7063                             ordered->file_offset + ordered->len > lock_start) {
7064                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7065                                               lock_start, lock_end, GFP_NOFS);
7066                                 btrfs_start_ordered_extent(inode, ordered, 1);
7067                                 btrfs_put_ordered_extent(ordered);
7068                                 key.offset += num_bytes;
7069                                 goto skip;
7070                         }
7071                         if (ordered)
7072                                 btrfs_put_ordered_extent(ordered);
7073
7074                         extent_locked = 1;
7075                         continue;
7076                 }
7077
7078                 if (nr_extents == 1) {
7079                         /* update extent pointer in place */
7080                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7081                                                 new_extents[0].disk_bytenr);
7082                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7083                                                 new_extents[0].disk_num_bytes);
7084                         btrfs_mark_buffer_dirty(leaf);
7085
7086                         btrfs_drop_extent_cache(inode, key.offset,
7087                                                 key.offset + num_bytes - 1, 0);
7088
7089                         ret = btrfs_inc_extent_ref(trans, root,
7090                                                 new_extents[0].disk_bytenr,
7091                                                 new_extents[0].disk_num_bytes,
7092                                                 leaf->start,
7093                                                 root->root_key.objectid,
7094                                                 trans->transid,
7095                                                 key.objectid);
7096                         BUG_ON(ret);
7097
7098                         ret = btrfs_free_extent(trans, root,
7099                                                 extent_key->objectid,
7100                                                 extent_key->offset,
7101                                                 leaf->start,
7102                                                 btrfs_header_owner(leaf),
7103                                                 btrfs_header_generation(leaf),
7104                                                 key.objectid, 0);
7105                         BUG_ON(ret);
7106
7107                         btrfs_release_path(root, path);
7108                         key.offset += num_bytes;
7109                 } else {
7110                         BUG_ON(1);
7111 #if 0
7112                         u64 alloc_hint;
7113                         u64 extent_len;
7114                         int i;
7115                         /*
7116                          * drop old extent pointer at first, then insert the
7117                          * new pointers one bye one
7118                          */
7119                         btrfs_release_path(root, path);
7120                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7121                                                  key.offset + num_bytes,
7122                                                  key.offset, &alloc_hint);
7123                         BUG_ON(ret);
7124
7125                         for (i = 0; i < nr_extents; i++) {
7126                                 if (ext_offset >= new_extents[i].num_bytes) {
7127                                         ext_offset -= new_extents[i].num_bytes;
7128                                         continue;
7129                                 }
7130                                 extent_len = min(new_extents[i].num_bytes -
7131                                                  ext_offset, num_bytes);
7132
7133                                 ret = btrfs_insert_empty_item(trans, root,
7134                                                               path, &key,
7135                                                               sizeof(*fi));
7136                                 BUG_ON(ret);
7137
7138                                 leaf = path->nodes[0];
7139                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7140                                                 struct btrfs_file_extent_item);
7141                                 btrfs_set_file_extent_generation(leaf, fi,
7142                                                         trans->transid);
7143                                 btrfs_set_file_extent_type(leaf, fi,
7144                                                         BTRFS_FILE_EXTENT_REG);
7145                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7146                                                 new_extents[i].disk_bytenr);
7147                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7148                                                 new_extents[i].disk_num_bytes);
7149                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7150                                                 new_extents[i].ram_bytes);
7151
7152                                 btrfs_set_file_extent_compression(leaf, fi,
7153                                                 new_extents[i].compression);
7154                                 btrfs_set_file_extent_encryption(leaf, fi,
7155                                                 new_extents[i].encryption);
7156                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7157                                                 new_extents[i].other_encoding);
7158
7159                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7160                                                         extent_len);
7161                                 ext_offset += new_extents[i].offset;
7162                                 btrfs_set_file_extent_offset(leaf, fi,
7163                                                         ext_offset);
7164                                 btrfs_mark_buffer_dirty(leaf);
7165
7166                                 btrfs_drop_extent_cache(inode, key.offset,
7167                                                 key.offset + extent_len - 1, 0);
7168
7169                                 ret = btrfs_inc_extent_ref(trans, root,
7170                                                 new_extents[i].disk_bytenr,
7171                                                 new_extents[i].disk_num_bytes,
7172                                                 leaf->start,
7173                                                 root->root_key.objectid,
7174                                                 trans->transid, key.objectid);
7175                                 BUG_ON(ret);
7176                                 btrfs_release_path(root, path);
7177
7178                                 inode_add_bytes(inode, extent_len);
7179
7180                                 ext_offset = 0;
7181                                 num_bytes -= extent_len;
7182                                 key.offset += extent_len;
7183
7184                                 if (num_bytes == 0)
7185                                         break;
7186                         }
7187                         BUG_ON(i >= nr_extents);
7188 #endif
7189                 }
7190
7191                 if (extent_locked) {
7192                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7193                                       lock_end, GFP_NOFS);
7194                         extent_locked = 0;
7195                 }
7196 skip:
7197                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7198                     key.offset >= search_end)
7199                         break;
7200
7201                 cond_resched();
7202         }
7203         ret = 0;
7204 out:
7205         btrfs_release_path(root, path);
7206         if (inode) {
7207                 mutex_unlock(&inode->i_mutex);
7208                 if (extent_locked) {
