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