7209                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7210                                       lock_end, GFP_NOFS);
7211                 }
7212                 iput(inode);
7213         }
7214         return ret;
7215 }
7216
7217 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7218                                struct btrfs_root *root,
7219                                struct extent_buffer *buf, u64 orig_start)
7220 {
7221         int level;
7222         int ret;
7223
7224         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7225         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7226
7227         level = btrfs_header_level(buf);
7228         if (level == 0) {
7229                 struct btrfs_leaf_ref *ref;
7230                 struct btrfs_leaf_ref *orig_ref;
7231
7232                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7233                 if (!orig_ref)
7234                         return -ENOENT;
7235
7236                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7237                 if (!ref) {
7238                         btrfs_free_leaf_ref(root, orig_ref);
7239                         return -ENOMEM;
7240                 }
7241
7242                 ref->nritems = orig_ref->nritems;
7243                 memcpy(ref->extents, orig_ref->extents,
7244                         sizeof(ref->extents[0]) * ref->nritems);
7245
7246                 btrfs_free_leaf_ref(root, orig_ref);
7247
7248                 ref->root_gen = trans->transid;
7249                 ref->bytenr = buf->start;
7250                 ref->owner = btrfs_header_owner(buf);
7251                 ref->generation = btrfs_header_generation(buf);
7252
7253                 ret = btrfs_add_leaf_ref(root, ref, 0);
7254                 WARN_ON(ret);
7255                 btrfs_free_leaf_ref(root, ref);
7256         }
7257         return 0;
7258 }
7259
7260 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7261                                         struct extent_buffer *leaf,
7262                                         struct btrfs_block_group_cache *group,
7263                                         struct btrfs_root *target_root)
7264 {
7265         struct btrfs_key key;
7266         struct inode *inode = NULL;
7267         struct btrfs_file_extent_item *fi;
7268         struct extent_state *cached_state = NULL;
7269         u64 num_bytes;
7270         u64 skip_objectid = 0;
7271         u32 nritems;
7272         u32 i;
7273
7274         nritems = btrfs_header_nritems(leaf);
7275         for (i = 0; i < nritems; i++) {
7276                 btrfs_item_key_to_cpu(leaf, &key, i);
7277                 if (key.objectid == skip_objectid ||
7278                     key.type != BTRFS_EXTENT_DATA_KEY)
7279                         continue;
7280                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7281                 if (btrfs_file_extent_type(leaf, fi) ==
7282                     BTRFS_FILE_EXTENT_INLINE)
7283                         continue;
7284                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7285                         continue;
7286                 if (!inode || inode->i_ino != key.objectid) {
7287                         iput(inode);
7288                         inode = btrfs_ilookup(target_root->fs_info->sb,
7289                                               key.objectid, target_root, 1);
7290                 }
7291                 if (!inode) {
7292                         skip_objectid = key.objectid;
7293                         continue;
7294                 }
7295                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7296
7297                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7298                                  key.offset + num_bytes - 1, 0, &cached_state,
7299                                  GFP_NOFS);
7300                 btrfs_drop_extent_cache(inode, key.offset,
7301                                         key.offset + num_bytes - 1, 1);
7302                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7303                                      key.offset + num_bytes - 1, &cached_state,
7304                                      GFP_NOFS);
7305                 cond_resched();
7306         }
7307         iput(inode);
7308         return 0;
7309 }
7310
7311 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7312                                         struct btrfs_root *root,
7313                                         struct extent_buffer *leaf,
7314                                         struct btrfs_block_group_cache *group,
7315                                         struct inode *reloc_inode)
7316 {
7317         struct btrfs_key key;
7318         struct btrfs_key extent_key;
7319         struct btrfs_file_extent_item *fi;
7320         struct btrfs_leaf_ref *ref;
7321         struct disk_extent *new_extent;
7322         u64 bytenr;
7323         u64 num_bytes;
7324         u32 nritems;
7325         u32 i;
7326         int ext_index;
7327         int nr_extent;
7328         int ret;
7329
7330         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7331         BUG_ON(!new_extent);
7332
7333         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7334         BUG_ON(!ref);
7335
7336         ext_index = -1;
7337         nritems = btrfs_header_nritems(leaf);
7338         for (i = 0; i < nritems; i++) {
7339                 btrfs_item_key_to_cpu(leaf, &key, i);
7340                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7341                         continue;
7342                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7343                 if (btrfs_file_extent_type(leaf, fi) ==
7344                     BTRFS_FILE_EXTENT_INLINE)
7345                         continue;
7346                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7347                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7348                 if (bytenr == 0)
7349                         continue;
7350
7351                 ext_index++;
7352                 if (bytenr >= group->key.objectid + group->key.offset ||
7353                     bytenr + num_bytes <= group->key.objectid)
7354                         continue;
7355
7356                 extent_key.objectid = bytenr;
7357                 extent_key.offset = num_bytes;
7358                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7359                 nr_extent = 1;
7360                 ret = get_new_locations(reloc_inode, &extent_key,
7361                                         group->key.objectid, 1,
7362                                         &new_extent, &nr_extent);
7363                 if (ret > 0)
7364                         continue;
7365                 BUG_ON(ret < 0);
7366
7367                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7368                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7369                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7370                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7371
7372                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7373                                                 new_extent->disk_bytenr);
7374                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7375                                                 new_extent->disk_num_bytes);
7376                 btrfs_mark_buffer_dirty(leaf);
7377
7378                 ret = btrfs_inc_extent_ref(trans, root,
7379                                         new_extent->disk_bytenr,
7380                                         new_extent->disk_num_bytes,
7381                                         leaf->start,
7382                                         root->root_key.objectid,
7383                                         trans->transid, key.objectid);
7384                 BUG_ON(ret);
7385
7386                 ret = btrfs_free_extent(trans, root,
7387                                         bytenr, num_bytes, leaf->start,
7388                                         btrfs_header_owner(leaf),
7389                                         btrfs_header_generation(leaf),
7390                                         key.objectid, 0);
7391                 BUG_ON(ret);
7392                 cond_resched();
7393         }
7394         kfree(new_extent);
7395         BUG_ON(ext_index + 1 != ref->nritems);
7396         btrfs_free_leaf_ref(root, ref);
7397         return 0;
7398 }
7399
7400 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7401                           struct btrfs_root *root)
7402 {
7403         struct btrfs_root *reloc_root;
7404         int ret;
7405
7406         if (root->reloc_root) {
7407                 reloc_root = root->reloc_root;
7408                 root->reloc_root = NULL;
7409                 list_add(&reloc_root->dead_list,
7410                          &root->fs_info->dead_reloc_roots);
7411
7412                 btrfs_set_root_bytenr(&reloc_root->root_item,
7413                                       reloc_root->node->start);
7414                 btrfs_set_root_level(&root->root_item,
7415                                      btrfs_header_level(reloc_root->node));
7416                 memset(&reloc_root->root_item.drop_progress, 0,
7417                         sizeof(struct btrfs_disk_key));
7418                 reloc_root->root_item.drop_level = 0;
7419
7420                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7421                                         &reloc_root->root_key,
7422                                         &reloc_root->root_item);
7423                 BUG_ON(ret);
7424         }
7425         return 0;
7426 }
7427
7428 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7429 {
7430         struct btrfs_trans_handle *trans;
7431         struct btrfs_root *reloc_root;
7432         struct btrfs_root *prev_root = NULL;
7433         struct list_head dead_roots;
7434         int ret;
7435         unsigned long nr;
7436
7437         INIT_LIST_HEAD(&dead_roots);
7438         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7439
7440         while (!list_empty(&dead_roots)) {
7441                 reloc_root = list_entry(dead_roots.prev,
7442                                         struct btrfs_root, dead_list);
7443                 list_del_init(&reloc_root->dead_list);
7444
7445                 BUG_ON(reloc_root->commit_root != NULL);
7446                 while (1) {
7447                         trans = btrfs_join_transaction(root, 1);
7448                         BUG_ON(!trans);
7449
7450                         mutex_lock(&root->fs_info->drop_mutex);
7451                         ret = btrfs_drop_snapshot(trans, reloc_root);
7452                         if (ret != -EAGAIN)
7453                                 break;
7454                         mutex_unlock(&root->fs_info->drop_mutex);
7455
7456                         nr = trans->blocks_used;
7457                         ret = btrfs_end_transaction(trans, root);
7458                         BUG_ON(ret);
7459                         btrfs_btree_balance_dirty(root, nr);
7460                 }
7461
7462                 free_extent_buffer(reloc_root->node);
7463
7464                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7465                                      &reloc_root->root_key);
7466                 BUG_ON(ret);
7467                 mutex_unlock(&root->fs_info->drop_mutex);
7468
7469                 nr = trans->blocks_used;
7470                 ret = btrfs_end_transaction(trans, root);
7471                 BUG_ON(ret);
7472                 btrfs_btree_balance_dirty(root, nr);
7473
7474                 kfree(prev_root);
7475                 prev_root = reloc_root;
7476         }
7477         if (prev_root) {
7478                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7479                 kfree(prev_root);
7480         }
7481         return 0;
7482 }
7483
7484 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7485 {
7486         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7487         return 0;
7488 }
7489
7490 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7491 {
7492         struct btrfs_root *reloc_root;
7493         struct btrfs_trans_handle *trans;
7494         struct btrfs_key location;
7495         int found;
7496         int ret;
7497
7498         mutex_lock(&root->fs_info->tree_reloc_mutex);
7499         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7500         BUG_ON(ret);
7501         found = !list_empty(&root->fs_info->dead_reloc_roots);
7502         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7503
7504         if (found) {
7505                 trans = btrfs_start_transaction(root, 1);
7506                 BUG_ON(!trans);
7507                 ret = btrfs_commit_transaction(trans, root);
7508                 BUG_ON(ret);
7509         }
7510
7511         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7512         location.offset = (u64)-1;
7513         location.type = BTRFS_ROOT_ITEM_KEY;
7514
7515         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7516         BUG_ON(!reloc_root);
7517         btrfs_orphan_cleanup(reloc_root);
7518         return 0;
7519 }
7520
7521 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7522                                     struct btrfs_root *root)
7523 {
7524         struct btrfs_root *reloc_root;
7525         struct extent_buffer *eb;
7526         struct btrfs_root_item *root_item;
7527         struct btrfs_key root_key;
7528         int ret;
7529
7530         BUG_ON(!root->ref_cows);
7531         if (root->reloc_root)
7532                 return 0;
7533
7534         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7535         BUG_ON(!root_item);
7536
7537         ret = btrfs_copy_root(trans, root, root->commit_root,
7538                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7539         BUG_ON(ret);
7540
7541         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7542         root_key.offset = root->root_key.objectid;
7543         root_key.type = BTRFS_ROOT_ITEM_KEY;
7544
7545         memcpy(root_item, &root->root_item, sizeof(root_item));
7546         btrfs_set_root_refs(root_item, 0);
7547         btrfs_set_root_bytenr(root_item, eb->start);
7548         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7549         btrfs_set_root_generation(root_item, trans->transid);
7550
7551         btrfs_tree_unlock(eb);
7552         free_extent_buffer(eb);
7553
7554         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7555                                 &root_key, root_item);
7556         BUG_ON(ret);
7557         kfree(root_item);
7558
7559         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7560                                                  &root_key);
7561         BUG_ON(!reloc_root);
7562         reloc_root->last_trans = trans->transid;
7563         reloc_root->commit_root = NULL;
7564         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7565
7566         root->reloc_root = reloc_root;
7567         return 0;
7568 }
7569
7570 /*
7571  * Core function of space balance.
7572  *
7573  * The idea is using reloc trees to relocate tree blocks in reference
7574  * counted roots. There is one reloc tree for each subvol, and all
7575  * reloc trees share same root key objectid. Reloc trees are snapshots
7576  * of the latest committed roots of subvols (root->commit_root).
7577  *
7578  * To relocate a tree block referenced by a subvol, there are two steps.
7579  * COW the block through subvol's reloc tree, then update block pointer
7580  * in the subvol to point to the new block. Since all reloc trees share
7581  * same root key objectid, doing special handing for tree blocks owned
7582  * by them is easy. Once a tree block has been COWed in one reloc tree,
7583  * we can use the resulting new block directly when the same block is
7584  * required to COW again through other reloc trees. By this way, relocated
7585  * tree blocks are shared between reloc trees, so they are also shared
7586  * between subvols.
7587  */
7588 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7589                                       struct btrfs_root *root,
7590                                       struct btrfs_path *path,
7591                                       struct btrfs_key *first_key,
7592                                       struct btrfs_ref_path *ref_path,
7593                                       struct btrfs_block_group_cache *group,
7594                                       struct inode *reloc_inode)
7595 {
7596         struct btrfs_root *reloc_root;
7597         struct extent_buffer *eb = NULL;
7598         struct btrfs_key *keys;
7599         u64 *nodes;
7600         int level;
7601         int shared_level;
7602         int lowest_level = 0;
7603         int ret;
7604
7605         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7606                 lowest_level = ref_path->owner_objectid;
7607
7608         if (!root->ref_cows) {
7609                 path->lowest_level = lowest_level;
7610                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7611                 BUG_ON(ret < 0);
7612                 path->lowest_level = 0;
7613                 btrfs_release_path(root, path);
7614                 return 0;
7615         }
7616
7617         mutex_lock(&root->fs_info->tree_reloc_mutex);
7618         ret = init_reloc_tree(trans, root);
7619         BUG_ON(ret);
7620         reloc_root = root->reloc_root;
7621
7622         shared_level = ref_path->shared_level;
7623         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7624
7625         keys = ref_path->node_keys;
7626         nodes = ref_path->new_nodes;
7627         memset(&keys[shared_level + 1], 0,
7628                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7629         memset(&nodes[shared_level + 1], 0,
7630                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7631
7632         if (nodes[lowest_level] == 0) {
7633                 path->lowest_level = lowest_level;
7634                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7635                                         0, 1);
7636                 BUG_ON(ret);
7637                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7638                         eb = path->nodes[level];
7639                         if (!eb || eb == reloc_root->node)
7640                                 break;
7641                         nodes[level] = eb->start;
7642                         if (level == 0)
7643                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7644                         else
7645                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7646                 }
7647                 if (nodes[0] &&
7648                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7649                         eb = path->nodes[0];
7650                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7651                                                       group, reloc_inode);
7652                         BUG_ON(ret);
7653                 }
7654                 btrfs_release_path(reloc_root, path);
7655         } else {
7656                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7657                                        lowest_level);
7658                 BUG_ON(ret);
7659         }
7660
7661         /*
7662          * replace tree blocks in the fs tree with tree blocks in
7663          * the reloc tree.
7664          */
7665         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7666         BUG_ON(ret < 0);
7667
7668         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7669                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7670                                         0, 0);
7671                 BUG_ON(ret);
7672                 extent_buffer_get(path->nodes[0]);
7673                 eb = path->nodes[0];
7674                 btrfs_release_path(reloc_root, path);
7675                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7676                 BUG_ON(ret);
7677                 free_extent_buffer(eb);
7678         }
7679
7680         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7681         path->lowest_level = 0;
7682         return 0;
7683 }
7684
7685 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7686                                         struct btrfs_root *root,
7687                                         struct btrfs_path *path,
7688                                         struct btrfs_key *first_key,
7689                                         struct btrfs_ref_path *ref_path)
7690 {
7691         int ret;
7692
7693         ret = relocate_one_path(trans, root, path, first_key,
7694                                 ref_path, NULL, NULL);
7695         BUG_ON(ret);
7696
7697         return 0;
7698 }
7699
7700 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7701                                     struct btrfs_root *extent_root,
7702                                     struct btrfs_path *path,
7703                                     struct btrfs_key *extent_key)
7704 {
7705         int ret;
7706
7707         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7708         if (ret)
7709                 goto out;
7710         ret = btrfs_del_item(trans, extent_root, path);
7711 out:
7712         btrfs_release_path(extent_root, path);
7713         return ret;
7714 }
7715
7716 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7717                                                 struct btrfs_ref_path *ref_path)
7718 {
7719         struct btrfs_key root_key;
7720
7721         root_key.objectid = ref_path->root_objectid;
7722         root_key.type = BTRFS_ROOT_ITEM_KEY;
7723         if (is_cowonly_root(ref_path->root_objectid))
7724                 root_key.offset = 0;
7725         else
7726                 root_key.offset = (u64)-1;
7727
7728         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7729 }
7730
7731 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7732                                         struct btrfs_path *path,
7733                                         struct btrfs_key *extent_key,
7734                                         struct btrfs_block_group_cache *group,
7735                                         struct inode *reloc_inode, int pass)
7736 {
7737         struct btrfs_trans_handle *trans;
7738         struct btrfs_root *found_root;
7739         struct btrfs_ref_path *ref_path = NULL;
7740         struct disk_extent *new_extents = NULL;
7741         int nr_extents = 0;
7742         int loops;
7743         int ret;
7744         int level;
7745         struct btrfs_key first_key;
7746         u64 prev_block = 0;
7747
7748
7749         trans = btrfs_start_transaction(extent_root, 1);
7750         BUG_ON(!trans);
7751
7752         if (extent_key->objectid == 0) {
7753                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7754                 goto out;
7755         }
7756
7757         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7758         if (!ref_path) {
7759                 ret = -ENOMEM;
7760                 goto out;
7761         }
7762
7763         for (loops = 0; ; loops++) {
7764                 if (loops == 0) {
7765                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7766                                                    extent_key->objectid);
7767                 } else {
7768                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7769                 }
7770                 if (ret < 0)
7771                         goto out;
7772                 if (ret > 0)
7773                         break;
7774
7775                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7776                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7777                         continue;
7778
7779                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7780                 BUG_ON(!found_root);
7781                 /*
7782                  * for reference counted tree, only process reference paths
7783                  * rooted at the latest committed root.
7784                  */
7785                 if (found_root->ref_cows &&
7786                     ref_path->root_generation != found_root->root_key.offset)
7787                         continue;
7788
7789                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7790                         if (pass == 0) {
7791                                 /*
7792                                  * copy data extents to new locations
7793                                  */
7794                                 u64 group_start = group->key.objectid;
7795                                 ret = relocate_data_extent(reloc_inode,
7796                                                            extent_key,
7797                                                            group_start);
7798                                 if (ret < 0)
7799                                         goto out;
7800                                 break;
7801                         }
7802                         level = 0;
7803                 } else {
7804                         level = ref_path->owner_objectid;
7805                 }
7806
7807                 if (prev_block != ref_path->nodes[level]) {
7808                         struct extent_buffer *eb;
7809                         u64 block_start = ref_path->nodes[level];
7810                         u64 block_size = btrfs_level_size(found_root, level);
7811
7812                         eb = read_tree_block(found_root, block_start,
7813                                              block_size, 0);
7814                         btrfs_tree_lock(eb);
7815                         BUG_ON(level != btrfs_header_level(eb));
7816
7817                         if (level == 0)
7818                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7819                         else
7820                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7821
7822                         btrfs_tree_unlock(eb);
7823                         free_extent_buffer(eb);
7824                         prev_block = block_start;
7825                 }
7826
7827                 mutex_lock(&extent_root->fs_info->trans_mutex);
7828                 btrfs_record_root_in_trans(found_root);
7829                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7830                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7831                         /*
7832                          * try to update data extent references while
7833                          * keeping metadata shared between snapshots.
7834                          */
7835                         if (pass == 1) {
7836                                 ret = relocate_one_path(trans, found_root,
7837                                                 path, &first_key, ref_path,
7838                                                 group, reloc_inode);
7839                                 if (ret < 0)
7840                                         goto out;
7841                                 continue;
7842                         }
7843                         /*
7844                          * use fallback method to process the remaining
7845                          * references.
7846                          */
7847                         if (!new_extents) {
7848                                 u64 group_start = group->key.objectid;
7849                                 new_extents = kmalloc(sizeof(*new_extents),
7850                                                       GFP_NOFS);
7851                                 nr_extents = 1;
7852                                 ret = get_new_locations(reloc_inode,
7853                                                         extent_key,
7854                                                         group_start, 1,
7855                                                         &new_extents,
7856                                                         &nr_extents);
7857                                 if (ret)
7858                                         goto out;
7859                         }
7860                         ret = replace_one_extent(trans, found_root,
7861                                                 path, extent_key,
7862                                                 &first_key, ref_path,
7863                                                 new_extents, nr_extents);
7864                 } else {
7865                         ret = relocate_tree_block(trans, found_root, path,
7866                                                   &first_key, ref_path);
7867                 }
7868                 if (ret < 0)
7869                         goto out;
7870         }
7871         ret = 0;
7872 out:
7873         btrfs_end_transaction(trans, extent_root);
7874         kfree(new_extents);
7875         kfree(ref_path);
7876         return ret;
7877 }
7878 #endif
7879
7880 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7881 {
7882         u64 num_devices;
7883         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7884                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7885
7886         num_devices = root->fs_info->fs_devices->rw_devices;
7887         if (num_devices == 1) {
7888                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7889                 stripped = flags & ~stripped;
7890
7891                 /* turn raid0 into single device chunks */
7892                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7893                         return stripped;
7894
7895                 /* turn mirroring into duplication */
7896                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7897                              BTRFS_BLOCK_GROUP_RAID10))
7898                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7899                 return flags;
7900         } else {
7901                 /* they already had raid on here, just return */
7902                 if (flags & stripped)
7903                         return flags;
7904
7905                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7906                 stripped = flags & ~stripped;
7907
7908                 /* switch duplicated blocks with raid1 */
7909                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7910                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7911
7912                 /* turn single device chunks into raid0 */
7913                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7914         }
7915         return flags;
7916 }
7917
7918 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
7919 {
7920         struct btrfs_space_info *sinfo = cache->space_info;
7921         u64 num_bytes;
7922         int ret = -ENOSPC;
7923
7924         if (cache->ro)
7925                 return 0;
7926
7927         spin_lock(&sinfo->lock);
7928         spin_lock(&cache->lock);
7929         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7930                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7931
7932         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7933             sinfo->bytes_may_use + sinfo->bytes_readonly +
7934             cache->reserved_pinned + num_bytes < sinfo->total_bytes) {
7935                 sinfo->bytes_readonly += num_bytes;
7936                 sinfo->bytes_reserved += cache->reserved_pinned;
7937                 cache->reserved_pinned = 0;
7938                 cache->ro = 1;
7939                 ret = 0;
7940         }
7941         spin_unlock(&cache->lock);
7942         spin_unlock(&sinfo->lock);
7943         return ret;
7944 }
7945
7946 int btrfs_set_block_group_ro(struct btrfs_root *root,
7947                              struct btrfs_block_group_cache *cache)
7948
7949 {
7950         struct btrfs_trans_handle *trans;
7951         u64 alloc_flags;
7952         int ret;
7953
7954         BUG_ON(cache->ro);
7955
7956         trans = btrfs_join_transaction(root, 1);
7957         BUG_ON(IS_ERR(trans));
7958
7959         alloc_flags = update_block_group_flags(root, cache->flags);
7960         if (alloc_flags != cache->flags)
7961                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7962
7963         ret = set_block_group_ro(cache);
7964         if (!ret)
7965                 goto out;
7966         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7967         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7968         if (ret < 0)
7969                 goto out;
7970         ret = set_block_group_ro(cache);
7971 out:
7972         btrfs_end_transaction(trans, root);
7973         return ret;
7974 }
7975
7976 int btrfs_set_block_group_rw(struct btrfs_root *root,
7977                               struct btrfs_block_group_cache *cache)
7978 {
7979         struct btrfs_space_info *sinfo = cache->space_info;
7980         u64 num_bytes;
7981
7982         BUG_ON(!cache->ro);
7983
7984         spin_lock(&sinfo->lock);
7985         spin_lock(&cache->lock);
7986         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7987                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7988         sinfo->bytes_readonly -= num_bytes;
7989         cache->ro = 0;
7990         spin_unlock(&cache->lock);
7991         spin_unlock(&sinfo->lock);
7992         return 0;
7993 }
7994
7995 /*
7996  * checks to see if its even possible to relocate this block group.
7997  *
7998  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7999  * ok to go ahead and try.
8000  */
8001 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8002 {
8003         struct btrfs_block_group_cache *block_group;
8004         struct btrfs_space_info *space_info;
8005         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8006         struct btrfs_device *device;
8007         int full = 0;
8008         int ret = 0;
8009
8010         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8011
8012         /* odd, couldn't find the block group, leave it alone */
8013         if (!block_group)
8014                 return -1;
8015
8016         /* no bytes used, we're good */
8017         if (!btrfs_block_group_used(&block_group->item))
8018                 goto out;
8019
8020         space_info = block_group->space_info;
8021         spin_lock(&space_info->lock);
8022
8023         full = space_info->full;
8024
8025         /*
8026          * if this is the last block group we have in this space, we can't
8027          * relocate it unless we're able to allocate a new chunk below.
8028          *
8029          * Otherwise, we need to make sure we have room in the space to handle
8030          * all of the extents from this block group.  If we can, we're good
8031          */
8032         if ((space_info->total_bytes != block_group->key.offset) &&
8033            (space_info->bytes_used + space_info->bytes_reserved +
8034             space_info->bytes_pinned + space_info->bytes_readonly +
8035             btrfs_block_group_used(&block_group->item) <
8036             space_info->total_bytes)) {
8037                 spin_unlock(&space_info->lock);
8038                 goto out;
8039         }
8040         spin_unlock(&space_info->lock);
8041
8042         /*
8043          * ok we don't have enough space, but maybe we have free space on our
8044          * devices to allocate new chunks for relocation, so loop through our
8045          * alloc devices and guess if we have enough space.  However, if we
8046          * were marked as full, then we know there aren't enough chunks, and we
8047          * can just return.
8048          */
8049         ret = -1;
8050         if (full)
8051                 goto out;
8052
8053         mutex_lock(&root->fs_info->chunk_mutex);
8054         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8055                 u64 min_free = btrfs_block_group_used(&block_group->item);
8056                 u64 dev_offset, max_avail;
8057
8058                 /*
8059                  * check to make sure we can actually find a chunk with enough
8060                  * space to fit our block group in.
8061                  */
8062                 if (device->total_bytes > device->bytes_used + min_free) {
8063                         ret = find_free_dev_extent(NULL, device, min_free,
8064                                                    &dev_offset, &max_avail);
8065                         if (!ret)
8066                                 break;
8067                         ret = -1;
8068                 }
8069         }
8070         mutex_unlock(&root->fs_info->chunk_mutex);
8071 out:
8072         btrfs_put_block_group(block_group);
8073         return ret;
8074 }
8075
8076 static int find_first_block_group(struct btrfs_root *root,
8077                 struct btrfs_path *path, struct btrfs_key *key)
8078 {
8079         int ret = 0;
8080         struct btrfs_key found_key;
8081         struct extent_buffer *leaf;
8082         int slot;
8083
8084         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8085         if (ret < 0)
8086                 goto out;
8087
8088         while (1) {
8089                 slot = path->slots[0];
8090                 leaf = path->nodes[0];
8091                 if (slot >= btrfs_header_nritems(leaf)) {
8092                         ret = btrfs_next_leaf(root, path);
8093                         if (ret == 0)
8094                                 continue;
8095                         if (ret < 0)
8096                                 goto out;
8097                         break;
8098                 }
8099                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8100
8101                 if (found_key.objectid >= key->objectid &&
8102                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8103                         ret = 0;
8104                         goto out;
8105                 }
8106                 path->slots[0]++;
8107         }
8108 out:
8109         return ret;
8110 }
8111
8112 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8113 {
8114         struct btrfs_block_group_cache *block_group;
8115         u64 last = 0;
8116
8117         while (1) {
8118                 struct inode *inode;
8119
8120                 block_group = btrfs_lookup_first_block_group(info, last);
8121                 while (block_group) {
8122                         spin_lock(&block_group->lock);
8123                         if (block_group->iref)
8124                                 break;
8125                         spin_unlock(&block_group->lock);
8126                         block_group = next_block_group(info->tree_root,
8127                                                        block_group);
8128                 }
8129                 if (!block_group) {
8130                         if (last == 0)
8131                                 break;
8132                         last = 0;
8133                         continue;
8134                 }
8135
8136                 inode = block_group->inode;
8137                 block_group->iref = 0;
8138                 block_group->inode = NULL;
8139                 spin_unlock(&block_group->lock);
8140                 iput(inode);
8141                 last = block_group->key.objectid + block_group->key.offset;
8142                 btrfs_put_block_group(block_group);
8143         }
8144 }
8145
8146 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8147 {
8148         struct btrfs_block_group_cache *block_group;
8149         struct btrfs_space_info *space_info;
8150         struct btrfs_caching_control *caching_ctl;
8151         struct rb_node *n;
8152
8153         down_write(&info->extent_commit_sem);
8154         while (!list_empty(&info->caching_block_groups)) {
8155                 caching_ctl = list_entry(info->caching_block_groups.next,
8156                                          struct btrfs_caching_control, list);
8157                 list_del(&caching_ctl->list);
8158                 put_caching_control(caching_ctl);
8159         }
8160         up_write(&info->extent_commit_sem);
8161
8162         spin_lock(&info->block_group_cache_lock);
8163         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8164                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8165                                        cache_node);
8166                 rb_erase(&block_group->cache_node,
8167                          &info->block_group_cache_tree);
8168                 spin_unlock(&info->block_group_cache_lock);
8169
8170                 down_write(&block_group->space_info->groups_sem);
8171                 list_del(&block_group->list);
8172                 up_write(&block_group->space_info->groups_sem);
8173
8174                 if (block_group->cached == BTRFS_CACHE_STARTED)
8175                         wait_block_group_cache_done(block_group);
8176
8177                 btrfs_remove_free_space_cache(block_group);
8178                 btrfs_put_block_group(block_group);
8179
8180                 spin_lock(&info->block_group_cache_lock);
8181         }
8182         spin_unlock(&info->block_group_cache_lock);
8183
8184         /* now that all the block groups are freed, go through and
8185          * free all the space_info structs.  This is only called during
8186          * the final stages of unmount, and so we know nobody is
8187          * using them.  We call synchronize_rcu() once before we start,
8188          * just to be on the safe side.
8189          */
8190         synchronize_rcu();
8191
8192         release_global_block_rsv(info);
8193
8194         while(!list_empty(&info->space_info)) {
8195                 space_info = list_entry(info->space_info.next,
8196                                         struct btrfs_space_info,
8197                                         list);
8198                 if (space_info->bytes_pinned > 0 ||
8199                     space_info->bytes_reserved > 0) {
8200                         WARN_ON(1);
8201                         dump_space_info(space_info, 0, 0);
8202                 }
8203                 list_del(&space_info->list);
8204                 kfree(space_info);
8205         }
8206         return 0;
8207 }
8208
8209 static void __link_block_group(struct btrfs_space_info *space_info,
8210                                struct btrfs_block_group_cache *cache)
8211 {
8212         int index = get_block_group_index(cache);
8213
8214         down_write(&space_info->groups_sem);
8215         list_add_tail(&cache->list, &space_info->block_groups[index]);
8216         up_write(&space_info->groups_sem);
8217 }
8218
8219 int btrfs_read_block_groups(struct btrfs_root *root)
8220 {
8221         struct btrfs_path *path;
8222         int ret;
8223         struct btrfs_block_group_cache *cache;
8224         struct btrfs_fs_info *info = root->fs_info;
8225         struct btrfs_space_info *space_info;
8226         struct btrfs_key key;
8227         struct btrfs_key found_key;
8228         struct extent_buffer *leaf;
8229         int need_clear = 0;
8230         u64 cache_gen;
8231
8232         root = info->extent_root;
8233         key.objectid = 0;
8234         key.offset = 0;
8235         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8236         path = btrfs_alloc_path();
8237         if (!path)
8238                 return -ENOMEM;
8239
8240         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8241         if (cache_gen != 0 &&
8242             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8243                 need_clear = 1;
8244         if (btrfs_test_opt(root, CLEAR_CACHE))
8245                 need_clear = 1;
8246         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8247                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8248
8249         while (1) {
8250                 ret = find_first_block_group(root, path, &key);
8251                 if (ret > 0)
8252                         break;
8253                 if (ret != 0)
8254                         goto error;
8255
8256                 leaf = path->nodes[0];
8257                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8258                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8259                 if (!cache) {
8260                         ret = -ENOMEM;
8261                         goto error;
8262                 }
8263
8264                 atomic_set(&cache->count, 1);
8265                 spin_lock_init(&cache->lock);
8266                 spin_lock_init(&cache->tree_lock);
8267                 cache->fs_info = info;
8268                 INIT_LIST_HEAD(&cache->list);
8269                 INIT_LIST_HEAD(&cache->cluster_list);
8270
8271                 if (need_clear)
8272                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8273
8274                 /*
8275                  * we only want to have 32k of ram per block group for keeping
8276                  * track of free space, and if we pass 1/2 of that we want to
8277                  * start converting things over to using bitmaps
8278                  */
8279                 cache->extents_thresh = ((1024 * 32) / 2) /
8280                         sizeof(struct btrfs_free_space);
8281
8282                 read_extent_buffer(leaf, &cache->item,
8283                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8284                                    sizeof(cache->item));
8285                 memcpy(&cache->key, &found_key, sizeof(found_key));
8286
8287                 key.objectid = found_key.objectid + found_key.offset;
8288                 btrfs_release_path(root, path);
8289                 cache->flags = btrfs_block_group_flags(&cache->item);
8290                 cache->sectorsize = root->sectorsize;
8291
8292                 /*
8293                  * check for two cases, either we are full, and therefore
8294                  * don't need to bother with the caching work since we won't
8295                  * find any space, or we are empty, and we can just add all
8296                  * the space in and be done with it.  This saves us _alot_ of
8297                  * time, particularly in the full case.
8298                  */
8299                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8300                         exclude_super_stripes(root, cache);
8301                         cache->last_byte_to_unpin = (u64)-1;
8302                         cache->cached = BTRFS_CACHE_FINISHED;
8303                         free_excluded_extents(root, cache);
8304                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8305                         exclude_super_stripes(root, cache);
8306                         cache->last_byte_to_unpin = (u64)-1;
8307                         cache->cached = BTRFS_CACHE_FINISHED;
8308                         add_new_free_space(cache, root->fs_info,
8309                                            found_key.objectid,
8310                                            found_key.objectid +
8311                                            found_key.offset);
8312                         free_excluded_extents(root, cache);
8313                 }
8314
8315                 ret = update_space_info(info, cache->flags, found_key.offset,
8316                                         btrfs_block_group_used(&cache->item),
8317                                         &space_info);
8318                 BUG_ON(ret);
8319                 cache->space_info = space_info;
8320                 spin_lock(&cache->space_info->lock);
8321                 cache->space_info->bytes_readonly += cache->bytes_super;
8322                 spin_unlock(&cache->space_info->lock);
8323
8324                 __link_block_group(space_info, cache);
8325
8326                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8327                 BUG_ON(ret);
8328
8329                 set_avail_alloc_bits(root->fs_info, cache->flags);
8330                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8331                         set_block_group_ro(cache);
8332         }
8333
8334         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8335                 if (!(get_alloc_profile(root, space_info->flags) &
8336                       (BTRFS_BLOCK_GROUP_RAID10 |
8337                        BTRFS_BLOCK_GROUP_RAID1 |
8338                        BTRFS_BLOCK_GROUP_DUP)))
8339                         continue;
8340                 /*
8341                  * avoid allocating from un-mirrored block group if there are
8342                  * mirrored block groups.
8343                  */
8344                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8345                         set_block_group_ro(cache);
8346                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8347                         set_block_group_ro(cache);
8348         }
8349
8350         init_global_block_rsv(info);
8351         ret = 0;
8352 error:
8353         btrfs_free_path(path);
8354         return ret;
8355 }
8356
8357 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8358                            struct btrfs_root *root, u64 bytes_used,
8359                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8360                            u64 size)
8361 {
8362         int ret;
8363         struct btrfs_root *extent_root;
8364         struct btrfs_block_group_cache *cache;
8365
8366         extent_root = root->fs_info->extent_root;
8367
8368         root->fs_info->last_trans_log_full_commit = trans->transid;
8369
8370         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8371         if (!cache)
8372                 return -ENOMEM;
8373
8374         cache->key.objectid = chunk_offset;
8375         cache->key.offset = size;
8376         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8377         cache->sectorsize = root->sectorsize;
8378         cache->fs_info = root->fs_info;
8379
8380         /*
8381          * we only want to have 32k of ram per block group for keeping track
8382          * of free space, and if we pass 1/2 of that we want to start
8383          * converting things over to using bitmaps
8384          */
8385         cache->extents_thresh = ((1024 * 32) / 2) /
8386                 sizeof(struct btrfs_free_space);
8387         atomic_set(&cache->count, 1);
8388         spin_lock_init(&cache->lock);
8389         spin_lock_init(&cache->tree_lock);
8390         INIT_LIST_HEAD(&cache->list);
8391         INIT_LIST_HEAD(&cache->cluster_list);
8392
8393         btrfs_set_block_group_used(&cache->item, bytes_used);
8394         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8395         cache->flags = type;
8396         btrfs_set_block_group_flags(&cache->item, type);
8397
8398         cache->last_byte_to_unpin = (u64)-1;
8399         cache->cached = BTRFS_CACHE_FINISHED;
8400         exclude_super_stripes(root, cache);
8401
8402         add_new_free_space(cache, root->fs_info, chunk_offset,
8403                            chunk_offset + size);
8404
8405         free_excluded_extents(root, cache);
8406
8407         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8408                                 &cache->space_info);
8409         BUG_ON(ret);
8410
8411         spin_lock(&cache->space_info->lock);
8412         cache->space_info->bytes_readonly += cache->bytes_super;
8413         spin_unlock(&cache->space_info->lock);
8414
8415         __link_block_group(cache->space_info, cache);
8416
8417         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8418         BUG_ON(ret);
8419
8420         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8421                                 sizeof(cache->item));
8422         BUG_ON(ret);
8423
8424         set_avail_alloc_bits(extent_root->fs_info, type);
8425
8426         return 0;
8427 }
8428
8429 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8430                              struct btrfs_root *root, u64 group_start)
8431 {
8432         struct btrfs_path *path;
8433         struct btrfs_block_group_cache *block_group;
8434         struct btrfs_free_cluster *cluster;
8435         struct btrfs_root *tree_root = root->fs_info->tree_root;
8436         struct btrfs_key key;
8437         struct inode *inode;
8438         int ret;
8439         int factor;
8440
8441         root = root->fs_info->extent_root;
8442
8443         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8444         BUG_ON(!block_group);
8445         BUG_ON(!block_group->ro);
8446
8447         memcpy(&key, &block_group->key, sizeof(key));
8448         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8449                                   BTRFS_BLOCK_GROUP_RAID1 |
8450                                   BTRFS_BLOCK_GROUP_RAID10))
8451                 factor = 2;
8452         else
8453                 factor = 1;
8454
8455         /* make sure this block group isn't part of an allocation cluster */
8456         cluster = &root->fs_info->data_alloc_cluster;
8457         spin_lock(&cluster->refill_lock);
8458         btrfs_return_cluster_to_free_space(block_group, cluster);
8459         spin_unlock(&cluster->refill_lock);
8460
8461         /*
8462          * make sure this block group isn't part of a metadata
8463          * allocation cluster
8464          */
8465         cluster = &root->fs_info->meta_alloc_cluster;
8466         spin_lock(&cluster->refill_lock);
8467         btrfs_return_cluster_to_free_space(block_group, cluster);
8468         spin_unlock(&cluster->refill_lock);
8469
8470         path = btrfs_alloc_path();
8471         BUG_ON(!path);
8472
8473         inode = lookup_free_space_inode(root, block_group, path);
8474         if (!IS_ERR(inode)) {
8475                 btrfs_orphan_add(trans, inode);
8476                 clear_nlink(inode);
8477                 /* One for the block groups ref */
8478                 spin_lock(&block_group->lock);
8479                 if (block_group->iref) {
8480                         block_group->iref = 0;
8481                         block_group->inode = NULL;
8482                         spin_unlock(&block_group->lock);
8483                         iput(inode);
8484                 } else {
8485                         spin_unlock(&block_group->lock);
8486                 }
8487                 /* One for our lookup ref */
8488                 iput(inode);
8489         }
8490
8491         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8492         key.offset = block_group->key.objectid;
8493         key.type = 0;
8494
8495         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8496         if (ret < 0)
8497                 goto out;
8498         if (ret > 0)
8499                 btrfs_release_path(tree_root, path);
8500         if (ret == 0) {
8501                 ret = btrfs_del_item(trans, tree_root, path);
8502                 if (ret)
8503                         goto out;
8504                 btrfs_release_path(tree_root, path);
8505         }
8506
8507         spin_lock(&root->fs_info->block_group_cache_lock);
8508         rb_erase(&block_group->cache_node,
8509                  &root->fs_info->block_group_cache_tree);
8510         spin_unlock(&root->fs_info->block_group_cache_lock);
8511
8512         down_write(&block_group->space_info->groups_sem);
8513         /*
8514          * we must use list_del_init so people can check to see if they
8515          * are still on the list after taking the semaphore
8516          */
8517         list_del_init(&block_group->list);
8518         up_write(&block_group->space_info->groups_sem);
8519
8520         if (block_group->cached == BTRFS_CACHE_STARTED)
8521                 wait_block_group_cache_done(block_group);
8522
8523         btrfs_remove_free_space_cache(block_group);
8524
8525         spin_lock(&block_group->space_info->lock);
8526         block_group->space_info->total_bytes -= block_group->key.offset;
8527         block_group->space_info->bytes_readonly -= block_group->key.offset;
8528         block_group->space_info->disk_total -= block_group->key.offset * factor;
8529         spin_unlock(&block_group->space_info->lock);
8530
8531         memcpy(&key, &block_group->key, sizeof(key));
8532
8533         btrfs_clear_space_info_full(root->fs_info);
8534
8535         btrfs_put_block_group(block_group);
8536         btrfs_put_block_group(block_group);
8537
8538         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8539         if (ret > 0)
8540                 ret = -EIO;
8541         if (ret < 0)
8542                 goto out;
8543
8544         ret = btrfs_del_item(trans, root, path);
8545 out:
8546         btrfs_free_path(path);
8547         return ret;
8548 }